From a Heuristic Point of View

From a Heuristic Point of View:

Essays in Honour of Carlo Cellucci

Edited by

Emiliano Ippoliti and Cesare Cozzo

 From a Heuristic Point of View:
Essays in Honour of Carlo Cellucci,
Edited by Emiliano Ippoliti and Cesare Cozzo

This book first published 2014

Cambridge Scholars Publishing

12 Back Chapman Street, Newcastle upon Tyne, NE6 2XX, UK

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

Copyright 2014 by Emiliano Ippoliti, Cesare Cozzo and contributors

All rights for this book reserved. No part of this book may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or
otherwise, without the prior permission of the copyright owner.

ISBN (10): 1-4438-5649-5, ISBN (13): 978-1-4438-5649-2

 TABLE OF CONTENTS

Introduction ............................................................................................... vii

Section I: Mathematical Logic, Mathematical Knowledge and Truth

Chapter One ................................................................................................. 3

Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind
Rosaria Egidi

Chapter Two .............................................................................................. 23

Serendipity and Mathematical Logic
Donald Gillies

Chapter Three ............................................................................................ 41

What Mathematical Logic Says about the Foundations of Mathematics
Claudio Bernardi

Chapter Four .............................................................................................. 55

Dedekind, Hilbert, Gdel: the Comparison between Logical Sentences
and Arithmetical Sentences
Michele Abrusci

Chapter Five .............................................................................................. 73

The Status of Mathematical Knowledge
Dag Prawitz

Chapter Six ................................................................................................ 91

Empiricism and Experimental Mathematics
Gabriele Lolli

Chapter Seven.......................................................................................... 107

Is Truth a Chimera?
Cesare Cozzo
vi Table of Contents

Section II: The Project for a Logic of Discovery

Chapter Eight ........................................................................................... 127

To Establish New Mathematics, We Use our Mental Models
and Build on Established Mathematics
Reuben Hersh

Chapter Nine............................................................................................ 147

Fermats Last Theorem and the Logicians
Emily Grosholz

Chapter Ten ............................................................................................. 163

Christiaan Huygenss On Reckoning in Games of Chance: A Case Study
on Celluccis Heuristic Conception of Mathematics
Daniel G. Campos

Chapter Eleven ........................................................................................ 179

Natural Mathematics and Natural Logic
Lorenzo Magnani

Chapter Twelve ....................................................................................... 195

For a Bottom-Up Approach to the Philosophy of Science
Maria Carla Galavotti

Chapter Thirteen ...................................................................................... 213

Mathematizing Risk: A Heuristic Point of View
Emiliano Ippoliti

Chapter Fourteen ..................................................................................... 241

Reflections on the Objectivity of Mathematics
Robert Thomas

Contributors ............................................................................................. 257

Index ........................................................................................................ 261

 INTRODUCTION

Strange as it may seem, some people believe that logic is a branch of

mathematics which has little to do with philosophy. Carlo Celluccis work
is a refutation of this bizarre prejudice. For a long time Cellucci has been a
distinguished logician in Italy. His most significant contributions to
mathematical logic concern proof-theory. Teoria della dimostrazione,
published in 1978, was the first book on proof-theory to appear in Italian.
But Celluccis work in logic has always been connected with a concern for
philosophical problems. One problem, in particular that of discovery
became prominent. It can be summarized in the question: how is it
possible that we acquire new knowledge? Cellucci gradually came to the
conviction that logic can play a relevant role in mathematics, science and
even philosophy (Cellucci 2013, p. 18) only if it helps us to solve this
problem. In this respect mathematical logic, the paradigm of logic initiated
by Gottlob Frege, is wholly inadequate. It is a logic of justification, not a
logic of discovery. Its aim is to provide a secure foundation for
mathematics, and Gdels incompleteness theorems show that it cannot
fulfill this purpose. Thus, Cellucci has gradually developed a critique of
mathematical logic and, on the other hand, the project for a new logic,
framed in the context of an original naturalistic conception of knowledge
and philosophy. The most recent and articulated statement of these views
can be found in his books Perch ancora la filosofia (2008) and
Rethinking Logic (2013). According to Celluccis naturalistic conception
logic is a continuation of the problem solving processes with which
biological evolution has endowed all organisms (2013, p. 18).
The aforementioned path from logic to a naturalistic conception of
knowledge and philosophy partly explains the title From a heuristic point
of view chosen by the editors of this volume in honour of Carlo Cellucci. It
partly explains the title because it suggests an analogy with the author of
the celebrated collection From a logical point of view, Willard Van Orman
Quine, the initiator of contemporary naturalized epistemology. Another
part of the explanation, of course, lies in the word heuristic, which
emphasizes the most characteristic trait of Celluccis thought and the main
difference between him and Quine: the emphasis on discovery. Both
philosophers maintain that knowledge is fallible and corrigible. Fallibilism
is thus a further idea that they share. But Quines picture of human
viii Introduction

knowledge embodies a principle of conservatism, Quines maxim of

minimum mutilation: when we encounter recalcitrant experience, we make
the smallest possible number of changes to our beliefs and do not alter our
overall theory more than necessary. According to the maxim of minimum
mutilation classical logic, though in principle not immune to revision, will
not be revised, because, Quine says, losing its convenience, simplicity and
beauty would be too high a price. Quines picture of knowledge and logic
thus appears almost static, compared with Celluccis insistence on
discovery methods and on the continuous growth of knowledge. Quines
doctrine that no statement, not even the logical law of the excluded
middle, is in principle immune to revision appears a rather lukewarm
revisionism, when compared with Celluccis harsh critique, not only of
classical logic, but of all the logics developed within the paradigm of
mathematical logic.
The essays collected in this book were written by some of Carlos
colleagues and two former students (the editors) who have shared his
interests and investigated topics that he dealt with. Each contributor in her
or his way, has to varying degrees studied, collaborated, discussed, agreed
or disagreed with Cellucci. The variety of these different attitudes is
reflected in the essays. A first group of essays, collected in Part I, more
directly concern mathematical logic, mathematical knowledge and
truth. Celluccis critique of mathematical logic is part of a more general
critique of the axiomatic method and of the idea of providing a secure
foundation for mathematics. The picture of mathematical knowledge as
based on the axiomatic method and the attempt to establish the absolute
certainty of mathematics by providing a firm foundation for axiomatic
theories characterize the foundational schools of logicism and Hilbertian
formalism in the twentieth century, which used mathematical logic as a
tool for their philosophical programmes. In her paper Regelfolgen in
Wittgensteins Philosophy of Mathematics and Mind Rosaria Egidi
examines some central concepts of Ludwig Wittgensteins writings. She
reminds us of the philosophers severe judgment in the Remarks on the
Foundations of Mathematics: Mathematical logic has completely
deformed the thinking of mathematicians and of philosophers. Though
Cellucci usually dissociates himself from Wittgenstein, Egidi expresses
the conviction that there is a point of agreement between Celluccis
arguments against foundational strategies and axiomatic method and
Wittgensteins idea that mathematics is embedded in the natural history
of mankind and bound up with the social and anthropological forms of
life. Celluccis arguments in Le Ragioni della logica (1998), Filosofia e
matematica (2002) and La filosofia della matematica del Novecento
 From a Heuristic Point of View ix

(2007) are summarized by Donald Gillies in the first part of his essay:
Cellucci maintains that the logicist programme of Frege and Russell and
Hilberts formalist programme ended in failure because of the limitative
results of Gdel and Tarski. Gillies agrees that Gdels first incompleteness
theorem gives a fatal blow to logicism and that Gdels second
incompleteness theorem shows the impossibility of carrying out Hilberts
programme. He thus concludes that Celluccis criticisms of mathematical
logic show that mathematical logic did not succeed in establishing a
foundation of mathematics of the kind that the inventors of mathematical
logic had hoped to create. However, he remarks, mathematical logic has
proved to be a very useful tool for computer science. This idea lies
behind the title of Gillies essay Serendipity and Mathematical Logic. He
defines serendipity as looking for one thing and finding another. The
concept can be applied to mathematical logic because the pioneers of
mathematical logic were looking for something they did not find: a
foundation for mathematics that would make mathematical statements
certain; but they found something other than what they were looking for:
invaluable, perhaps indeed essential, tools for the development of
computers. In the last part of his essay Gillies explains how this came
about.
In his essay What Mathematical Logic Says about the Foundations of
Mathematics Claudio Bernardi grants that perhaps in the nineteenth
century logic was regarded as a way to guarantee the certainty of
mathematics and that nowadays it seems nave, and perhaps even futile,
to hope for a definitive, proven certainty of mathematics. Nevertheless,
Bernardi thinks that mathematical logic offers a fruitful model of
mathematical activity. Though it is not a faithful description of how a
mathematician works, mathematical logic provides us with a theoretical
framework (model theory, proof theory, recursion theory) in which various
implicit features of mathematical practice are made explicit, a precise
definition of proof is given, rigorous methods and procedures to develop
mathematical theories are suggested. Bernardi emphasizes the fact that
the theoretical framework of mathematical logic yields significant results:
mathematical logic, exactly like mathematical analysis, is justified by its
results, which clarify the sense and the limits of mathematical activity.
Cellucci maintains that the founders of mathematical logic identified the
mathematical method with the axiomatic method. The axiomatic method is
the main object of study of mathematical logic. Mathematical logic is
based on the concept of a formal system as a closed system. A closed
system is : a system based on principles that are given once for all and
cannot change, and whose development consists in deducing conclusions
x Introduction

from them. Therefore, the development of the system requires no input

from the outside (Cellucci 2013, p 178, cfr. 1993). Against the picture of
mathematical knowledge as consisting of closed systems Cellucci
proposes the view that all knowledge, including mathematical knowledge,
consists of open systems: an open system is a system which initially
consists only of the problem to be solved, and possibly other data already
available, and whose development consists in obtaining more and more
hypotheses for solving the problem from the problem itself, and possibly
other data already available, by non-deductive rules. The hypotheses are
then checked by means of the plausibility test procedure. The other data
already available from which hypotheses are possibly obtained come from
other open systems, therefore the development of an open system involves
interactions with other open systems (Cellucci 2013, p. 294). Bernardi
agrees with Cellucci that open systems can better account for the fact that
the mathematical community searches continuously for new axioms,
which are deeper or more general, or more suited for some purpose, trying
to give a more comprehensive explanation of a subject, but he thinks that
both open systems and axiomatic systems reflect mathematical
activities. Bernardis defence of the axiomatic method and of
mathematical logic is strengthened by some final remarks concerning their
educational value and the role of mathematical logic in the contribution of
computer science to mathematics.
As Gillies explains in his essay, Gdels incompleteness theorems are
the linchpin of Celluccis verdict that the basic assumptions of mathematical
logic are untenable. A completely different interpretation of the
significance of incompleteness is proposed by Michele Abrusci in his
paper, Dedekind, Hilbert, Gdel: the Comparison Between Logical
Sentences and Arithmetical Sentences. The equivalence of arithmetical and
logical sentences established by Dedekind in Was sind und was sollen die
Zahlen and Gdels completeness theorem for first order logic combine to
give a picture of the relation between logic and arithmetic according to
which every existential arithmetical first-order sentence belonging to 610
is equivalent to a universal logical sentence belonging to 31 , and vice
versa every universal logical sentence belonging to 31 is equivalent to an
existential arithmetical first-order sentence belonging to 610. Moreover
every universal arithmetical first-order sentence belonging to 310 is
equivalent to an existential logical sentence belonging to 61, and vice
versa every existential logical sentence belonging to 61 is equivalent to an
universal arithmetical first-order sentence belonging to 310 . Abrusci
comments that these results reveal strong mutual dependencies between
logic and arithmetic and the need [] to seek simultaneous foundations
 From a Heuristic Point of View xi

of both logic and arithmetic. He remarks that given this correspondence

between logic and arithmetic, a universal arithmetical quantifier
corresponds to an existential logical quantifier, and an existential
arithmetical quantifier corresponds to a universal logical quantifier.
Perhaps, this exchange of quantifiers when we go from an arithmetical
sentence to a logical sentence, and vice versa is the most important
discovery provided by these results, surely an unexpected result. The
completeness theorem for first order logic can be reformulated as the
principle that every logical sentence belonging to 31 is equivalent to its
logical provability. Abrusci calls this a weak form of the completeness
of logic. The simplest extension of this principle would be every logical
sentence belonging to 31 or belonging to 61 is equivalent to its logical
provability. But if we accept the hypothesis that every logical sentence
belonging to 61 is equivalent to its logical provability, we reach rather
paradoxical consequences: every existential logical sentence belonging
to 61 is equivalent to a universal logical sentence belonging to 31 and
every universal arithmetical sentence belonging to 310 is equivalent to an
existential arithmetical sentence belonging to 601. This is a kind of
collapse of universal and existential quantifiers both in logic and in
arithmetic. But the Incompleteness Theorem proved by Gdel shows that
the extension of the weak form of completeness does not hold, because
not every logical sentence belonging to 61 is equivalent to its logical
provability. Thus, Gdels Incompleteness Theorem avoids the collapse
of quantifiers, in arithmetic and in logic! So, it is not a disaster for logic or
for arithmetic: rather, it saves logic and arithmetic from a collapse of
quantifiers!.
The essays by Abrusci, Bernardi and Gillies concern mathematical
logic. Celluccis criticisms, however, are not levelled only at mathematical
logic, they are also aimed at a deeper target: the role of deduction itself in
mathematics. The axiomatic method is deductive. But the method of
obtaining hypotheses for solving problems that Cellucci calls the analytic
method is non-deductive. Cellucci rejects the common view that
mathematical knowledge is obtained by deductive inference. In his essay
The status of mathematical knowledge Dag Prawitz intends to defend the
common view. Prawitz agrees that after Gdels incompleteness theorem,
one cannot think that mathematics is rightly characterized by saying that it
aims at proving theorems in given axiomatic systems. But he maintains
that the fall of the axiomatic method does not affect the view that
mathematical knowledge is acquired by deductive proofs from obvious
truths, because this view is not tied to the idea that one can specify once
and for all a set of axioms from which all deductive proofs are to start. For
xii Introduction

instance, in arithmetic a deductive proof of an assertion can start from

reflective principles that are not given in advance but are formulated in the
context of the assertion in question and are then seen to be obviously
true. Cellucci rejects the common view for two reasons: 1) the process by
which one finds a hypothesis that is capable of solving a mathematical
problem is a proof process, but not a deductive one; 2) there is no rational
way of knowing whether primitive premisses are true; the initial premisses
of a deductive proof are only more or less plausible hypotheses. Prawitz
agrees that there is a heuristic phase in the solution of a mathematical
problem in which guesses are made and different strategies are tried. The
search for suitable premisses from which it would be possible to deduce an
answer can to some extent be described as a rule-governed process, and
we may choose to call it proving. But there are good arguments against
stretching the term proof or inference that far. So, (1) above is partly, at
least, a question of terminology and the main argument for the
inadequacy of the deductive method is (2). Prawitz thinks, however, that
argument (2) goes too far. He contends that we often do know that the
premises of deductive inferences are true (and not only plausible). We can
know the truth of the premises in three ways. A fallible method [] for
getting to know that the initial premisses are true [is] the analytic method
as Cellucci describes it, or just careful arguments pro and con. But
Prawitz thinks that we can also find conclusive grounds for the initial
premises [] in virtue of what one takes the involved concepts to mean.
A third way of knowing that certain premisses are true is by means of
deductive inference. The conclusion of one deductive inference can
become the premiss of another deductive inference. Prawitz is aware that,
on the basis of a philosophical dictum [] that the content of the
conclusion of a deductive inference is already contained in the content of
the premisses, some philosophers, including Cellucci, claim that
deductive inference cannot generate new knowledge. Prawitz thinks that
the notion of a deductively valid inference ought to be analysed in a new
way, not in terms of model-theoretic logical consequence. He has
proposed a new analysis of this notion in terms of conclusive grounds.
According to Prawitzs analysis, if an inference is deductively valid, it can
provide new knowledge: it yields a conclusive ground for the conclusion
when conclusive grounds for the premises are given. The general idea that
mathematical knowledge is obtained deductively, Prawitz writes, can be
vindicated by showing that for all deductive inferences used in
mathematics there are operations that transform conclusive grounds for the
premises to a conclusive ground for the conclusion. This is a project of
whose realization, Prawitz admits, he has given here only a hint. In the
 From a Heuristic Point of View xiii

closing sentences he remarks, however, that it is an open question whether

the project outlined can be carried through and that he sees Celluccis
criticism as a stimulating challenge. The status of mathematical
knowledge is also the topic of Empiricism and Experimental Mathematics,
by Gabriele Lolli. Celluccis fallibilist conception of mathematical
knowledge might perhaps be seen as sharing some features of a general
trend in the philosophy of mathematics: neo-empiricism. After dealing
with neo-empiricism and its relation with so-called experimental
mathematics, Lolli argues that the limitations of neo-empiricism result
from considering mathematics only from the point of view of procedures.
Lolli criticizes the idea that mathematics can be characterised by its
method: the limit of the empiricist vision is that of restricting the focus
only on method of validation, thus ending in the stale antithesis between
deduction and induction, between logic and experiment. Even Cellucci,
although he is not an empiricist, suffers from this type of limit; in many
writings (e.g. 1998) he opposes the method of logic to different methods,
in particular the method of analysis, or the open systems. The discussion
of the method is inevitably shallow; knowledge is not given by the
reasoning method, but by the models of reality one builds, and the
concepts one uses. I am sure Cellucci would agree. In the last part of his
essay, Lolli proposes the ideas of the Russian thinker Pavel A. Florenskij
as a model for a philosophy of experimental mathematics.
A discussion of mathematical knowledge and of knowledge in general
often leads to the problem of truth. For example, in his essay Dag Prawitz
implicitly endorses the widespread view that knowledge and truth are
strictly connected notions. Analytical epistemologists think that x knows
that p implies it is true that p. Cellucci, however, proposes a completely
different conception of knowledge: knowledge does not aim at truth, it
aims at plausibility (Cellucci 2008, p. 177). Plausibility is a key notion in
Celluccis heuristic view of knowledge. Knowledge consists of plausible
hypotheses: a hypothesis is plausible if, and only if, it is compatible with
the existing data (p. 177). By compatibility with the existing data,
Cellucci means that if we compare the arguments for and against the
hypothesis based on the existing data, the arguments in favour of the
hypothesis outweigh the arguments against it (pp. 177-8). Truth does not
play any role in science: it is only a chimera that prevents us from
adequately understanding the character of knowledge and therefore
must be disposed of. The question Is Truth a Chimera? provides the
title for Cesare Cozzos contribution to this volume. Cozzo summarizes
Celluccis arguments for the thesis that truth is a chimera and then raises
four objections to Celluccis views on truth. He argues that, Celluccis
xiv Introduction

arguments notwithstanding, a notion of truth is necessary for the human

activity of problem solving and therefore for an adequate understanding of
the phenomenon of knowledge.

In the essays hitherto considered the contributors agree with some of

Celluccis ideas, but argue in favour of some author, some notion, some
distinction, some standard view, or scientific programme that Cellucci
criticizes, rejects or at least seems to put aside in pursuit of his programme
for a heuristic conception of knowledge. Other essays develop lines of
thought that parallel and sometimes follow Celluccis suggestions for
research and some of the main features of his programme. The latter
essays, collected in Part II, generally concern the project for a logic of
discovery. In particular they deal with the nature of mathematical objects,
the role of ampliative inference and axiomatization, the continuity between
mathematics and the empirical sciences and naturalistic epistemology.
The idea of a logic of discovery is central to the papers by Reuben
Hersh and Emily Grosholz. In his To establish new mathematics, we use
our mental models and build on established mathematics Reuben Hersh
maintains that mathematical knowledge grows by means of two tools:
established mathematics and heuristic reasoning, in particular mental
models of mathematical entities. At the beginning of his paper Hersh
rejoices at the fact that, thanks in part to Celluccis work, the topic of
mathematical practice arrived as a legitimate theme of philosophical
investigation. The question is What do real mathematicians really do?.
A mathematicians proof is not an axiomatic proof. To illustrate his point
Hersh considers Andrew Wiles proof of Fermats Last Theorem (FLT). A
mathematicians proof, like Wiles proof, starts from established
mathematics and establishes new mathematics by means of mental models.
Established mathematics is the body of mathematics that is accepted as
the basis for mathematicians proofs and every mathematician has
complete confidence in it, since confidence in established mathematics is
for a mathematician as indispensable as confidence in the mechanics of a
piano for a piano virtuoso, or confidence in the properties of baseballs and
bats for a big league baseball player. If youre worried about that, you
arent even in the game. Hersh argues that established mathematics is a
historical product, and does not need a foundation, because it is what any
mathematician is trying to build on. The status of established
mathematics is not absolute truth, but rather, warranted assertibility. So
Hersh partly shares Celluccis criticism of the notion of truth, arguing that
questions of truth versus fiction are irrelevant to practice and that
truth in the sense of perfect certainty is unattainable. For Hersh, in
 From a Heuristic Point of View xv

mathematics the true should be understood as well justified or firmly

established, in the sense of John Deweys warranted assertibility. The
second essential ingredient of mathematical knowledge, mental models of
mathematical objects, are constituted by the properties and capabilities
effectively employed in using these objects in proofs and in general in
mathematical practice. Mental models are socially regulated: they are
acquired and developed in the course of a social practice of interaction
with other mathematicians and not simply because the single mathematician
has memorized a formal definition. In this sense, a number theorist
knows what a Galois representation is, [and] knows what a semistable
elliptic curve is: he (or she) knows how to use these objects effectively
because his (or her) mathematical practice has shaped them and molded
them to be congruent or matching to the models possessed by other experts
[in the same field]. Mental models are the candidates for the new
semantics of mathematics. In this sense Hersh characterises mathematical
reasoning as essentially semantic.
The FLT proof is also the starting point of Fermats Last Theorem and
the Logicians by Emily Grosholz, who employs it as an exemplary case
study to show the interplay of two essential tasks in scientific discovery,
namely analysis and reference, which usually generate internally
differentiated texts because thinking requires us to carry out two distinct
though closely related tasks in tandem. Analysis requires the abstract,
more discursive project of theorizing, the search for conditions of
intelligibility of problematic objects or solvability of objective problems,
while reference requires the choice of good predicates, durable
taxonomy, and useful notation and icons. The central claim of her paper
is that productive mathematical discourse must carry out these two distinct
tasks in parallel: more abstract discourse that promotes analysis, and
more concrete discourse (often involving computation or iconic
representations) that enables reference, are typically not the same, so that
the resultant composite text characteristic of successful mathematical
research will thus be heterogeneous and multivalent, a fact that has been
missed by philosophers who begin from the point of view of logic.
According to Grosholz, the integration of the various discourses into a
rational relationship generates the growth of knowledge, and she gives tthe
proof of FLT, as well as some of its proposed logical reconstructions, as
examples. In this sense, the FLT proof suggests that in the first stage,
modular forms are investigated as the objects of reference, and treated
geometrically as holomorphic differentials on a certain Riemann surface,
while elliptic curves are treated as instruments of analysis; and conversely
in the second stage, Wiles proof, elliptic curves serve initially as objects
xvi Introduction

of reference, while modular forms become the instruments of analysis.

Sharing two crucial theses of Celluccis heuristic conception of the nature
of mathematical objects and the axiomatic method, Grosholz argues that in
real mathematics, the discovery, identification, classification and epistemic
stability of objects is controversial and that it is not generally true that
what we know about a mathematical domain can be adequately expressed
by an axiomatized theory in a formal language, nor that the objects of a
mathematical domain can be mustered in a philosophical courtyard,
assigned labels, and treated as a universe of discourse.
A core feature of Celluccis heuristic conception, that is the detection
of inferential means of discovery, is the central theme of Christiaan
Huygenss On reckoning in games of a chance: a case study on
Celluccis heuristic conception of mathematics by Daniel G. Campos. He
offers an insightful discussion of the effectiveness of Celluccis account of
scientific discovery by examining a case study in probability theory.
Campos endorses Celluccis philosophical conception of mathematics as
an open-ended, heuristic practice as opposed to the foundationalist view
of mathematics as a closed-ended body of knowledge that is completely
determined by self-evident axioms though he also briefly mentions a
few examples from the history of mathematics to raise some questions
about Celluccis strong claim that axioms never have a heuristic function
or cannot be regarded as hypotheses. In his essay, he examines the case
of Christiaan Huygenss On Reckoning in Games of Chance (1657) to
show that Celluccis heuristic conception provides an insightful way to
account for Huygenss approach to solving mathematical problems in
probability theory. In particular he shows that Huygenss practice consists
in problem-solving that can be described by the analytic method and its
ampliative inferences to search for hypotheses. More specifically, he
argues that Huygens uses three rulesof those explicitly treated by
Celluccito generate hypotheses, as Huygens employs the heuristic
methods of particularization, generalization, and reduction to solve one of
the main problems in his Reckoning.
In Celluccis programme, the rules for finding hypotheses (such as
those discussed by Campos) are correlated to his characterization of the
naturalistic approach to heuristics, a theme that is the topic of Natural
Mathematics and Natural Logic by Lorenzo Magnani. In this paper
Magnani builds on Celluccis version of the naturalistic approach, and in
particular on the distinction between natural and artificial logic and
mathematics. Magnani aims to provide new insights into distributed
cognition, especially into the role of logical models as forms of cognitive
externalizations of preexistent in-formal human reasoning performances.
 From a Heuristic Point of View xvii

He argues that the ideal of a formal logical deduction is an optical

illusion and that the growth of scientific knowledge relies on the
interplay between internal and external semantic, pragmatic and non-
demonstrative representations. As a consequence, he endorses Celluccis
skeptical conclusion about the superiority of demonstrative over non-
demonstrative reasoning, since to know whether an argument is
demonstrative one must know whether its premises are true. But knowing
whether they are true is generally impossible, as Gdels theorems show.
As a matter of fact, premises in demonstrative arguments have the same
status of the premises of non-demonstrative reasoning and demonstrative
reasoning cannot be more cogent than the premises from which it starts;
the justification of deductive inferences in any absolute sense is
impossible, they can be justified as much, or as little, as non-deductive
ampliative inferences.
A cornerstone of Celluccis work is his contrast between a bottom-up
and a top-down approach to mathematics. Cellucci (2013, p. 10) writes:
most mathematicians follow the top-down approach to mathematics,
which has been the mathematics paradigm for the past one and a half
centuries. According to the top-down approach: 1) a mathematics field is
developed from above, that is, from general principles concerning that
field; 2) it is developed by the axiomatic method, which is a downward
path from principles to conclusions derived deductively from them.
Against the top-down approach, Cellucci argues in favour of a bottom-up
approach, according to which 1) A mathematics field is developed from
below, that is, from problems of that field or from problems of other
mathematics, natural science or social science fields. 2) It is developed by
the analytic method, which is an upward path from problems to hypotheses
derived non-deductively from them (2013, p. 11). The aim of Maria
Carla Galavottis essay, For A Bottom-Up Approach To The Philosophy
Of Science, is to extend Celluccis bottom-up approach [] to the
philosophy of science at large. Galavotti is convinced that a bottom-up
approach is capable of promoting a better understanding of the nature of
scientific knowledge. In her paper she expounds some lines of thought
concerning the philosophy of science and statistical methodology that can
contribute to this extension of the bottom-up approach. In the first place,
Patrick Suppes probabilistic empiricism can be deemed bottom-up,
although this is not the terminology he uses. Suppes holds that the
relation between empirical theories and data calls for a hierarchy of
models characterized by different degrees of abstraction, where there is a
continuous interplay between theoretical and observational model
components. The hierarchy is developed from bottom to top because
xviii Introduction

given a model of the data exhibiting a certain statistical structure of some

phenomenon under investigation a fitting theoretical model is sought.
Suppes rejects a clear-cut distinction between theories and data:
depending on the desired level of abstraction different pieces of
information [] will count as data, and what qualifies as relevant will
inevitably depend on a cluster of context-dependent elements. Suppes
approach is contextual and pluralist: scientific structures admit of a
multiplicity of representations, the choice of which depends on contextual
factors. For Suppes scientific activity is a kind of perpetual problem-
solving and in this respect too it is clear that probabilistic empiricism has
much in common with Celluccis bottom up approach. In his book
Probabilistic Metaphysics Suppes rejects the chimeras of a traditional
conception of science: certainty, determinism, the idea of the unity of
science, the idea that science is converging towards some fixed result that
will give us complete knowledge of the universe. Galavotti compares
Suppes views with Celluccis theses on the chimeras that have prevented
philosophy from adequately understanding the character of knowledge (cf.
Cellucci 2008, pp. 77-8). She concludes that Suppes and Cellucci can be
seen as complementary to the extent that they develop different aspects of
the bottom-up approach. A thesis emphasized by Galavotti is that in the
bottom-up movement each step from problem to model depends on the
context in which the problem arises. Therefore, context constitutes the
bedrock on which the bottom-up approach to the philosophy of science
rests. This thesis is exemplified by Galavottis description of Christian
Hennig's approach to statistics that can be regarded as an expansion of
Celluccis approach to mathematics and Suppes view of philosophy of
science and is further illustrated by her outline of the literature on the
statistical methodology for assessing causal relations. In the last section of
her essay Galavotti points out that any account of the notion of context has
to include (i) the disciplinary framework in which some problem
originates, and more specifically its conceptual reference setting,
compounded by the body of theoretical and methodological knowledge
shared by the scientific community addressing the problem in question,
(ii) the nature and amount of the available evidence (in the case of
statistical data the composition and size of the population from which they
are obtained), and (iii) the aims of a given investigation (explanation or
prediction).
The bottom-up approach and the heuristic perspective challenge the
received view on the characterization of mathematical objects and the
nature of mathematical modelling. This point is a central theme of
Mathematizing risk. A heuristic point of view by Emiliano Ippoliti. In his
 From a Heuristic Point of View xix

paper, Ippoliti endorses Celluccis view on heuristics, examining its

consequences for the mathematization of several phenomenathe
applicability and the effectiveness of mathematics. Using the notion of risk
as an example, Ippoliti shows how the heuristic view accounts for it and its
mathematical treatment. To this end he investigates the main approaches
to risk, namely the probabilistic, the psychological, the fractal and the
evolutionary (not to mention the co-evolutionary), and shows that the lack
of success of the various approaches to the treatment of risk is due to the
ways in which they conceptualize and mathematize it. Then, taking as a
starting point Celluccis characterization of mathematical objects (namely,
mathematical objects are hypotheses tentatively introduced to solve
problems) he sets out to show that the heuristic point of view can offer a
better characterization of risk and can improve the other approaches, but
this requires a different conceptual path, bottom-up, local and oriented
towards problem-solving. He argues that risk is a complex object from a
mathematical point of view, whose identity is continuously open to new
determinations and properties and in this sense, it shows the extent to
which mathematical objects are hypotheses tentatively introduced to solve
problems and that the hypotheses through which mathematical objects
are introduced characterize their identity. The identity of a mathematical
object can be characterized differently by different hypotheses, and this
implies that hypotheses do not characterize the identity of mathematical
objects completely and conclusively, but only partially and provisionally.
For the identity of mathematical objects is always open to receiving new
determinations through interactions between hypotheses and experience
(Cellucci 2013, p. 104).
This characterization of mathematical objects shapes Celluccis
approach to two long-standing issues in the philosophy of mathematics,
namely the ontology of mathematical entities and the relations between
mathematics and empirical science. In particular, the heuristic view argues
both for continuity between mathematics and empirical science and for the
dismissal of the relevance of the ontological issue of mathematical entities.
These two tenets are the starting point for Reflections on the objectivity of
mathematics by Robert Thomas. As regards the first tenet, Thomas
employs the notion of assimilation to argue for the objectivity of
mathematics. His concept of assimilation is roughly that of Piaget in
which new things we meet are assimilated to notions we already have,
which notions in their turn are accommodated to accept the new arrivals.
The assimilation process makes mathematics similar to empirical science,
in the way that new concepts are acquired and made to fit with other
concepts, classes and the corpus of knowledge we already have. As for the
xx Introduction

second tenet, Thomas argues that objectivity does not imply a dependence
on an ontology. More specifically, he maintains that objectivity is
achieved in mathematics by public agreement (including agreements to
differ) on styles of inference and definitions in terms of relations.
Moreover mathematics is about relations and not about mathematical
objects, as the mathematical objects are just things that we wish on the
relations we want to talk about in order to be able to do the talking: the
existence or otherwise of the mathematical entity is completely irrelevant
to the construction of mathematics and its inquiry. Thomas also clarifies
this point by introducing an evolutionary characterization of the role of
mathematics and reasoning in general. He argues that since we must be
able to reason as dependably about what does not existeven in a
mathematical senseas about what does, for instance in reductio proofs,
whether some things exist or not is not of any practical importance. It is
not just in mathematics that we need to be able to reason effectively about
what does not exist; it seems to me that the evolutionary advantage to our
reasoning ability is primarily our capacity for reasoning about the future.

Here the editors end their introduction. We thank all the authors who
have made this volume possible by kindly accepting our request to
participate in the project for a book in honour of Carlo Cellucci and we
thank the Department of Philosophy of Sapienza University of Rome for
financial support. We hope that this book will allow the reader to form an
opinion of the variety of perspectives inspired or challenged by Carlos
work. Thank you, Carlo, for your contribution to philosophy and to our
personal development.

Cesare Cozzo, Emiliano Ippoliti

References
Cellucci, C. (1978). Teoria della dimostrazione, Torino: Boringhieri.
. (1998). Le ragioni della logica, Laterza: Roma-Bari.
. (2002). Filosofia e matematica, Laterza: Roma-Bari.
. (2007). La filosofia della matematica del Novecento, Laterza: Roma-
Bari.
. (2008). Perch ancora la filosofia, Laterza: Roma-Bari.
. (2013). Rethinking Logic: Logic in Relation to Mathematics, Evolution
and Method, Springer: Dordrecht.
 SECTION I

MATHEMATICAL LOGIC,
MATHEMATICAL KNOWLEDGE
AND TRUTH
 CHAPTER ONE

REGELFOLGEN IN WITTGENSTEINS
PHILOSOPHY OF MATHEMATICS AND MIND

ROSARIA EGIDI

Der mathematische Satz hat die Wrde einer Regel.

Das ist wahr daran, da Mathematik Logik ist: sie
bewegt sich in den Regeln unserer Sprache. Und das
gibt ihr ihre besondere Festigkeit, ihre abgesonderte
und unangreifbare Stellung.

Remarks I, 165

SUMMARY The existence of mutual influences between the philosophy of

mathematics and philosophical psychology was widely documented from
Wittgensteins 1930s writings onwards. In particular, his notes on
mathematics in the Proto-Investigations (1937), later included in Part I
of the Remarks on the Foundations of Mathematics, continued the analysis
dedicated to following-a-rule. The paper aims, firstly, to highlight the
Wittgensteinian project for a general examination of normative concepts,
characteristic of mathematics and of rule-governed activities; and secondly,
to exemplify the affinities between mathematical and psychological
methods in the critique of the conceptual confusions lurking in their
languages and in the treatment of key concepts such as rule and proof.

KEYWORDS Wittgenstein, following-a-rule, mathematics, mind, normativity,

proof.

1. The project for a general investigation of normative

contexts
In a recent paper, Carlo Cellucci (2008) returned to one of his favourite
themes: the opposition between the notions of axiomatic and analytic
proof, employing fresh arguments to defend the superiority of the latter.
Though Celluccis notion of analytic differs significantly from the
4 Chapter One

normative concept Wittgenstein assigns to the mathematical proof and

though the vindication of the biological and evolutionary bases of
mathematical procedures is far removed from Wittgensteins anti-
Darwinian attitude, I think that his idea that mathematics is embedded in
the natural history of mankind and bound up with the social and
anthropological forms of life, is a point of agreement for all the arguments
used in the refutation of foundationalist strategies as well as for those
exemplified in axiomatic methods. Prompted by this conviction, I present
here a short examination of some central concepts of Wittgensteins
writings on the foundations of mathematics.
The existence of mutual influences and parallel methods between the
philosophy of mathematics and philosophical psychology is widely
documented in Wittgensteins writings from the 1930s onwards. One of
the most important documents in this sense is taken from the second half
of the so-called Proto-Investigations of 1937, which examines the
philosophical problems of mathematics and follows the first half, devoted
to issues of language, meaning, understanding and following rules. Taken
together, the two parts, along with the 1938 Preface, make up the
Frhfassung of the last version (the Sptfassung composed in 1945-
46) of the Philosophische Untersuchungen,1 published posthumously in
1953 with an English translation and with the title Philosophical
Investigations (hereafter Investigations).2
The closeness, present in the Proto-Investigations, of the theme of
following rules (Regelfolgen) and verbs related to the investigation of
the foundations of mathematics, in the special non-logicist meaning
Wittgenstein assigns to the term, is in no way accidental. It is, as we shall
see, integral to Wittgensteins second philosophy and to the task of
rejecting the hegemonic model of the name-object relation and supporting

1
See the critical-genetic edition including the five known versions of the
Philosophische Untersuchungen in Wittgenstein (2002). On their composition and
for the tables of correspondence between them, see J. Schultes Einleitung and
Anhang. While Wittgenstein included the first half of the Proto-Investigations
in 1-190 of the Philosophische Untersuchungen, with a few variations, the
second half appeared as Part I in his Remarks on the Foundations of Mathematics
(hereafter Remarks). This work contains a selection made by the editors of
Wittgensteins writings on the philosophy of mathematics composed between 1937
and 1944, the year in which he stopped writing about this topic and turned his
attention exclusively to the philosophy of mind and related issues up until his death
in 1951.
2
With the reference to Investigations I mean only Part I of the work.
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 5

the plurality of language-games, hence avoiding the drawbacks of the

one-sided diet which in his opinion was the main cause of the
philosophical disease (Investigations, 593). Mathematics or rather
mathematical logic is also plagued, according to Wittgenstein, by the same
disease, so that the therapeutic method he suggests for philosophy will also
be applied to mathematics. His diagnosis is concisely expressed in a
passage of the Remarks that is worth recalling:

Mathematical logic has completely deformed the thinking of

mathematicians and of philosophers, by setting up a superficial
interpretation of the forms of our everyday language as an analysis of the
structures of the facts. Of course in this it has only continued to build on
the Aristotelian logic (V, 48).

In particular, the second half of the Proto-Investigations reveals the

point at which Wittgensteins philosophy of mathematics is grafted onto
the trunk of his analysis of propositions containing verbs such as can,
to be able, to understand and finally to follow-a-rule. This shows
that these two lines of research belong to the same categorical domain, in
that they share the status of normative constructs and do not describe, as in
the case of empirical propositions, states of affairs or mental states (in
other words: physical, psychological or abstract objects), but are rules or
norms of description, consisting in grammatical stipulations about the
meaning to be attributed to the words comprising them. In this way, from
the mid-1930s onwards, Wittgenstein confirms the distinction between
descriptive propositions and rules of description, i.e. the different use
(empirical and normative) of propositions that he would gradually develop
until it became canonical in On Certainty (167).
The second half of the Proto-Investigations is also interesting
because from it emerge traces of the perplexities that in 1937 marked
Wittgensteins analysis of Regelfolgen and momentarily blocked its
continuation, prompting him to shift his attention to mathematical
propositions and procedures. As a matter of fact, this deviation was to
last seven years, during which Wittgenstein devoted himself almost
exclusively to the philosophy of mathematics3, only returning in 1944-45

3
In addition to the hundreds of manuscript pages, most of which we find in the
Remarks, Wittgensteins activity in this field is testified to by the lectures he gave
in English in 1939. These were collected by his pupils, later edited by Cora
Diamond in 1975 and entitled Lectures on the Foundations of Mathematics
(hereafter Lectures).
6 Chapter One

to work with more assured self-awareness on the theme of rules. But

where did what I have called the blocking of Wittgensteins
investigation of following-a-rule occur?
In the Proto-Investigations, after devoting a detailed series of
remarks to the problems arising from the particular language-game of the
pupil asked by the teacher to follow the rule of adding 2 after 1000 in the
succession of natural numbers, i.e. to apply the mathematical formula [
+n], Wittgenstein appears to stumble upon a question asked by his
imaginary interlocutor: But then the steps [from the formula add n to its
application] are not determined by the mathematical formula? and replies
that there is an error in this question (Proto-Investigations, 167-
168; Investigations, 189). Although Wittgenstein reveals the error, he
does not apply himself, as is the case later in the celebrated 191-242 of
the Investigations, to explaining the reasons why it is a mistake to ask how
the rule given by the teacher determines the steps in advance, i.e. the
pupils subsequent applications of the formula, as if the connection
between rule and application existed even before the application takes
place. In the next two remarks Wittgenstein confines himself to reiterating
the argument already adopted in the transition writings and in particular
in the Big Typescript about expecting, wishing, intending, thinking,
believing4 and later extended to following rules, inferring and proofing
according to which these procedures are not the result of mental processes
that magically contain all the steps already preformed, in the same way
as in inferring the conclusions are already logically present in the
premises. In the Remarks this argument will be confirmed when
Wittgenstein denies that the result of a calculation exists even before the
proof has been established and affirms, recalling an old sentence of the
Tractatus, that process and result are equivalent (I, 82). In normative
contexts, and in particular in mathematical procedures, the connections
between process and result do not pre-exist, the steps are not already
taken, are not predetermined (I, 22), but are in some way created ex novo.
The proof makes new connexions, and it creates the concepts of these
connexions (it does not establish that they are there; they do not exist until
it makes them (III, 31). Here Wittgenstein only mentions the basic idea
that the answer to the problem of clarifying what is the criterion for
establishing how a rule is followed and applied should not be looked for in
underlying mental processes, but in the fact that people are trained in such
a way that when ordered to follow a given rule or apply a given formula

4
Cf. Wittgenstein (2005, 76-84, pp. 354-399).
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 7

they always take the same step at the same point. It is therefore on the
basis of our received education and learning that we follow certain rules,
and the criterion for establishing how the rule is applied lies in the way in
which it was taught and how we learnt the technique for using it (Proto-
Investigations, 168; Investigations, 189-90).
At this point of the dialogue, Wittgensteins reflections on Regelfolgen
appear to encounter a setback. From 170 of the Proto-Investigations,
his analysis of understanding and following-a-rule breaks off and his
attention turns to issues of an apparently different nature, concerning the
inexorability of counting and calculating, the difference between
calculation and experiment, the procedures for inferring and proofing,
logical compulsion and the role of invention in mathematics. The
impression that the argument has been abruptly changed appears to be
confirmed by the fact that in the later versions of the Investigations,
composed from 1944 onwards, Wittgenstein deleted the sections devoted
in the Proto-Investigations to the philosophy of mathematics, replacing
them with his more recent reflections on Regelfolgen at the point at which
he had broken them off.5 But on closer inspection, the impression is
deceptive and the brusque transition to other subjects only apparent. In
tackling the mathematical themes indicated above, Wittgenstein actually
completes his analysis of the constructs representing language-games
whose propositions express rule-governed activities.6
We can say that the Proto-Investigations give a germinal idea of the
grand design pursued by Wittgenstein of a systematic inquiry into
language-games in which sentences occur whose meaning is not given by
reference to physical or mental states but depends on the rules of the
grammar of our language. The inquiry starts from the analysis of the
propositions we call normative, including, according to Wittgenstein,

5
See in Wittgenstein (2002) the intermediate version (Zwischenfassung), 190
and the final version (Sptfassung), 191. In these two versions the treatment of
Regelfolgen is followed by a series of remarks on the critique of private language
which is incorporated into the theme of the normative contexts insofar as, in my
opinion, it should be read as a detailed objection to those who claim that language
can be built on entirely subjective and therefore private bases. This would
contradict the assumption Wittgenstein has just made that what people say does
not depend on the agreement of individual opinions but on shared practices and
interpersonal conventions, in other words on what Wittgenstein comprises under
the concept of form of life (Investigations, 241).
6
An extensive and well-documented treatment of the theme of rules is contained in
volume II of Baker & Hackers commentary (1985).
8 Chapter One

both sentences containing verbs such as can, to be able, to understand,

to follow-a-rule, and mathematical propositions. But in the subsequent
versions of the Investigations the field of inquiry gradually widens. For
example, the 1944 intermediate version includes the analysis of language
of sense data, thoughts and representations, and the 1945-46 late version
extends to the multifarious forms of intentional contexts. The mere fact
that in the Proto-Investigations he chose an arithmetical example such as
the succession of natural numbers to clarify the rule-following procedure
shows us the link between the two themes and clarifies Wittgensteins
original idea that they should both be addressed in the same work. This is
testified to by the inclusion, in the two Prefaces to the Investigations of
1938 and 1945,7 of the foundations of mathematics in the list of subjects
that will be explored in the work. Moreover, the very fact that
Wittgenstein did not complete the last version of the Investigations
supports the hypothesis that he may not have abandoned his original idea
and that, if he had had time, he would have completed the treatment of the
variety of language-games and added to that of intentional contexts,
contained in the last one hundred and seventy sections or so of the
Investigations, the analysis of mathematical propositions.

2. Mathematical propositions as rules

The manuscripts of 1943-44, the bulk of which are included in Parts VI
and VII of the Remarks, offer a mature exposition of what I have called
the grafting of the philosophy of mathematics onto the trunk of
Regelfolgen. These contain in nuce the arguments that will be developed in
185-242 of the Investigations and illuminate a variety of aspects. In
particular, Wittgenstein highlights the typical properties that mathematical
concepts and procedures share with all normative contexts, conferring on
them a status that preserves them from the conceptual confusion stemming
from traditional views. The first of these properties is that according to
which mathematical propositions do not have a descriptive status:

There is no doubt at all he states that in certain language-games

mathematical propositions play the part of rules of description, as opposed
to descriptive propositions (Remarks VII, 6a).

The distinction between rules and descriptions is in fact one of the most

7
Cf. Wittgenstein (2002, pp. 207-9; 565-68).
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 9

important consequences of Wittgensteins critique, from the transition

writings onwards, of the denominative theories of meaning, exemplified
by the so-called Augustinian conception of language and which in the
Remarks is extended to include the mathematical Platonism inherent in the
logic of Frege-Russell. The circumstance that within this logic
mathematics is regarded as the natural history of mathematical objects,
which introduces us to the mysteries of the mathematical world is
properly the aspect against which Wittgenstein says I want to give a
warning (II, 40). Even the very mysteriousness of a reference to such a
world suggests the comparison of mathematics with alchemy:

It is already mathematical alchemy, that mathematical propositions are

regarded as statements about mathematical objects, and mathematics as
the exploration of these objects? [] All that I can do, is to shew an easy
escape from this obscurity and this glitter of the concepts (V, 16 b, e).

Wittgensteins critique begins with the rejection of the logistic ideal of

founding mathematics on logical objects and establishes itself as an
attempt to replace the task of a foundation of mathematical propositions
with a clarification of their grammar (VII, 16a). Unlike propositions
that describe objects, be they physical, psychological or abstract,
mathematical propositions are not genuine propositions with an objective
reference point but derive their meaning from the system of rules to which
they belong; they do not have a verifiable cognitive status, and are neither
true nor false but are the fruit of grammatical stipulations that do not obey
the truth-functional logic but have a normative status, of a conventional
nature, insofar as they are expressions of rule-governed activities.
In this sense, the attribution of a non-descriptive status to mathematical
propositions appears to exhibit a special affinity in Wittgensteins
arguments against mathematical Platonism with formalistic concepts of
mathematics and in particular, as F. Mhlhlzer points out (2008; 2010,
pp. 29ff, 72ff), with the mathematician Johannes Thomae, whom Frege
had criticized in the Grundgesetze der Arithmetik (II, pp. 98ff). According
to Thomae, mathematical calculations do not describe anything but are
pure games of marks on paper, manipulations of symbols, devoid of any
referential meaning. Wittgensteins normative conception of mathematics,
however, differs substantially from the formalistic theory since
mathematical constructions, on a par with codes, rules of law, rules of
calculation, and, as we shall see, mathematical proofs, cannot be reduced
to a mere manipulation of signs but have practical needs and ends and
consist in the possibility of their application a circumstance that is not
10 Chapter One

contemplated in formalistic theories and for which Wittgenstein also

criticises Russell for not having taken it into sufficient consideration
(Remarks III, 29d). Despite having no connection with facts, which is by
contrast the case for descriptive propositions, normative constructs have a
pragmatic connotation because they refer to actions and therefore fall into
the domain of the phenomena of doing (Phnomene des Tuns). We can
therefore conclude that the pragmatic or applicative dimension of
normative contexts enables us to clarify that mathematical propositions
can in no sense be assimilated to mere manipulations of symbols in the
style of formalistic conceptions. In this way, Wittgenstein recovers a
constructive role for mathematics that links it to the creation of shared
customs, to consolidated practices and techniques based on the learning
and use of mathematical operations.
The pragmatic and applicative dimension that Wittgenstein attributes
to normative contexts, and therefore also to mathematical propositions,
finds further confirmation in the idea that all rule-guided activities can be
assimilated to creative acts (schpferische Akten) and that the most
important mathematical procedures, such as proofs, are aimed at the
creation of ever new techniques of proof. It is obvious that this
Wittgensteinian conception whereby mathematics is a motley of techniques
of proof. And upon this is based its manifold applicability and its
importance (III, 46a) shows his sharp opposition to the foundationalist
theories that reduce mathematics to the unique and exclusive model of
logic.8 The well-known pronouncements of the Remarks lead back to this
creative aspect, according to which mathematical concepts and procedures,
lacking any referential meaning, do not belong to the context of discovery
(Entdeckung) but rather to that of invention (Erfindung) (I, 168,
Appendix II, 2; II, 38; V, 11f).
If it is true that the applicative dimension of normative contexts confers
on them the dignity of procedures that are different from mere
manipulations of signs with no objective content, this does not, however,
seem a sufficient requisite to protect the language-games in which these
contexts recur from the intrusion of subjectivist elements that could expose
the connection between the expression of a rule and its application to
sceptical doubt; in the case in point between a mathematical construct and
the result of calculations and measurements (Investigations, 242). Doubt
is thereby cast on the logically necessary nature of the connection, which
marks an important part of our lifes activities, and by virtue of which

8
Cf. Mhlhlzer (2010, pp. 309ff).
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 11

one insists inexorably on the fact that we all say two after one and
three after two, and so on (Remarks I, 4). The commonly held claim
that the connection must be guaranteed by some kind of interpretation,
that there are multiple and different ways of interpreting the rule and that it
must be interpreted before it is applied, seems to lead inevitably to a
paradox, often called the sceptical paradox. How can I follow a rule,
when after all whatever I do can be interpreted as following it? (VI, 38f-
g). Wittgenstein is opposed to this idea, and determined to dismantle the
basic misunderstanding on which the paradox is based.
In the lectures of 1939 Wittgenstein clarifies that the transition from
the rule to its application does not depend on an interpretation:

Suppose someone said, Surely the use I make of the rule for continuing
the series depends on the interpretation I make of the rule or the meaning I
give it. But is it ones criterion for meaning a certain thing by a rule the
using of the rule in a certain way, or is it a picture or another rule or
something of the sort?. In that case, it is still a symbol which can be
reinterpreted in any way whatsoever (Lectures, p. 183).

If following rules were determined by an interpretation, every rule-

following activity would depend on a subjective act such as believing or
thinking you are following-a-rule; in other words it would be the result of
an individual and private experience, made by a single person and perhaps
once in their lifetime, with the result that any action by any agent, whether
in accordance with the rule, or in contradiction to it, can be made to count
as its application. A similar consequence appears to confer a markedly
sceptical connotation on Regelfolgen. If any way of acting can be
determined by a rule, the very concept of following-a-rule becomes
hollow: any result can be in accord with the rule, which results in precisely
that paradoxical situation whereby no rule is to be followed. It is really the
interpretative background which according to Wittgenstein becomes
responsible for the sceptical paradox: he contrasts this with the normative
background, which dissociates following-a-rule from believing or
thinking it is being followed, neutralizing the paradox and thereby
preserving normative contexts from the threat of scepticism.
But if following-a-rule is not an interpretation, on what basis should a
person behave and on what basis would we be willing to say that a person
heeds a sign, executes a calculation, obeys an order, and so on?
Wittgensteins response can be summed up in a few words: it is on the
basis of a practice (Praxis) that a person behaves, and accordingly the
concept of Regelfolgen relies on habits, customs, institutions, on shared
behaviour, consolidated practices and ultimately on practical requirements
12 Chapter One

(Investigations, 198). To those who ask where the connection between

the rule and its execution is made, between the calculation and the result,
and what preserves it from subjective discretion and gives it instead a
character of necessity, Wittgenstein responds that the answer is contained
in the list of rules of the game, in the teaching of it, in the day-to-day
practice of playing (Remarks I, 130; Investigations, 197).
According to Wittgenstein there is therefore a way of conceiving
following-a-rule that is not an interpretation (Investigations, 201) but a
practice, a way that eliminates any reference to elements that could be
called individualistic, such as opinions, interpretations, intuitions, beliefs
and which therefore removes the grammar of Regelfolgen from the
jurisdiction of subjective practices, individual decisions, arbitrary
interpretations, placing it in contexts ruled by shared practices, habitual
uses and customs. As will be confirmed in the Investigations, the
language-game of following rules is absolutely different from the
subjective one of believing or thinking you are following them, insofar as
it precludes the possibility of a rule being obeyed privately:

And hence also obeying a rule is a practice. And to think one is obeying a
rule is not to obey a rule. Hence it is not possible to obey a rule privately:
otherwise thinking one was obeying a rule would be the same think as
obeying it (202).

Rather, the expressions of rule-governed activities occur in contexts that

prescribe their application once and for all and manifest themselves in
what we call obeying the rule and going against it in actual cases
(201); in other words, they are expressions of a highly generalized nature,
not an empirical but a higher kind of generality, similar to that of the laws
of logic.
In Wittgensteins concept of normative contexts there are, if I am not
mistaken, echoes of the Fregean distinction between the domain of truths
bearing on logic from that of the doxa, which concerns instead the
menschliches Frwahrhalten, or the set of expressions relating to beliefs,
opinions and interpretations that individuals hold to be true (Frege 1893,
Einleitung, pp. XVI-XVII). And again if I am not mistaken, Wittgensteins
concept is also a useful argument both for showing how normative
contexts are in principle immune from scepticism, and for avoiding these
being associated with the so-called community view of following-a-rule,
often indicated, in the wake of Kripkes conception, as the way out of the
sceptical paradox. It is not based on the majority of human opinions,
beliefs and interpretations that the contexts guided by rules and among
them mathematical propositions take on what Wittgenstein calls their
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 13

peculiar solidity, their unassailable position, set apart (Remarks I,

165). It is rather by virtue of the intrinsic normative character that
qualifies these contexts, of the regularity of the language-games of which
they are a part, that the condition is realized whereby we can agree on our
judgments, i.e. on establishing what is true and false.

3. The proof as part of an institution

The language-games whose sentences do not coincide with descriptive
propositions but express rules of description are the subject of
Wittgensteins reflections on the main subjects which he proposes to
address in the manuscripts forming the core of the Investigations. While in
the 1938 Preface to the Proto-Investigations he mentions alongside the
concepts of meaning, understanding, proposition and logic the
foundations of mathematics, sense-data, the opposition between idealism
and realism, and other things (Wittgenstein 2002, p. 207), in the 1945
Preface only states of consciousness (Bewutseinzustnde), and other
things are indicated after the foundations of mathematics among the
topics in the programme (Investigations, p. vi). Within the broader concept
of states of consciousness Wittgenstein includes statements on
sensations, on which he had written in the Big Typescript in the context of
his refutation of the theses supported by idealists and realists.9 Despite
Wittgensteins failure to complete his programme and, as we know, the
separate publication of the writings on mathematics, he confirms his idea
that mathematical propositions and states of consciousness share the status
of non-descriptive propositions, whose meaning, as in the case of
normative contexts, is given by the rules prescribing the use of the words
which comprise them. The further specifications of the difference between
descriptive propositions and rules of description illustrated in the Remarks
and in the versions of the Investigations in the 1940s, build on and develop
the view of the analogy of concepts and methods that, according to the
pronouncement later expressed in the well-known passage of Part II of the
Investigations (1949), links the inquiry into the foundations of
mathematics to that into psychology, a passage often quoted but rarely
considered in light of the themes that provide a background for the
philosophical undertaking of the late Wittgenstein:

An investigation is possible in connection with mathematics which is

9
Cf. Wittgenstein (2005, 101-107, pp. 487-527).
14 Chapter One

entirely analogous to our investigation of psychology. It is just as little a

mathematical investigation as the other is a psychological one. It will not
contain calculations, so it is not for example logistic. It might deserve the
name of an investigation of the foundations of mathematics
(Investigations, II, xiv, p. 232).

In the same way that in Wittgensteins philosophical psychology it is the

concept of following-a-rule that exemplifies the main properties of
normative constructs, in the philosophy of mathematics it is the concept of
mathematical proof that highlights the most subtle facets of these
constructs. The intersection of the two concepts of Regelfolgen and
mathematischer Beweis is one of the focal points of the Remarks, as is the
exposition, in part contemporary, documented in the Lectures of 1939.
In the Remarks, which collect only in part and on the basis of the
editors personal selection criteria the wealth of material from the
Wittgensteinian manuscripts dedicated to mathematics in the years 1937-
44, the sections on the issue of mathematical proof are certainly among
those that occupy the most space and that received most attention from
Wittgenstein. Within the general conception that mathematics is
normative and that it forms a network of norms (Remarks VII, 61m,
67d), the proof is undoubtedly the normative procedure par excellence,
since in the general context of Wittgensteins late philosophy it plays a
role of paramount importance, which sheds light not only on his analysis
of mathematical concepts but also on the entire conception of normativity
developed in his late work. Unfortunately, the fragmentary, unsystematic
and frequently inaccurate character of Wittgensteins writings on
mathematics a primary cause of the profound misunderstandings and
criticisms aimed at these writings since the publication of the Remarks in
1956 does not give a sufficient account of the conceptions underlying
them, except when seen within the broader and more refined analyses of
normative contexts contained in the Investigations. The fact that only the
writings in Part I of the Remarks were drawn from typewritten versions,
and therefore elaborated on and corrected by the author, should also not be
underestimated, while the texts from the other six Parts were selected from
manuscripts that Wittgenstein did not revise.
Part I of the Remarks, the texts of which, as we know, stem from the
second half of the Proto-Investigations, contain the earlier writings
collected in the work, the bulk of which belongs to 1937. Together with
the Preface of 1938 and the three Appendices, Part I is an invaluable
introduction to the arguments developed later on. However, the more
extensive and mature treatment of the concept of mathematical proof is
developed in 21-44 of Part III of the Remarks, which includes texts
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 15

dating from about two years after Part I, as well as in various passages of
the other Parts from the later manuscripts of 1943-44. In the pages that
follow, I aim to summarize a variety of arguments dedicated by
Wittgenstein to mathematical proof.
The recurring theme in the analysis of the proof is the difference
between the empirical and normative use of propositions, a difference that
is reflected in the version of the relationship between experiment and
proof proposed in the Remarks.10 This gives an account of the general
features assigned by Wittgenstein to the concept of mathematical proof
and also clarifies the meaning of the multifaceted and often metaphorical
epithets of paradigm, model and their synonyms (yardstick, guide-
line, track, path, road), as well as of picture of an experiment,
perspicuous procedure, instrument, creation, and finally institution,
used in the Remarks and Lectures to designate the various connotations of
the proofing procedure. The thesis that the proof is not or does not serve
as an experiment tends to confirm the idea that the inquiry into the
foundations of mathematics, i.e. into the grammar of fundamental
concepts of mathematics, entails a metabasis from objects to concepts
(Remarks V, 9i), and hence from the context of facts to that of norms,
from descriptive propositions to rules of description.11 The concept of
experiment belongs to the former context, that of proof to the latter:
the former has to do with the facts, of what is the case, and it shows us
how they are, the latter has to do with the world of rules, hence of what
ought to be. The normative or, we might say, deontic character of
proofing is summed up in the sentences that in Part III of the Remarks
epitomize the Wittgensteinian assumption, almost in the form of maxims
enclosed in quotation marks: A proof ought to show not merely that this
is how it is, but this is how it has to be (III, 9a) and again Proof must
be a procedure of which I say: Yes, this is how it has to be; this must come
out if I proceed according to this rule (III, 23b; III, 30a, b, 39c, 55a).
In Part I of the Remarks (25-27) and in the Lectures (pp. 71-77) an
effective example is given that makes explicit the meaning of the two
maxims and of the dichotomy is-ought/is-must, in other words the
different use of empirical and proved propositions. The example is
intended to show that the correspondence between a certain pattern (a) of

10
The implicit rejection of the empiricist conception of mathematics, embedded in
Wittgensteins analysis of this relationship, is pointed out by Wright (1980, ch.
XVII).
11
Cf. Frascolla (1994, ch. 3; 2001, pp. 181ff).
16 Chapter One

five strokes and pattern (b) of the angles of a pentacle has two uses, i.e. it
can be treated as an experiment and as a proof.12 The equal number of
elements recurring in (a) and in (b) can be the experimental result of
counting, which is not yet a proof; given that it is subject to errors and can
be submitted to verification this result counts as an experiment, which is
temporarily determined and therefore falls within the domain of empirical
propositions, of the to be. How, then, do I convince myself
Wittgenstein wonders of the correctness of an experimental result, i.e. of
the result of a calculation? Drawing projection-lines between the strokes
and angles of the pentacle I obtain the figure (c) of a five-pointed star with
five threadlike appendages, observing which I become convinced and from
which I deduce that the two patterns are like-numbered. If I call hand
(H) pattern (a) and pentacle (P) pattern (b), I can consider figure (c) a
mathematical proof and the proposition that H has so many strokes as P
has angles a proved proposition (Remarks I, 27). The proved proposition
is therefore: From now on an H and a P are called the same in
number, a proposition that does express what I am going to count as
belonging to the essence of the figures from now on and that, given that it
belongs to the essence can be deposited among the paradigms of the
language (32). In light of this proof, one is therefore induced to say
Yes, this is how it has to be; I must fix the use of my language in this
way (30b) and to file the proved proposition in, so to speak, the archives
of language (29f).13 Unlike the result of a calculation, the proved
proposition works like an atemporal rule, which is valid once and for
all, in the sense that it does not admit of exceptions, contemplates just
one possibility, and therefore falls within the domain of must be or
ought to be:

It must be like this, does not mean: it will be like this. On the contrary: it
will be like this chooses between one possibility or another. It must be
like this sees only one possibility (IV, 31a).

12
A reconstruction of Wittgensteins example is given by Frascolla (1994, pp. 135-
36).
13
What Wittgenstein says about the proved proposition will be referred in the
Lectures to calculation: I once said: A calculation could always be laid down in
the archive of measurements. It can be regarded as a picture of an experiment. We
deposit the picture in the archives, and say, This is now regarded as a standard of
comparison by means of which we describe future experiments. It is now the
paradigm with which we compare (pp. 104-12).
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 17

Further qualifications of the mathematical proof as a model or picture

of an experiment (III, 1b, 23c, 24b, 33a, 36b, 39a; VII, 9a; Lectures,
p. 72) or instrument of conviction (Remarks III, 29f, 39c-e, i-j; VII,
72-73) can be more easily understood if we recall the example of figure
(c) and its role as proof. This is a role that consists in giving a new rule for
ascertaining the numerical equality or rather the biunivocal correspondence
(1-1 correlation) between the elements of the two patterns (I, 40), i.e.
according to a mode of representation which differs from experimental
ascertainment and does not rely on the usual instruments to verify facts,
but brings into play categories such as Bildhaftigkeit, berzeugung and
anschauliche Vorgnge, from which any experimental element disappears
(das Experimentalhafte verschwindet, I, 80) and where causality
plays no part (IV, 41a; VII, 74b). But perhaps the most substantial and
articulate qualification, related to the properties mentioned so far, is the
perspicuous or surveyable or plain to view character which
according to Wittgenstein pertains to the concept of proof (I, 154; III,
1-2, 21-44, 55c; IV, 41a; VII, 20f).
The idea of the perspicuity of the mathematical proof is epitomized by
two statements that respectively open and conclude Wittgensteins
treatment of the theme and are presented, once again, in the form of
maxims enclosed in quotes: A mathematical proof must be perspicuous
(bersichtlich) (III, 1) and Proof must be surveyable (bersehbar)
(55c). I will omit here the details of the main treatment contained in
21-4414 and examine instead the connection between the perspicuity of
the proof and the proof as the image of an experiment, which Wittgenstein
himself recalls in an annotation in parentheses in 1b with reference to his
old example of figure (c) in 25-27 of Part I of the Remarks. Defining the
proof as perspicuous adds an essential specification to the idea of the
image of an experiment, whereby it must be an easy, certain and exact
reproduction, which is plain to see and ensures a new vision of the
experiments result, a vision, so to speak, from on high, similar to that
provided not by a photograph but by an aerial map, which disregards the
reproduction of particular objects and focuses instead on the internal
relationships between them.15 In this sense the proof is a perspicuous
representation of the experiment, which shows and resolves uncertainties

14
The conceptual density of the theme as well as the controversial aspects it
exhibits vis vis the Principia Mathematica and related systems is highlighted in
Mhlhlzers commentary (2010, pp. 102-196).
15
On proof as an internal relation see the Remarks VII, 8a.
18 Chapter One

and errors, an instrument whose use Wittgenstein will transfer in the

Investigations from mathematics to philosophy, to the well-known concept
of bersichtliche Darstellung, which designates the new way of looking at
things and dispels the philosophical problems which take the form of I
dont know my way out (Ich kenne mich nicht aus) (123).
In so far as it is perspicuous, the proof is elevated to the rank of
paradigm, of yardstick, which must be followed for future experiments or
applications, as if in that way a track had been laid down in language. In
this sense, the normative status of the proof is confirmed: the perspicuity
of the proof means precisely that this is not an experiment, because it gives
us the reason for saying that this must be the result (Remarks III, 39c).
But given that it is perspicuous, the proof also constitutes a new way of
seeing the connections and, unlike an experiment, is a rule of conceptual
change, which prescribes how to operate with concepts in a new way
(VII, 45k), i.e. a Begriffsnderung that remodels our way of seeing (IV,
30i-j). Also linked to this new way of operating with concepts, which is
typical of mathematics and, according to Wittgenstein, of its intrinsic
attitude towards invention, is the creative nature of the proof, to which he
returns in many passages of the Remarks to emphasize the distance that
separates his conception from the foundationalist one:

The idea that the proof creates a new concept might also be roughly put as
follows: the proof is not its foundations plus the rules of inference, but a
new building although it is an example of such and such a style. A
proof is a new paradigm. (III, 41g).

The perspective supporting the analysis of the proof in 21-44 had been
effectively outlined in a previous passage of the Remarks:

When I said that a proof introduces a new concept, I meant something like:
the proof puts a new paradigm among the paradigms of the language; []
One would like to say: the proof changes the grammar of our language,
changes our concepts. It makes new connections, and it creates the
concepts of these connections. (It does not establish that they are there;
they do not exist until it makes them.) (III, 31b).16

The remark in parentheses concluding this passage deserves particular

attention. Here, too, there is an allusion to the anti-foundationalist attitude

16
See also III, 41g-i; IV, 45b.
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 19

pervading Wittgensteins alternative to Russells inquiry into the

foundations of mathematics, which is the leitmotif of the Remarks, and
inspires the task of substituting for the investigation of the connections
which, so to speak, already exist, the creation of new connections that do
not exist before the proof has introduced them.
In the context of the crucial 21-44 there is a final qualification of the
concept of mathematical proof, which is given this time in interrogative
form and which, unlike the others previously examined, is not subject to
further scrutiny in the text: Is a proof not also part of an institution in this
way? (36b). Wittgenstein replies with a rhetorical question to the
previous, also rhetorical, question about whether the concept of proof, on a
par with that of a standard metre, can be learnt without knowing the
institution of measuring and its connexion with the standard metre
(36a). The concept of institution is not new to Wittgensteins
philosophical lexicon. From the manuscripts from the early 1930s it is
used to qualify the grammar of a language or a game or a rule as generally
recognized institutions (Mss 110, p.143; 164, p. 95).17 Only in the writings
after 1939 does it appear to have taken on the more specific meaning it has
in the Remarks and later in the Investigations: without an institution the
rules are as if suspended in the air, because the institution of their
application is missing (Ms 129, p.183).18. The same words are used in the
Investigations to deny that an interpretation is required and a subjective
and arbitrary criterion is needed for knowing if a rule has been applied
correctly (198). The basis comprising uses, customs and established
practices which provides the background and gives meaning and validity
to the application of rules is not an interpretation but an institution. In this
sense every rule-guided activity and therefore, according to Wittgenstein,
also a mathematical proof is part of or participates in an institution.
The precise meaning of institution, its role in applying rules, obeying
orders, performing proofs, and more in general the nature of its role that is
somehow compulsive in determining human actions, is not stated in Part
III of the Remarks; clarifying this is the task that Wittgenstein will
undertake in the Investigations, when he completes his analysis of
Regelfolgen. In the context of his reflections on the philosophy of
mathematics, Wittgenstein perhaps believed that it would have been

17
Cf. Wittgensteins Nachlass. The Bergen Electronic Edition. Ed. by the
Wittgenstein Archives at the University of Bergen. Oxford: Oxford University
Press, 2000.
18
Ibid.
20 Chapter One

sufficient to underscore the institutional nature of the mathematical proof

by confirming its normative status, without further comments. However,
various passages scattered here and there in the Remarks appear intent on
clarifying that the distinctive trait of an institution is that of being the point
of convergence at which consensus is achieved, i.e. the agreement
(bereinstimmung) of persons who are members of a community is an
essential part of human action, not excluding mathematical operating (III,
67a-c). Indeed, Wittgenstein insists on the correlation between
agreement and following rules: the word agreement and the word
rule are related, they are cousins. The phenomenon of agreement and of
acting according to a rule hang together (VI, 41; Lectures, p. 83).
Acting according to a rule and the agreement or consensus behind it is
not, however, sufficient to guarantee the correct outcome of every kind of
acting and the absence of this guarantee is responsible for the so-called
scepticism of the rules which, as Wittgenstein will show in the
Investigations (199-202), has paradoxical consequences. In order to
avoid the sceptical paradox the agreement must take an institutionalized
form, which confers on every rule-guided activity a role that protects it
from the risk of being considered a private matter, which can be
undertaken by just one human being and just once in his life. But
what about this consensus doesnt it mean that one human being by
himself could not calculate? Well, one human being could at any rate not
calculate just once in his life (Remarks III, 67c). In conclusion, the
assignment of an institutional character to every rule-guided activity
means that the criterion for the agreement cannot be given only with
reference to the opinions of a majority of people, but depends on an
agreement of ratifications (VII, 9a-d), on an agreement in judgments, i.e.
in standards for judging what is true or false (VI, 39e; Investigations,
242). Certainly, the qualification of the mathematical proof as part of an
institution is little more than hinted at, and will receive much more
attention in the Investigations, where the question will be highlighted
fully. This does not mean that Wittgensteins writings on mathematics are
only a prologue to that important part of the philosophical psychology
which in his second masterpiece he reserves for normative contexts, but
remains, even in the fragmentary form in which it has been handed down
to us, an integral part of it.
 Regelfolgen in Wittgensteins Philosophy of Mathematics and Mind 21

References
Baker G.P. & Hacker P.M.S. (1985). Wittgenstein: Rules, Grammar and
Necessity. An Analytical Commentary on the Philosophical
Investigations. Vol. 2, Oxford: Blackwell.
Cellucci C. (2008). Why Proof? What is a Proof? In: Lupacchini R., Corsi
G., eds. Deduction, Computation, Experiment. Exploring the
Effectiveness of Proof. Berlin: Springer: 127.
Frascolla P. (1994). Wittgensteins Philosophy of Mathematics. London-
New York: Routledge.
. (2001). Philosophy of Mathematics. In: Glock H.-J., ed. Wittgenstein:
A Critical Reader. Oxford: Blackwell.
Frege G. (1893). Grundgesetze der Arithmetik, begriffsschriftlich
abgeleitet, Band, I. Jena: Verlag Hermann Pohle.
Mhlhlzer F. (2008). Wittgenstein und der Formalismus. In: Kro M.,
Hrsg. Ein Netz von Normen: Wittgenstein und die Mathematik. Berlin:
Parerga Verlag: 107-48.
. (2010). Braucht die Mathematik eine Grundlegung?. Ein Kommentar
des Teils III von Wittgensteins Bemerkungen ber die Grundlagen
der Mathematik. Frankfurt: Klostermann.
Wittgenstein L. (1953). Philosophische Untersuchungen. Philosophical
Investigations. Ed. by G.E.M. Anscombe and R. Rhees. Oxford:
Blackwell (Italian transl.: Ricerche filosofiche. Torino: Einaudi, 1967).
. (1969). On Certainty. Ed. by G.E.M. Anscombe and G.H. von Wright.
Oxford: Blackwell (Ital. transl.: Della certezza. Torino: Einaudi).
. (1976). Wittgensteins Lectures on the Foundations of Mathematics.
Cambridge 1939. Ed. by C. Diamond. Ithaca, NY: Cornell University
Press (Ital. transl.: Lezioni sui fondamenti della matematica.
Cambridge 1939, Torino: Boringhieri, 1982).
. (19783). Remarks on the Foundations of Mathematics. Ed. by G.H. von
Wright, R. Rhees, G.E.M. Anscombe. Oxford: Blackwell (Ital. transl.:
Osservazioni sopra i fondamenti della matematica. Torino. Einaudi,
1979).
. (2002). Philosophische Untersuchungen. Kritisch-genetische Edition.
Hrsg. v. J. Schulte. Frankfurt: Suhrkamp.
. (2005). The Big Typescript: Ts 213. German-English Scholars
Edition. Ed. and transl. by C.G. Luckhardt, M.A.E. Aue. Oxford:
Blackwell (Ital. transl.: The Big Typescript. Torino: Einaudi, 2002).
Wright C. (1980). Wittgenstein on the Foundations of Mathematics.
Cambridge, MA: Harvard University Press.
 CHAPTER TWO

SERENDIPITY AND MATHEMATICAL LOGIC

DONALD GILLIES

SUMMARY This paper is an attempt to apply the concept of serendipity to

mathematical logic. Serendipity is defined as looking for one thing and
finding another. It is argued that mathematical logic was developed as
part of the search for a foundation, which would render mathematics
absolutely certain. Following Carlo Cellucci, it is then argued that this
foundational quest failed. However, and here serendipity comes in,
mathematical logic proved unexpectedly to be a very useful tool for
computer science. The paper concludes by considering briefly whether the
concept of serendipity could be useful in the context of Carlo Celluccis
programme for developing a logic of mathematical discovery.

KEYWORDS serendipity, mathematical logic, computer science, logic of

discovery, foundationalism

1. Serendipity
There is one definite fact about the term serendipity. It was coined by
Horace Walpole in a letter to his friend Horace Mann, dated 28 January
1754. However, Merton and Barber (2004) shows that the subsequent
history of the term was tortuous and complicated. It has been, and still is,
used in a variety of different senses. In this paper, I will use a definition of
serendipity which, according to Merton and Barber (2004, p. 112) was first
formulated by Edward Solly in 1880 and is cited in the Oxford English
Dictionary. According to this definition, serendipity consists in looking
for one thing and finding another. This definition does have some basis in
Horace Walpoles original letter, for he says there (quoted from Merton
and Barber, 2004, p. 2): you must observe that no discovery of a thing
you are looking for comes under this description. However, serendipity is
24 Chapter Two

sometimes used in broader senses which also have some basis in Horace
Walpoles original letter.
The classic instance of serendipity, as it has just been defined, is
Columbus discovery of America. Columbus was looking for a sea route to
the East Indies obtained by sailing west. However, what he actually found
was a new continent, whose existence was unknown to European
geographers.
Let us now turn to considering how the concept of serendipity might be
applied to the development of mathematical logic. The main ideas of
Mathematical Logic were developed between 1879 and 1931 by a number
of different researchers, of whom Frege, Peano, Russell, Hilbert, and
Gdel are the most famous. Van Heijenoort (1967) is a well-known
collection of papers by the pioneers of mathematical logic. The title of the
collection: From Frege to Gdel. A Source Book in Mathematical Logic,
1879-1931 gives an accurate idea of its content.
How then can the concept of serendipity be applied to mathematical
logic? I have defined serendipity as looking for one thing, and finding
another. To show that serendipity applies to mathematical logic, we have
therefore to show two things: (1) that the pioneers of mathematical logic
were looking for something which they did not find, and (2) that they did
find something different from what they were looking for.
As regards (1), my claim will be that Frege, Russell, Peano, and
Hilbert were looking for a foundation, which would render mathematics
absolutely certain. They thought that mathematical logic was an essential
component of this foundation. We can summarise this by saying that the
pioneers of mathematical logic were looking for a foundation for
mathematics. However, this foundational quest ended in failure.
As regards (2), my claim will be that mathematical logic turned out to
have a surprising and unexpected use. It proved to be a valuable tool for
computer science. Now the pioneers of mathematical logic did not
anticipate this application. They could hardly have done so. There are
arguments among historians as to what should be counted as the first
computer in the modern sense, but no serious candidate for this title was
produced before 1945, that is to say over 15 years after Frege, Russell,
Peano and Hilbert had created mathematical logic. This is why the concept
of serendipity clearly applies to mathematical logic.
The above claims, though without using the concept of serendipity, are
to be found in Doxiadis and Papadimitriou (2009) Logicomix. This is a
truly remarkable attempt to present the history of mathematical logic,
focussing mainly on Russell, in the form of a comic book or graphic novel.
In self-referential fashion, the authors appear themselves as characters in
 Serendipity and Mathematical Logic 25

the book. By the time we have reached p. 303, Apostolos Doxiadis seems
convinced that the foundational quest of Russell et. al. has had an unhappy
ending. However, Christos Papadimitriou says: follow the quest for
ten more years and you get a brand-new, triumphant finale with the
creation of the computer, which is the quests real hero! We are even
promised a further book: The Story of Computers as sequel to
Logicomix. Let us hope this is indeed produced.
Let me now turn to providing arguments for the two parts of my
serendipity claim. I have first to show that mathematical logic failed in its
original aim of providing a foundation for mathematics. I will try to do this
in the next section 2, using mainly the arguments of Cellucci in his (1998),
(2002) and (2007). Then I have to show that mathematical logic turned
out, after all, to be a useful tool for computer science. I will attempt to do
this in section 3.

2. Celluccis Critique of Mathematical Logic

Mathematical logic was developed in the context of two research
programmes which had philosophical motivations. The first of these was
logicism, that is to say the attempt to show that mathematics1 could be
reduced to logic. This was first developed by Frege and then by Russell.
Both Frege and Russell (at least initially) thought that, if logicism were
successful, it would provide a foundation, which would render
mathematics absolutely certain. They thought that the fundamental
principles of logic would be a priori certain, and that the theorems of
mathematics, following from the fundamental principles by strict logical
reasoning, would also be certain.
The second research programme was formalism. In Hilberts version of
this programme, mathematics consisted of a collection of formal systems
in each of which the theorems were deduced from the axioms using
mathematical logic. Hilbert aimed to prove that each of the formal
systems, which comprised mathematics, could be proved consistent using
only the indubitable methods of finitary arithmetic. If he and his school
had succeeded, this would have rendered mathematics absolutely certain in
a slightly different sense from that of the logicist school. Peano, whose
work preceded that of Hilbert, also had a formalist point of view. He and

1
Strictly speaking Frege limited his version of logicism to the claim that arithmetic
can be reduced to logic. Russell later extended logicism to the whole of
mathematics. However, I will ignore this difference in this preliminary exposition.
26 Chapter Two

his followers in their Formulaire de Mathmatiques (1895-1908) tried to

present the whole of mathematics as a collection of formal systems.
However, Peano did not have Hilberts requirement that these formal
systems should be proved consistent.2
In chapters 8, 9 and 11 of his 2002, Cellucci argues that the
foundational programme of Frege and Russell, and that of Hilbert both
ended in failure because of the limitative results of Gdel (1931) and
Tarski (1936). Let me now give an informal exposition of why Gdels
two incompleteness theorems might be taken as having refuted both
logicism and formalism.
The relation of Gdels incompleteness theorems to the logicist
philosophy of mathematics of Russell and Whiteheads Principia
Mathematica is clearly indicated by the title of Gdels 1931 paper which
runs On Formally Undecidable Propositions of Principia Mathematica
and Related Systems I. However, as the title also shows, his results do not
apply just to Principia Mathematica, but to any similar formal system. I
will now state the first of Gdels incompleteness theorems in a later
version proved by Rosser (1936), and sometimes known as the Gdel-
Rosser theorem.
Gdels First Incompleteness Theorem. Given any formal system S,
such that (1) S is consistent and (2) a sufficiently large amount of
arithmetic can be derived in S, then we can find an undecidable proposition
p in S that is to say, a proposition p such that p cannot be proved in S,
and the negation of p (not-p) also cannot be proved in S. p can, however,
be shown to be a true statement of arithmetic by an informal argument
outside S.
Let us see how this theorem applies to Russell and Whiteheads system
Principia Mathematica or PM for short. Russell and Whitehead had
claimed that the axioms of PM were truths of logic. They also thought that
it should be possible to prove any theorem of mathematics within PM. Had
this in fact been true, then it would have been shown that the whole of
mathematics is nothing but an extension of logic. Of course, Russell and
Whitehead could not prove within PM all the theorems of existing
mathematics, let alone of all future mathematics. In their three massive
volumes, however, they did prove a very large number of mathematical
theorems, so that by the end of volume III it did begin to look plausible
that it would be possible to prove any further mathematical theorem within

2
For some further discussion of Peano as a forerunner of Hilberts philosophy of
mathematics, see Gillies, 1982, pp. 69-70.
 Serendipity and Mathematical Logic 27

their formal system. Gdels first incompleteness theorem, however,

showed that, if PM is consistent, then there is a true statement of
arithmetic which cannot be proved within PM. The theorem thus showed
that the logicist philosophy of mathematics of Russell and Whitehead was
not correct. At first sight, it might be thought that the logicist position
might still be rescued by constructing some logicist system S more
powerful than PM, but of course Gdels first incompleteness theorem
would apply to S if S were consistent, and the same objection could be
raised against S as against PM. It thus looks as if Gdels first
incompleteness theorem gives a fatal blow to logicism. To explain the
objection to Hilberts formalism, we will need to state Gdels second
incompleteness theorem.

Gdels Second Incompleteness Theorem Given any formal system S,

such that (1) S is consistent and (2) a sufficiently large amount of
arithmetic can be derived in S, then the consistency of S cannot be proved
within S.

It is easy to see why this shows the impossibility of carrying out

Hilberts programme. Let S be some complicated formal system for some
branch of mathematics e.g. set theory. Certainly we will be able to
formulate very elementary arithmetic within S, and so we will not be able
to prove S consistent using only the methods of elementary arithmetic, as
Hilbert has hoped.
Prima facie then, Gdels incompleteness theorems refute both logicism
and formalism. However, this conclusion is not accepted by all
philosophers of mathematics, and, in the last few decades, there have been
attempts to formulate neo-logicist or neo-formalist positions which are
designed to evade the problems created by Gdels incompleteness
theorems. Cellucci criticizes these new versions of logicism and formalism.
Let us start by considering his treatment of neo-logicism.
Cellucci deals with neo-logicism in his (2007). He criticizes the
version of neo-logicism developed by Hale and Wright in their (2001).
Hale and Wright, like Frege, defend the thesis that the truths of arithmetic
are analytic. However, they use analytic in a slightly broader sense than
Frege. Frege held that a statement is analytic if it is deducible from
primitive truths of logic. Hale and Wright maintain that a statement is
analytic if it is deducible from primitive truths of logic together with
primitive analytic propositions. A primitive analytic proposition is one
which gives a contextual definition of a concept. As far arithmetic is
concerned, the primitive analytic proposition is what they call Humes
Principle (or HP). This can be stated as follows:
28 Chapter Two

NxF(x) = NxG(x) if and only if F G (HP)

where NxF(x) stands for the number associated with the concept F, and F
G means that there is a 1-1 correspondence between the objects which
fall under the concept F and those which fall under the concept G.
Against this neo-logicist position, Cellucci marshals a number of
arguments (2007, pp. 86-89). First of all he argues that it just as much
undermined by Gdels incompleteness theorems as the earlier versions of
logicism. Then he goes on to say that the mathematical concept NxF(x) is
not completely eliminable on this approach, because HP only allows us to
eliminate NxF(x) in contexts of the form NxF(x) = NxG(x). However, if
neo-logicism allows a mathematical concept such as NxF(x) which is not
completely eliminable in favour of purely logical concepts, then it is really
no longer a form of logicism. Cellucci also has a rather telling comment on
the notion of analytic used by Hale and Wright. Cellucci writes (2007, pp.
87-8):

Neo-logicism considers HP to be analytic because it gives a contextual

definition of the concept of number. But then one ought also to consider as
analytic, for example, the principle of inertia: Every body remains in its
state of rest or uniform motion in a straight line unless acted on by a force,
because this gives a contextual definition of the concept of force. But this
sense of analytic takes all philosophical bite from the claim that the truths
of arithmetic are analytic, because the laws of physics would also be
analytic in this sense.

Cellucci has some further arguments against neo-logicism, and overall his
criticism of that position seems to me decisive.
Turning now to neo-formalism, there are really two versions of this
position. The stronger version hopes to rescue Hilberts original version of
formalism. The main proponents of this approach are Simpson (1988)
Partial realizations of Hilberts program, and Detlefsen (1990) On an
alleged refutation of Hilberts program using Gdels first incompleteness
theorem. The titles of these papers clearly indicate the authors intention of
reviving Hilberts programme in some form. Cellucci criticizes Simpson in
his 2002, Chapter 9, Section 5, pp. 78-9, and Detelefsen in the same work,
Chapter 8, Section 4, pp. 66-7. These criticisms seem to me convincing.
There is however a weaker version of neo-formalism which argues that
mathematics should consist of a set of formal systems, but gives up
Hilberts programme of trying to prove the consistency of these formal
systems of mathematics using finitary arithmetic. This point of view has
 Serendipity and Mathematical Logic 29

been most notably defended by Bourbaki. In a sense, it is a return to the

formalism of Peano.
Cellucci discusses Bourbakis position in his 1998, p. 174. As he
points out, Bourbaki, because of Gdels second incompleteness theorem,
completely renounce the attempt to prove that the formal systems of
mathematics are consistent using the methods of finitary arithmetic.
Instead they regard this consistency as an empirical fact. If, for example,
we are considering a version of axiomatic set theory, say ZFC, this is a
theory which was formulated in the 1930s, and within which
mathematicians have now worked for decades without encountering any
contradictions. So ZFC is likely to be consistent. Moreover, even if a
contradiction did appear in ZFC, the axioms of ZFC could no doubt be
quickly altered to repair the damage, just as happened when the
contradictions appeared in the early 20th century.
What does Cellucci say in reply to this? He argues (1998, p. 179): But
the recursive definitions by means of which the terms and formulas of a
formal system are introduced, presuppose the infinite repeatability of
certain operations, and so need an intuition which goes beyond the
empirical. So, Cellucci concludes, the consistency of a formal system
cannot be an empirical fact as Bourbaki claim.
That concludes my account of some of Celluccis criticisms of
mathematical logic. These criticisms do, in my view, establish that
mathematical logic did not succeed in establishing a foundation for
mathematics of the kind that the inventors of mathematical logic had
hoped to create.
While accepting many of Celluccis criticisms of mathematical logic, I
do not want to give the impression that he is an entirely negative thinker.
On the contrary, he has his own programme of developing a logic of
discovery for mathematics which will be quite different in character from
traditional mathematical logic. He makes a most interesting start with this
programme in his 2002, Parts 3 and 4, pp. 143-308, where his proposed
logic of discovery is illustrated by many striking mathematical examples. I
will say a little more about this programme for studying discovery in
mathematics in the final section, but in the next section I will try to
complete my argument for the claim that serendipity applies to the
development of mathematical logic.
30 Chapter Two

3. Mathematical Logic as a Useful Tool for Computer

Science3
Mathematical logic has proved to be a useful tool in many areas of
computer science. I will begin by discussing its use in programming
languages. There is no doubt that mathematical logic exercised a very
considerable influence on the development of such languages. The first
work of mathematical logic was Freges Begriffsschrift of 1879. This is the
first example of a fully formalised language, and so, since programming
languages need to be formalised, it is the precursor of all programming
languages.4 Moreover the logic programming language PROLOG is
directly based on the predicate calculus which Frege introduced in his
1879. PROLOG was intended to use standard Fregean logic. However, it
turned out that it actually used a logic of a different type non-monotonic
logic.5 This is rather characteristic of the way that mathematical logic
came to be applied in computer science. Ideas and theories were taken up
from mathematical logic, but usually they had to be modified in some
important respects before they could be applied in computer science.
Church developed the O-calculus with a view to providing a new
foundation for logic in the style of Russell and Whitehead. However, it
proved very useful for programming languages. It became the basis of
programming languages such as LISP, Miranda, and ML, and indeed is
used as a basic tool for the analysis of other programming languages.

3
The following account is a revised version of my treatments of this problem in
Gillies and Zheng (2001), and Gillies (2002). These in turn made great use of
Davis (1988a & 1988b). I was also able to have discussions with Martin Davis on
the problem when I read an earlier draft of Gillies and Zheng (2001) to the Logic
Club in Berkeley in 1998. The revisions of my earlier treatment are partly due to
some further thoughts of my own on the problem, and also to the impact of two
important works on this subject which have been published in the meantime
namely Numerico (2005) and Priestley (2011). Once again I have benefited from
lengthy discussions with Teresa Numerico and Mark Priestley over the years. I
would also like to thank a number of computer scientists with whom I have
discussed the problem and who make many helpful suggestions which have been
incorporated in my account. These include James Cussens, Mark Gillies, Stephen
Muggleton, David Page, and Ashwin Srinivasan. My earlier account included a
discussion of the influence of the theory of Turing machines on the invention of
the computer, but this has been omitted for reasons of space.
4
I owe this point to Martin Davis. See his 1988b, p. 316.
5
For details, see Gillies, 1996, pp. 72-79.
 Serendipity and Mathematical Logic 31

It could, however, be objected that logic-based languages such as LISP

or PROLOG are little used. However, Priestley has shown in his 2011 that
many ideas from mathematical logic were embodied in standard
programming languages. FORTRAN was the most successful
programming language of the 1950s, and as Priestley points out (2011, p.
200) in FORTRAN expressions were given a formal recursive
definition which was very similar in style to the definitions given in logic
texts of the terms of formal languages. Priestley goes on to comment
(2011, p. 201-2):

By and large, expressions in later languages were defined in a similar way

to Fortran This raises the question of why this particular definition
turned out to be so influential. One possible answer is that it was the
style of the definition that gave rise to its success. Although a number of
writers had perceived a general similarity between logic and programming,
this was the first time that techniques from formal logic had been applied
to a relatively mundane task like syntax definition. As well as providing a
concise and general definition of expressions, this suggested a general
approach to the design of programming languages which appealed to the
authority and established results and techniques of the discipline of logic.

The next important development on Priestleys account was ALGOL 60.

This incorporated many features drawn from logic. It was not itself very
successful in practical terms, but it did exert an enormous influence on the
development of programming languages, and led to new programming
languages incorporating features drawn from mathematical logic. As
Priestley comments (2011, p. 225) what changed the face of
programming was not simply the Algol 60 language, but rather a coherent
and comprehensive research programme within which the Algol 60 report
had the status of a paradigmatic achievement, in the sense defined by the
historian of science Thomas Kuhn.
Priestley also makes an interesting comparison between LISP and
ALGOL. He writes (2011, p. 223): Lisp can fairly be described as a
programming language which to a large extent is based on prior work in
formal logic. Unlike Algol, however, Lisp is not presented as a formal
language.
Let us now turn to the influence of Russells theory of types on
computer science. Russell discovered in 1900 that Freges original version
of logicism was defective because it gave rise to a contradiction, now
known as Russells paradox. Russell still supported logicism, and so
wanted to develop a version of logicism which avoided the paradox. After
exploring a number of possibilities, he opted for basing mathematical logic
on the theory of types, and published a paper developing this view in
32 Chapter Two

1908. The theory of types did not have a great success in the mathematical
community which preferred to resolve the paradox by using the axiomatic
set theory developed by Zermelo, a member of Hilberts school at
Gttingen. Indeed type theory is not taught at all in most mathematics
departments. The situation is quite different in computer science
departments where courses on type theory are a standard part of the
syllabus. This is because the theory of types is now a standard tool of
computer science. Functional programming languages such as Miranda
and ML are usually typed, and indeed some form of typing is incorporated
into most programming languages. It is desirable when specifying a
function such as f(x,y) to specify also the types of its variables x,y,
otherwise errors can be produced by substituting something of the wrong
type for one of the variables which will often produce a nonsensical
answer. Once again it should be stressed that the type theories used in
contemporary computer science are not the same as Russells original type
theory, but are descendants of Russells original system which have been
modified to make them suitable for computer science. An important link in
the chain was Churchs 1940 version of the theory of types which was
developed from Russells theory, and which influenced researchers in
computer science. Davis sums up the situation very well as follows
(1988b, p. 322):

Although the role of a hierarchy of types has remained important in the

foundations of set theory, strong typing has not. It has turned out that one
can function quite well with variables that range over sets of whatever
type. So, Russells ultimate contribution was to programming languages!

Another area of computer science in which mathematical logic has proved

very useful is artificial intelligence. Let us start with automated theorem
proving. Here theorems are first formalised in 1st order predicate calculus.
However, the version of the predicate calculus used is one which was
devised specially for computer science. This is the clausal form of logic
based on the resolution principle which was introduced by Alan Robinson
in 19656. This approach was developed by Larry Wos, George Robinson
(no relation of Alan Robinson!), and William McCune. In 1996, their logic
theorem prover managed to solve an open problem in Boolean algebra by
proving the Robbins conjecture. The final successful search for the proof

6
See Robinson (1965), and some philosophical comments on this version of
predicate calculus in Gillies (2002)
 Serendipity and Mathematical Logic 33

took about 8 days of computer time.7 Many human mathematicians,

including Tarski and his students, had tried unsuccessfully to prove this
conjecture in the preceding sixty years. So this was a remarkable triumph
for automated theorem proving.8
Robinsons clausal form of logic based on the resolution principle has
also proved useful in machine learning. Muggleton had the idea of
inverting Robinsons deductive logic to produce inductive logic
programming. His techniques have led to the discovery of new results in
biochemistry.9
These examples, and more could be given, establish beyond doubt that
mathematical logic has proved to be a very useful tool for computer
science. Yet mathematical logic was not designed for this purpose. In the
next, and last, section of the paper, I will try to explain how this
serendipity came about.

4. Explaining Serendipity
The word serendipity suggests something strange and mysterious. Yet
there is really nothing mysterious about many instances of serendipity as it
has been defined in this paper. Suppose I have mislaid my pen and search
for it in a drawer filled with miscellaneous objects. The pen is not there,
but, at the bottom of the drawer, I find an old notebook which had gone a-
missing more than a year before. This is an example of serendipity since I
was looking for one thing, and found another. However, there is nothing
strange or mysterious about it.
The classic example of serendipity, namely Columbus discovery of
America, was scarcely more mysterious than my everyday example of the
pen and the notebook. Anyone who followed Columbus strategy of
sailing west in order to discover a sea route to the East Indies would
inevitably, if his ship hadnt accidentally sunk, have discovered America.
America was simply there blocking the path, and any mariner following
Columbus search strategy, would have run into it.
On the other hand, the case of mathematical logic does seem more
surprising. The pioneers of mathematical logic such as Frege, Russell,

7
See Mackenzie (2001, p. 89).
8
For further details, see Wos and Pieper (1999). Corfield (2003, pp. 35-79) gives
some valuable philosophical reflections on automated theorem proving, including
the proof of the Robbins conjecture.
9
For further details, see Muggleton (1992) and Gillies (1996, pp. 41-44 & 50-54).
34 Chapter Two

Peano and Hilbert were working on philosophically motivated

programmes designed to provide a foundation for mathematics which
would make the results of mathematics certain. Their foundational
programmes failed, and yet the new concepts and theories which they had
created proved to be invaluable, perhaps indeed essential, tools for the
development of computers. How did this come about?
To answer this question, let us first look at the case of Frege. He was
very concerned to criticize Kants view of arithmetic as synthetic a priori,
and based on intuition. Frege had the alternative logicist view that the
truths of arithmetic could be derived from logical axioms by strictly
logical rules of inference without any appeal to intuition. To demonstrate
this, Frege had to limit himself to simple logical inferences which did not
make any appeal to intuition. As he puts it himself (Frege, 1879, p. 5): to
prevent anything intuitive [Anschauliches] from penetrating here
unnoticed, I had to bend every effort to keep the chain of inferences free of
gaps.
Ordinary language proved to be inadequate for achieving this goal, and
this led Frege to invent his Begriffsschrift, as he himself says (1884, p.
103):

To minimize these drawbacks, I invented my concept writing. It is

designed to produce expressions which are shorter and easier to take in,
and to be operated like a calculus by means of a small number of standard
moves, so that no step is permitted which does not conform to the rules
which are laid down once and for all. It is impossible, therefore, for any
premiss to creep into a proof without being noticed.

Now computers do not of course possess human intuition. So, if a proof is

to be handled by a computer, it must first be formalised in a system like
Freges Begriffsschrift. Thus Freges philosophical programme to prevent
anything intuitive from penetrating here unnoticed led him to create a
language suitable for computers.
Frege was, quite inadvertently of course, taking a step towards the
mechanisation of mathematics. The process of mechanisation in general
takes place in something like the following manner. The starting point is
handicraft production by skilled artisans. The next step is the division of
labour in the workshop in which the production process is broken down
into smaller and simpler steps, and an individual worker carries out only
one such step instead of the process as a whole. Since the individual steps
are now quite simple and straightforward, it becomes possible to get them
carried out by machine, and so production is mechanised.
 Serendipity and Mathematical Logic 35

Turning now to mathematics, if a proof is written out in the

characteristic human style, which is only partially formalised, then its
validity cannot be checked mechanically. One needs a skilled human
mathematician to apply his or her intuition to see whether a particular
line follows from the previous ones. Once a proof has been formalised,
however, it is a purely mechanical matter to check whether the proof is
valid using the prescribed set of rules of inference. Thus Freges work can
be seen as replacing the craft skills of a human mathematician with a
mechanical process.
Frege himself would not have seen his work in this way, yet it should
not be forgotten that Frege, Peano, Russell, Hilbert, etc. lived in a society
in which material production had been very successfully mechanised and
in which there was an ever increasing amount of mental (white collar)
labour. This social situation may have influenced them, albeit at an
unconscious level, towards the project of mechanising thought. Moreover
it was natural that any attempt to mechanise thought should begin with
mathematics, since mathematics, unlike other areas of thought, was
already partially formalised. This observation may partly explain why
philosophy of mathematics held such a central place in philosophy in the
period from the late 19th to the late 20th centuries.
Turning now to the other pioneer of mathematical logic, Giuseppe
Peano, we find a similar picture. Peano does not seem to have been, like
Frege, a strong critic of Kants philosophy of arithmetic, and to have
opposed the use of intuition in mathematics on those grounds. However,
there were other reasons to be sceptical of intuition, coming from within
mathematics itself. In 1872 Weierstrass gave a lecture in which he defined
a function which was continuous but nowhere differentiable. This is
clearly counter-intuitive, and was the first of a series of counter-intuitive
objects discovered by mathematicians in the next few decades. Chapter 12,
pp. 92-100, of Celluccis (2002) is entitled: Intuizione e Mostri (Intuition
and Monsters). It gives a nice account of these mathematical monsters
which contradict intuition, and argues, quite correctly in my view, that
their existence shows that intuition cannot provide a satisfactory
foundation for mathematics. Now Peano discovered one of the most
famous of these monsters, his space-filling curve, and published an article
about it in 1890. The very next year, 1891, Hilbert published an article
developing Peanos work and producing a new space-filling curve. It
seems reasonable to suppose that both Peano and Hilbert became
distrustful of intuition as a result of the counter-intuitive monsters, and so
were led towards a formalist approach to mathematics. Peano eliminated
intuition altogether. Hilbert allowed a weak form of intuition to support
36 Chapter Two

finitary arithmetic, but only at the meta-level. So both of them developed

object-level formal theories, which, because the dependence on intuition
was eliminated, were entirely suitable for use on computers.
Fully formalised mathematical theories may indeed be very suitable for
computers, but they are not very suitable for humans. Although Frege
claimed that his concept writing was designed to produce expressions
which are shorter and easier to take in, in practice formalised expressions
soon become too long for humans to understand, though such length is no
problem for a computer. As we saw earlier, variables of different types are
very suitable for computers, but humans prefer to have variables of a
single type. This was why human mathematicians preferred to solve
Russells paradox using axiomatic set theory, where variables all range
over sets, rather than to use Russells type theory. In general, human
mathematicians need to use intuitive short cuts to make concepts
comprehensible, and cannot work with fully formalised theories.
As we have seen, computers have now started to do mathematics, but,
in so far as mathematics continues to be carried out by humans, some
appeal to intuition is going to be indispensable. This is another reason why
both logicism and formalism failed to give an adequate account of
mathematics, but provided tools very suitable for use by computers.
Frege and Peano were both sceptical of intuition though for different
reasons. Another thing they had in common was that they both developed
an axiomatic-deductive version of arithmetic with the underlying logic
made fully explicit. Before them, geometry had, since the time of Euclid,
been developed as an axiomatic-deductive theory, but arithmetic was
introduced as a calculating tool. The significance of their new
development is that it provided a formal language in which arithmetical
statements could be expressed. This connection between arithmetic and a
corresponding formal language was carried further by Gdel who
introduced the Gdel numbering of formulas, and showed that some
operations with these formulas could be carried out within arithmetic.
These links between arithmetic and a corresponding formal language
were invaluable for developing computers. A common problem in
computing is to write a programme instructing the computer to carry out
an arithmetical calculation. Now such a programme has to give the
instructions in a completely precise and unambiguous fashion. So a formal
language for arithmetic is needed here. However, computers are universal
machines and so we may wish to programme them to carry out deductions
rather than calculations. Here the possibility of representing deductions as
arithmetical calculations is useful. Moreover programmes need to be
stored on the computer, and so coding a programme by its Gdel number
 Serendipity and Mathematical Logic 37

could again be a useful device. So we see once again why ideas developed
in the context of a formalist philosophy of mathematics proved very useful
for computing.
There is however an important difference between formalised Peano
arithmetic and a programme giving instructions about how to carry out an
arithmetical calculation. The formulas of Peano arithmetic are statements
about numbers, whereas the formulas of the computer programme are
orders to perform particular arithmetical operations. This gives part of the
explanation as to why, although computer scientists frequently make use
of ideas and theories from mathematical logic, they nearly always have to
modify these ideas and theories in order to make them applicable to
computing.
This completes my attempted explanation of the serendipity involved
in the development of mathematical logic. However, before closing the
paper, I would like to make an observation about how Celluccis positive
project for a logic of mathematical discovery might be developed. Now
researchers who have studied discovery in the natural sciences and
technology have often made use of the concept of serendipity. Could it
therefore be a useful concept for studies of discovery in mathematics? In
this paper, I have argued that serendipity occurred in the development of
mathematical logic. Are there other examples of serendipity in the history
of mathematics?
One obvious further example is the discovery of non-Euclidean
geometry. Saccheri, for example, was working on the attempt to prove
Euclids fifth postulate using reductio ad absurdum. He failed to find the
proof he was looking for, but discovered, without realising it, some
theorems of hyperbolic geometry. This seems undoubtedly a case of
serendipity. However, like the case of mathematical logic, it arises from a
foundational programme within mathematics. Are there examples of
serendipity, which occur in developments of mathematics which have no
connection with foundational issues?
Here, it is harder to find examples of serendipity, but there is a related
phenomenon which I suggest calling: additional serendipity. In serendipity
proper, someone fails to discover what he or she was looking for, but
discovers something else unexpected instead. In additional serendipity, the
researcher does discover what he or she was looking for, but, in addition,
discovers something else unexpected. Additional serendipity does seem to
be quite a common phenomenon in mathematics. For example, the basic
results of group theory were discovered in a successful investigation of the
solubility of polynomial equations, but group theory turned out
unexpectedly to provide, in addition, a useful tool for classifying
38 Chapter Two

geometries. Thus, it seems to me, that serendipity, and additional

serendipity, might indeed be useful concepts to use in an extension of
Celluccis investigations of discovery in mathematics.

References
Bourbaki, N. (1949) Foundations of mathematics for the working
mathematician, Journal of Symbolic Logic, 14, pp. 1-8.
Cellucci, C. (1998) Le Ragioni della Logica. Laterza.
. (2002) Filosofia e matematica. Laterza.
. (2007) La filosofia della matematica del Novecento. Laterza.
Church, A. (1940) A Formulation of the Simple Theory of Types, Journal
of Symbolic Logic, 5, pp. 56-68.
Corfield, D. (2003) Towards a Philosophy of Real Mathematics.
Cambridge University Press.
Davis, M. (1988a) Mathematical Logic and the Origins of Modern
computing. In Herken, 1988, pp. 149-174.
. (1988b) Influences of Mathematical Logic on Computer Science. In
Herken, 1988, pp. 315-326.
Detlefsen, M. (1990) On an alleged refutation of Hilberts program using
Gdels first incompleteness theorem, Journal of Philosophical Logic,
19. pp. 343-377.
Doxiadis, A. and Papadimitriou, C.H. (2009) Logicomix. An Epic Search
for Truth. Bloomsbury.
Frege, G. (1879) Begriffsschrift. Eine der arithmetischen nachgebildete
Formelsprache des reinen Denkens. English Translation in van
Heijenoort (1967), pp.1-82.
Frege, G. (1884) The Foundations of Arithmetic: A Logico-Mathematical
Enquiry into the Concept of Number. English Translation by
J.L.Austin, Blackwell, 1968.
Gillies, D.A. (1982) Frege, Dedekind, and Peano on the Foundations of
Arithmetic. 2nd Corrected Edition, Routledge, 2011.
. (1996) Artificial Intelligence and Scientific Method. Oxford University
Press.
. (2002) Logicism and the Development of Computer Science. In
A.C.Kakas and F.Sadri (eds.) Computational Logic: Logic
Programming and Beyond, Part II, Springer, pp. 588-604.
Gillies, D.A, and Zheng, Y. (2001) Dynamic Interactions with the
Philosophy of Mathematics, Theoria, 16(3), pp.437-459.
 Serendipity and Mathematical Logic 39

Gdel, K. (1931) On Formally Undecidable Propositions of Principia

Mathematica and related systems I. English Translation in Van
Heijenoort, 1967, pp. 566-616.
Herken, R. (ed.) (1988) The Universal Turing Machine. A Half-Century
Survey. Oxford University Press.
Hale, R. and Wright, C. (2001) The Reasons Proper Study. Essays
towards a neo-Fregean Philosophy of Mathematics. Oxford University
Press.
Mackenzie, D. (2001) Mechanizing Proof. MIT Press.
Merton, R.K., and Barber, E. (2004) The Travels and Adventures of
Serendipity. A Study in Sociological Semantics and the Sociology of
Science. Princeton University Press.
Muggleton, S. (ed.) (1992) Inductive Logic Programming. Academic
Press.
Numerico, T. (2005) Alan Turing e lIntelligenza delle Macchine. Franco
Angeli.
Priestley, M. (2011) A Science of Operations. Machines, Logic and the
Invention of Programming. Springer.
Robinson, J.A. (1965) A Machine-Oriented Logic Based on the Resolution
Principle, Journal for the Association for Computing Machinery, 12,
pp. 23-41.
Rosser, J.B. (1936) Extensions of some theorems of Gdel and Church,
Journal of Symbolic Logic, 1, pp. 87-91.
Russell, B.A.W. (1908) Mathematical logic as based on the theory of
types, American Journal of Mathematics, 30, pp. 222-262. Reprinted in
Van Heijenoort, 1967, pp.150-182.
Simpson, S.G. (1988) Partial realizations of Hilberts program, Journal of
Symbolic Logic, 53, pp. 349-363.
Tarski, A. (1936) Der Wahrheitsbegriff in den formalisierten Sprachen,
Studia Philosophica, 1, pp. 261-405.
Van Heijenoort, J. (Ed.) (1967) From Frege to Gdel. A Source Book in
Mathematical Logic, 1879-1931. Harvard University Press.
Wos, L. and Pieper, G.W. (1999) A Fascinating Country in the World of
Computing. Your Guide to Automated Reasoning. World Scientific.
 CHAPTER THREE

WHAT MATHEMATICAL LOGIC SAYS ABOUT

THE FOUNDATIONS OF MATHEMATICS

CLAUDIO BERNARDI

To Carlo

SUMMARY My purpose is to examine some concepts of mathematical

logic studied by Carlo Cellucci. Today the aim of classical mathematical
logic is not to guarantee the certainty of mathematics, but I will argue that
logic can help us to explain mathematical activity; the point is to discuss
what and in which sense logic can "explain". For example, let us consider
the basic concept of an axiomatic system: an axiomatic system can be very
useful to organize, present, and clarify mathematical knowledge. And,
more importantly, logic is a science with its own results: so, axiomatic
systems are also interesting because we know several revealing theorems
about them. Similarly, I will discuss other topics such as mathematical
definitions, and some relationships between mathematical logic and
computer science. I will also consider these subjects from an educational
point of view: can logical concepts be useful in teaching and learning
elementary mathematics?

KEYWORDS Mathematical logic, foundations of mathematics, axiomatic

systems, proofs, definitions, mathematical education.

1. Mathematical logic vs. the foundations of mathematics

There is no doubt that research in mathematical logic can contribute to the
study of the foundations of mathematics. For instance, mathematical logic
provides answers (both complete and partial) to the following questions:
- Given a precisely stated conjecture, can we be sure that eventually
a good enough mathematician will be able to prove or disprove it?
- Can all mathematics be formalized?
42 Chapter Three

- Is there a "right" set of axioms for arithmetic, or for mathematical

analysis? is there a proof (in some fixed theory) for any statement of
arithmetic which is true in N?
- Can we prove the consistency of standard mathematical theories?
and what does it mean to prove consistency?
- By adding a new axiom to a theory, we find new theorems; but can
we also expect to find shorter proofs for old theorems?
- Will we ever construct a computer that will be capable of
answering all mathematical problems?
- Is any function from N to N computable by an appropriate
computer? if not, how can we describe computable functions?
- If we know that a computation ends, can we estimate the time
necessary to complete the computation?
- Is it true that, if a "short" statement is a theorem, then there is a
short proof for it?
The list could be much longer. In some cases (as in the first question)
the answer given by logic contradicts the expectations of a working
mathematician, while in other cases (as in the last question) the answer
confirms that expectation. However, it is not true that the general purpose
of mathematical logic is to clarify the foundations of mathematics. First of
all, for the past few decades, much research in logic has been of mainly
technical value and does not deal directly with the foundation of
mathematics. Perhaps in the nineteenth century, logic was regarded as a
way to guarantee the certainty of mathematics. But nowadays we do not
expect that much: it seems nave, and perhaps even futile, to hope for a
definitive, proven certainty of mathematics.
Let us start from the beginning. Mathematical logic provides us with a
precise definition of a proof and suggests rigorous methods and procedures
for developing mathematical theories. But these are just the initial steps of
mathematical logic: if logic consisted only in giving detailed definitions of
proofs and theories, it would not be of great scientific importance. While
succeeding in formalizing statements and arguments is interesting, the
historical and cultural importance of proof theory, model theory, and
recursion theory strongly depends on the results achieved in these areas
(for example, on the answers given to the previous questions).
In other words, mathematical logic is a way of organizing mathematics
and solving paradoxes; but I find that logic is interesting also because its
organization of mathematics provides significant results. In fact, any
theory grows if and when results are found. So, we can distinguish
between two kinds of logical results which can be useful in the study of
foundations and, more generally, to working mathematicians.
 What Mathematical Logic Says about the Foundations of Mathematics 43

On the one hand, mathematical logic provides explicit rules that

mathematicians habitually use (often without being fully aware of it),
inserting them into a clear and consistent framework; in this way
omore complex situations can be tackled. For instance in logic:

- it is explained what a proof by contradiction, or a counterexample,

is; it is not impossible for a mathematician, who in his work usually gives
proofs by contradiction and counterexamples, to be unable to give clear
answers to explain these totally elementary concepts1;
- various forms of the principle of induction are stated explicitly and
compared;
- equivalents to the axiom of choice, or weaker forms of it, are stated
and recognized.

On the other hand, mathematical logic allows us to construct a theoretical

framework that clarifies the meaning and limitations of mathematical
activity. The study of logic can provide information of the following kind:
this theory is decidable, while this other one is not. Note that often logical
results contrast with the nave expectations of working mathematicians.

2. Mathematical experience vs. mathematical logic

In the introduction to the book The Mathematical Experience (Davis and
Hersh 1981), Gian-Carlo Rota, challenging the idea that mathematics
consists mainly of the demonstration of theorems, wrote a famous
sentence:
A mathematician's work is mostly a tangle of guesswork, analogy, wishful
thinking and frustration, and proof, far from being the core of discovery, is
more often than not a way of making sure our minds are not playing tricks.

1
Let us briefly recall that the word counterexample denotes an example that shows
that a statement is not correct, when the statement consists of an implication that is
preceded by a universal quantifier. To this end, we have to construct an object x
that satisfies the hypothesis but not the thesis. In formal terms, the explanation is
clearer: to conclude that the formula x [P(x)oQ(x)] does not hold, we have to
prove x [P(x) o Q(x)], that is, x [P(x) o Q(x)] and this formula is logically
equivalent to x [P(x) Q(x)].
44 Chapter Three

Of course Gian-Carlo Rota was right2. It is true that in mathematical

experience, when checking a method, testing a tool, or hoping that an
application will follow, there are very often trials and failures. But in a
logic book we will not find a chapter about hope or failure: it is not the
purpose of mathematical logic to describe how a mathematician works.
Regarding mathematical activity and its formalization, it is worth
quoting three amusing conversations. In Davis and Hersh (1981) a
philosophy grad student asks the Ideal Mathematician, What is a
mathematical proof, really?. A similar question is considered in Devlin
(1992); while in Hersh (2011) a Successful Mathematician is accosted by
the Stubborn Student, who has trouble when comparing the mathematical
concept of a limit with its concrete applications.
But there is a different question: what is the meaning of logical notions
such as mathematical proof and mathematical theories? Of course, these
notions have to do with the work of a mathematician.
First of all, a distinction must be made between the way in which a
mathematician works and the final presentation of a subject. Logic refers
to the organization of a mathematical subject in a clear form, so that other
people can understand it: the starting point (axioms), new concepts
(definitions), properties and consequences (theorems).
In any case, mathematical logic supplies juts a suitable frame for
mathematical theories, but we could also find other models. Is the logical
model a convenient one? When discussing this point, we have to face two
different questions:

- is the model faithful?

- is the model useful?

Let me give a rough example: in our framework, taking a photograph

could provide a faithful description, but would be totally useless. A
faithful description which does not yield results is much less interesting
than an unfaithful description which yields results. There is no doubt that
the logical formalization of the concept of a proof is far from a concrete
proof and even further from the way in which a proof is found, but many
results can be found in proof theory, which concern mathematicians, their

2
However, I do not agree with the idea that a proof is just a method of confirming
what we already know. Note that many mathematicians (including Gian-Carlo
Rota) have sought elegant proofs; moreover, very often a proof allows for a deeper
understanding of the subject.
 What Mathematical Logic Says about the Foundations of Mathematics 45

expectations, and the limits of mathematics. From this point of view, logic
is a fruitful model.We could make similar remarks about other areas of
mathematics. Consider mathematical analysis: it rests on the set of real
numbers, which can hardly be regarded as a faithful model of reality. But
mathematical analysis has enormous importance, because its theorems can
be fruitfully applied in physics and the natural sciences, in the study of the
real world.
Mathematical logic, exactly like mathematical analysis, is justified by
its results.

3. Axiomatic systems
The axiomatic method is a way of thinking. This is true in mathematical
research, as well as in mathematical education (I will deal with mathematical
education in 5 and 6). The concept of a mathematical axiom, and its
meaning, is part of our culture; several sciences other than mathematics
have tried to introduce something similar to axioms, in order to achieve a
more convincing approach to the matter studied.
In mathematics, the axiomatic method has had enormous influence
from the time of Euclid (consider the fifth postulate and non-Euclidean
geometries). Or, to give a present-day example, consider reverse
mathematics which would not have been born without the concept of an
axiomatic system; see (Marcone 2009) for a general introduction.
Instead of discussing the relationship between axioms and theorems in
general terms, I prefer to stress one particular aspect.
Using the terminology of recursion theory (see Rogers 2002 for
definitions), the set of axioms of a typical mathematical theory is recursive
(or decidable), in the sense that one can recognize if a given sentence is an
axiom. On the other hand, the set of theorems is not recursive, it is only
recursively enumerable, because, given a sentence, we are generally not
able to decide whether it is a theorem or not, we can only try and hope to
get an answer.
Mathematics allows us to jump from a recursive set up to a set which is
recursively enumerable but not recursive. So, even if the content of any
given theorem is implicit in the axioms, the set of theorems is more
complex, in a technical sense, than the set of axioms.
In my opinion, and from an abstract point of view, this is the ultimate
task of mathematics and, on the other hand, it explains the difficulty of
doing mathematics. Some remarks are necessary.
First of all, very often in the work of a mathematician, the set of
axioms is not explicitly stated; but, even if this is the case, I think it can be
46 Chapter Three

assumed to be recursive, in the sense that the mathematician is able to

recognize the axioms he is entitled to use. Moreover, a mathematician
sometimes changes his hypothesis when trying to give a proof: he hoped
that a result D held in general in a given theory T, but realized that it is
necessary to add a hypothesis E. The axiomatic approach is not affected in
this case, because it has been proven that EoD is a theorem of T.
It should be observed that the mathematical community searches
continuously for new axioms, which are deeper or more general, or more
suitable for some purpose, trying to give a more comprehensive
explanation of a subject.
Taking account of these situations, Carlo Cellucci (2002) introduced
and studied open systems, which can better describe the analytical
method adopted by mathematicians. A fairly similar approach was
suggested in strictly logical terms by Roberto Magari (1974) and (1975).
These ideas are profound and relevant, but I think that open systems are
not to be juxtaposed on axiomatic systems, because both reflect
mathematical activities. Note also that a new axiom, or a new hypothesis,
can be considered in different situations: as a partial step in the solution of
a problem, or because it allows us to find desired consequences, or when
introducing a new axiomatic system.
Let us go back to the set of theorems. A working mathematician may
be astonished by the fact that this set is recursively enumerable, since, at
first glance, it seems to be even more complex. The point is that a
mathematician does not proceed by listing the set of theorems and looking
for something interesting: his procedure is completely different. However,
the fact that the set of theorems is recursively enumerable corresponds to
the possibility of checking any proof.

4. Some remarks about definitions

a) A definition is just an abbreviation, since a long expression is
substituted by a single new term: it is introduced simply for the reader's
convenience. We could always replace the defined term with the defining
expression: we get a statement that is less clear than the original one, but
that has the same content.
b) Definitions are an indispensable part of any book on mathematics.
Definitions not only draw attention to what will be useful later, but
correspond to concepts, and therefore must be fully understood by
anybody who studies a subject. For instance, the definitions of limit,
continuity, derivative, are essential for studying calculus.
 What Mathematical Logic Says about the Foundations of Mathematics 47

In my opinion, claims a) and b) are both correct. Claim a) reflects a

theoretical and abstract point of view, whereas b) applies to every human
being who learns mathematics.
There is no contradiction between a) and b), just as there is no
contradiction between someone who says that a sphere is a locus of points
which have the same distance from a given point, and someone who thinks
of a sphere as something round, that can roll perfectly in all directions.
A typical question about definitions is the following: is a theorem
proved starting only from axioms or starting from both axioms and
definitions?
First note that, in mathematics, a definition gives no information about
the objects involved: in elementary geometry we can define the bisector of
an angle before knowing that any angle has a bisector. From this point of
view, a definition cannot be a starting point for proving something (even
though, of course, it can allow us to give other definitions).
On the other hand, in many cases it seems that, during a proof, we rely
on definitions, especially when basic concepts are involved. But let us
compare the situation to an algebraic one. When making a complicated
algebraic computation, it often happens that is convenient to introduce a
new letter, for instance setting y = x2 (where the letter y did not occur
previously). In doing this, we may be putting ourselves in a position to
directly apply a known formula or to recognize an algebraic pattern; but
note that the previous equality gives no information about x. The equality
y = x2 is nothing but a definition, introduced only to make what follows
accessible and clear.
Returning to theorems and definitions, we conclude that any proof is
supported by axioms (and already known theorems), and not by any
definitions. When proving a theorem we often read statements such as:
recalling the definition of a limit, we can say that ...; or, the number d is
a divisor of p, but p is prime and, by the definition of a prime number, we
conclude d = 1 or d = p. However, in these cases, the reference to a
definition is useful only for recalling the meaning of a term, that is, for
substituting a term with the proprieties used in defining that term.

5. Is mathematical logic useful at school?

which concepts should be taught?
Knowing mathematical logic does not provide a method for those who
want to do research in geometry or analysis; in the same way, the study of
logic does not provide a necessary introduction to high school
mathematics. In fact, were some logical concepts to be summarized in the
48 Chapter Three

first chapter of a mathematical textbook, this chapter would very likely be

nearly useless to students, and would soon be forgotten.
What can be useful in high school are frequent discussions of the
logical aspects of the mathematical concepts and procedures the students
are dealing with. Indeed, mathematical education has an educational value
that does not depend on applications. I am referring to skills involving the
use of language and argumentation; and language and argumentation are
obviously related to logic.
Let us examine some points.

a. Axiomatic systems
An axiomatic system is a way both of correctly organizing and presenting
knowledge in a specific mathematical area, and also in general of teaching
a correct way of proving and deducing.
In fact, without the concept of an axiomatic system, the teaching of
mathematics consists only of an unjustified set of rules and arguments,
based on common sense, or on the authority of the teacher. Such teaching
is appropriate to middle school, but is not always suitable for high school.
It has been remarked that, in the Elements of Euclid, the connections
between axioms and theorems are complex: if we try to specify, for any
theorem, the previous theorems and axioms upon which it depends, we
find an intricate structure. Even a good student cannot completely master
this logical structure.
This may be true, but I believe it is not a good reason to give up! On
the contrary, the teacher will pay attention to the ties between axioms and
theorems in some specific cases, stress the fact that a theorem depends or
does not depend e.g. on the Pythagorean theorem, and so on, even if he
knows in advance that no student will learn the entire structure of the
axiomatic system of Euclidean geometry.
Equally, a comparison between Euclidean geometry and non-Euclidean
geometries can be useful. The teacher will show that, in the new
geometries, some known theorems remain valid, while others (such as the
Pythagorean theorem, the sum of angles of a triangle, etc.) no longer hold;
on the other hand, there are also new theorems (such as the fourth criterion
for triangles: if the angles of a triangle are congruent to the angles of a
second triangle, the two triangles are congruent to each other). We must
not underestimate the educational importance of the fact that a
mathematical result holds in one theory but does not hold in another.
 What Mathematical Logic Says about the Foundations of Mathematics 49

b. Proving
The difference between verifying and proving is obviously fundamental
when teaching and learning mathematics (incidentally, it has been noted
by several people that the use of computers in geometry and arithmetic
does not help in this regard: why do we need to prove what is said or
shown by a computer?).
Of course, I am not referring to formal proofs. But even a student at the
beginning of stet high school education can understand some basic logical
facts about proofs, such as:

- any proof consists of several elementary steps;

- it is not easy to find a proof, but a given proof can be checked by
everybody who knows the concepts involved;
- in many cases there are different proofs for the same statement;
- some theorems are proved by contradiction;
- if a theorem is an implication, the inverse implication is not always a
theorem;
- proving DoE is logically equivalent to proving EoD;
- not for every x ... is different from for every x not ....

The importance of proving in teaching and learning mathematics has been

widely investigated; see for instance (Arzarello 2012), (Bernardi 1998),
(Bernardi 2010), (Francini 2010).

c. Formalizing statements
For students at the end of high school or the beginning of university,
formalizing statements is a useful exercise, and in particular finding the
explicit quantifiers and implications hidden in natural language. Think of a
trivial sentence like the square of an odd number is odd; in formalizing
it, we have to write a universal quantifier as well as an implication: for any
number n, if n is odd then n2 is odd.
On the other hand, I think that translating mathematical statements into
a first-order language is in general too difficult for students. Formalization
can only be done in specific cases. I refer for instance to the definition of
the limit of a function; in my opinion the difficulty in understanding this
notion also depends on the logical structure H G x (this is one of the
first times a student meets three alternating quantifiers).
Many current words and expressions used in mathematics are perhaps
useful in practice (and in particular in the practice of teaching), but are
subject to criticism for being unclear or ambiguous: e.g., "given a
number", "take a function", "we impose that ...", "the general term of a
50 Chapter Three

sequence", "fix the value x0 of a variable x", ... Formal language allows us
to clarify these situations. In this respect, logic can contribute (and in fact
has contributed) to improving rigour in natural mathematical language. For
other remarks, see (Bernardi 2011).

6. Formal notation and self-confidence at school,

in algebra, geometry, logic
Algebraic manipulation increases self-confidence in high school students.
Of course, not all students acquire good skills in algebra; but the
doubtful student seeks comfort in algebraic calculus and, in any case, tries
to perform some algebraic computations. Algebraic language is effective,
rules in algebra are clear and simple. Steps in computation do not require
too much thought; for this very reason, students usually prefer algebra to
geometry.
Students, as well as teachers, rely on algebraic formalism: this is
because algebraic formalism is artificial, and therefore governed by simple
rules. Sometimes the abstract is more natural than the concrete: algebraic
language is abstract, in the sense that it has been built by us and for us.
From an educational point of view, there is the obvious risk of a
mechanical and unconscious application of rules. On the other hand,
regular practice with algebraic computation develops other skills, such as
the capacity to make indirect controls: e.g., in some situations we will
automatically be expecting a polynomial to be homogeneous (and
therefore realize something is wrong if it turns out not to be).
The situation is completely different in the teaching and learning of
geometry. In my opinion, the teaching difficulties of Euclidean geometry
also result from the lack of convenient notation. Take for example the
angles RSP and SPQ of the quadrilateral PQRS: while these symbols
are not long, they have to be continuously interpreted within a diagram.
There is no formal manipulation, with the exception of some very
particular cases (for instance, referring to the sum of vectors the equality
AB + BC = AC holds, but a similar equality does not hold for segments).
Unlike algebraic expressions, geometrical symbols are not suitable for
computation, they have only schematic and mnemonic value.
So, in the teaching of mathematics, even though the logical structure
(axioms, theorems, primitive concepts, ...) is more evident in geometry
than in algebra, a formal calculus occurs only in algebra.
When introduced to the symbolic use of connectives and quantifiers at
the end of high school or at the beginning of university, students are
 What Mathematical Logic Says about the Foundations of Mathematics 51

amazed by the expressive power of logical language: any mathematical

statement seems to be expressible in symbolic logic.
However, at that level a calculus in logic can be presented very rarely,
and only in some specific cases. Symbols like connectives and quantifiers
allow us to express sentences in a clear and concise way, but they must be
interpreted each time, and cannot be directly manipulated. This situation
presents analogies with notation in geometry, rather than with algebraic
symbols.
In high school we can have a "logical calculus" only in the construction
of truth tables, but I think that this construction has limited value in
mathematical education. From this point of view, logical language cannot
provide a student with the same self-confidence as algebraic manipulation.

7. Mathematical logic and computer science

As is well known, computers can contribute to providing mathematical
proofs. Does mathematical logic play a role in this contribution?
To answer this question, I think that a distinction has to be made: there
are (at least) three different kinds of contributions made by computer
science to mathematics.
- First of all, there are computer-assisted proofs. This name usually
refers to a computer used to perform very long and complex computations,
or to examine a great number of possibilities. The first famous case
occurred in 1976, when the four colour theorem was proven. That proof
has been discussed for a long time (perhaps too long); I believe that the
point of discussion was not the certainty of the proof, but the change in the
style of proving (like an athletic world record which is achieved by
modifying technical equipment). A computer can also be used to find
approximate solutions of equations, to simulate the evolution of a
phenomenon, to visualize patterns. But these applications are not too
different from the previous one. In all these cases, the role of logic is
usually limited.
- The role of logic is greater when a computer is used directly to find
new theorems, or new proofs of known theorems, as happens particularly
in algebra and geometry. This area is related to artificial intelligence: a
computer searches for new facts by combining known facts. The
possibility that a computer may find and prove a theorem is fascinating.
In these cases, to plan a theorem prover, formalization of arguments
and proofs is obviously necessary. A problem is to direct the research
towards statements of some interest. It is very easy to give rules to obtain
new theorems: for instance, starting from A we can deduce A A as well
52 Chapter Three

as B o A, but how can we recognize interesting theorems among trivial

ones?
- Lastly, we must mention a more recent application, where the role
of formal logic is even greater.
Often, in the history of mathematics, a mistake has been found in a
proof that had previously been accepted. In this third application of
computer science, known as automated proof checking (or also, with a
slightly different meaning, automated theorem proving), computers are
used simply to formalize known proofs of theorems and check them in
detail. The purpose is to certify theorems and collect them in libraries. We
could mention, for instance, the Mizar system (see http://www.mizar.org/)
and the Coq Proof Assistant (see http://coq.inria.fr).
It is currently hard to predict just how much and how widely these
supports will be used, but the interesting point is that automated proof
checking not only guarantees greater accuracy (even if, in any case, we
cannot hope for the complete certainty of a statement): in fact, looking for
a better way to formalize also sheds new light, suggests new ideas, and
brings in new generalizations.

References
Arzarello F. (2012). Provare se, vedere che, sapere perch: la
dimostrazione in classe. In: Proceedings of the XIX Congress of the
Unione Matematica Italiana, to appear
Bernardi C. (1998). How formal should a proof be in teaching
mathematics?. Bulletin of the Belgian Mathematical Society, suppl. 5,
n. 5: 7-18
. (2010). Linguaggio algebrico e linguaggio logico nell'insegnamento e
nell'apprendimento della matematica. In G. Gerla, ed., Logica
matematica e processi cognitivi, Collana Scientifica dell'Universit di
Salerno, Rubbettino Editore: 39-45
. (2011). I linguaggi della matematica a scuola. Riflessioni di un logico,
L'insegnamento della Matematica e delle Scienze integrate, 34 A-B:
559-576
Cellucci C. (2002). Filosofia e matematica, Rome-Bari: Laterza
. (2010). Matematica e filosofia della matematica: presente e futuro, La
Matematica nella Societ e nella Cultura - Rivista della Unione
Matematica Italiana (1) 3: 201-234
Devlin K. (1992). Computers and Mathematics, Notices of the AMS, 39:
1065-1066
 What Mathematical Logic Says about the Foundations of Mathematics 53

Davis P.J. and Hersh R. (1981). The Mathematical Experience, Boston:

Birkhuser
Francini P. (2010). Cercare, mostrare, dimostrare, Archimede, LXII: 67-73
Hersh R. (2011). Mathematical Intuition (Poincare, Polya, Dewey). In
Cellucci C., Grosholz E., Ippoliti E., eds., Logic and Knowledge,
Cambridge Scholars Publishing, UK: 297-323 (partially translated into
Italian in Archimede, LXII (2010): 202-205)
Magari R. (1974). Su certe teorie non enumerabili, Annali di matematica
pura ed applicata (IV), XCVIII: 119-152
. (1975). Significato e verit nell'aritmetica peaniana, Annali di
matematica pura ed applicata (IV), CIII: 343-368
Marcone A. (2009). Equivalenze tra teoremi: il programma di ricerca della
reverse mathematics, La Matematica nella Societ e nella Cultura -
Rivista della Unione Matematica Italiana, Serie III: 101-126
Rogers H. (2002). Theory of recursive functions and effective
computability, Cambridge: MIT Press
 CHAPTER FOUR

DEDEKIND, HILBERT, GDEL:

THE COMPARISON BETWEEN LOGICAL
SENTENCES AND ARITHMETICAL SENTENCES

V. MICHELE ABRUSCI

To Carlo Cellucci, with friendship and gratitude

SUMMARY: Some of the main theorems of modern logic are considered in

this paper as theorems concerning the equivalence between logical
sentences and arithmetical sentences, in particular with respect to well-
investigated hierarchies of logical sentences and arithmetical sentences.
These theorems are: Dedekindss Theorem (the consequence of Dedekinds
definition of the set of natural numbers, each arithmetical sentence is
equivalent to a logical sentence), Gdels Completeness Theorem for first
order logic, the possibility of an arithmetization of syntax, Gdels
Incompleteness Theorem. An important role in this context is also played
by Hilberts Thesis (the equivalence between satisfiability and consistency),
which may be proved when it is restricted to first-order formulas.

KEYWORDS: Completeness, Incompleteness, Satisfiability, Consistency,

Logical Sentences, Arithmetical Sentences.

1. Logical sentences and hierarchies of logical sentences

Logical sentences are sentences which are closed (i.e. sentences without
free variables) and contain only logical concepts.
Note that every first-order formula contains only logical concepts, but
in general even if it is called a closed formula according to the usual
syntax of first-order logic it may be not a logical sentence since it may
contain many free variables:
56 Chapter Four

x implicitly, a free class variable X, the generic non-empty class of

objects on which the formula may be interpreted;
x a finite number of free variables for functions on X, usually
called (using rather misleading terminology) individual constants
(the free variables for 0-ary functions on X) and function symbols
(the free variables for m+1-ary functions on X with mz1);
x a finite number of free variables for predicates on X, usually
called (again using rather misleading terminology) propositional
letters (the free variables for 0.ary predicates on X) and predicate
symbols (the free variables for m+1-ary predicates on X with mz
1).

Note also that propositional formulas containing propositional letters, i.e.

proposisitional variables, are not logical sentences since they contain free
variables.
It is useful to classify a large number of logical sentences in the
following hierarchies
(6n )nN and (3n) nN ,
indexed by natural numbers. Both hierarchies have been thoroughly
investigated in books on proof-theory (e.g. in (Girard 1987) and (Girard,
2006-07)).

x 60 = 30 is the class of all the logical sentences which are

equivalent to a logical propositional sentence, i.e. to a logical
formula (of propositional logic) constructed from logical
constants 0 (falsehood) and 1 (truth) by means of logical
connectives (and) and (or).
x 61 is the class of all the logical sentences which are equivalent to
a logical sentence obtained from a first-order logical formula by
putting before the formula the second-order existential
quantification  of each logical variable occurring inside the
formula. So, a logical sentence A belongs to 61 when A is
equivalent to a logical sentence
X f1fnP1Pq B
where B is a closed first-order formula and the variables X,
f1fn,P1Pq are all the logical variables occurring in B.
Therefore, each logical sentence belonging to 61 is equivalent to
the satisfiability of a closed first-order formula, i.e. each logical
sentence belonging to 61 is equivalent to the existence of a model
for a first-order formula.
 Dedekind, Hilbert, Gdel 57

x 31 is the class of all the logical sentences which are equivalent to

a logical sentence obtained from a first-order logical formula by
putting before the formula the second-order universal
quantification  of each logical variable occurring inside the
formula. So, a logical sentence A belongs to 31 when A is
equivalent to a logical sentence
X f1fnP1Pq B
where B is a closed first-order formula and the variables X,
f1fn,P1Pq are all the logical variables occurring in B.
Therefore, each logical sentence belonging to 31 is equivalent to
the logical validity of a closed first-order formula, i.e. each
logical sentence belonging to 31 is equivalent to the validity of a
first-order formula in all possible worlds.
x 6n+1 is the class of all the logical sentences which are equivalent
to a logical sentence obtained from a closed first-order logical
formula by putting before the formula a second-order
quantification ( or ) of each logical variable occurring inside
the formula, beginning with a group of existential quantifiers,
followed by a group of universal quantifiers, and so on for a total
of n+1 groups of quantifiers.
x 3n+1 is the class of all the logical sentences which are equivalent
to a logical sentence obtained from a closed first-order logical
formula by putting before the formula a second-order
quantification ( or ) of each logical variable occurring inside
the formula, beginning with a group of universal quantifiers,
followed by a group of universal quantifiers, and so on for a total
of n+1 groups of quantifiers.

It is easy to see that a logical sentence belongs to 6n if and only if its

negation belongs to 3n (and so a logical sentence belongs to 3n if and only
if its negation belongs to 6n).
In these hierarchies we may find all the logical sentences which are
equivalent to a logical sentence obtained by quantifying in an arbitrary
way all the free logical variables occurring in a closed first-order logical
formula. So, in these hierarchies we find a large number of logical
sentences, and very interesting classes of logical sentences.

Note that the statements of the logical investigations (in proof-theory and
model theory) of first-order logic are usually logical sentences belonging
to the union of classes 61 and 31.
58 Chapter Four

2. Arithmetical sentences and hierarchies of arithmetical

sentences
First-order arithmetical sentences are closed first-order sentences containing
only logical and arithmetical concepts. In particular, first-order arithmetical
sentences are arithmetical sentences of first-order Peano Arithmetic (PA),
i.e. the closed formulas of the formal language of Peano Arithmetic (PA)
when they are interpreted in a standard way on the set of natural numbers
and so function symbols and predicate symbols are interpreted in a
standard way as arithmetical functions and arithmetical predicates and
each quantification becomes a quantification on a variable for natural
numbers.

In 20th-century proof-theory, the following classification of first-order

arithmetical sentences has been proposed by means of the hierarchies
(6n0)nN and (3n0)nN
indexed by natural numbers. Both these hierarchies have been thoroughly
investigated in books on proof-theory (e.g. in (Girard 1987) and (Girard,
2006-07)).

x 600 = 300 is the class of all the first-order arithmetical sentences

which are equivalent to a quantifier-free arithmetical sentence of
first-order PA.
x 610 is the class of all the first-order arithmetical sentences which
are equivalent to an arithmetical sentence of first-order PA
obtained from a quantifier-free formula of PA by putting before
the formula the existential quantification  of each number
variable occurring inside the formula. So, a first-order
arithmetical sentence A belongs to 610 when A is equivalent to an
arithmetical sentence of first-order PA
n1nk B
where B is a quantifier-free formula of the formal language of PA.
x 310 is the class of all the first-order arithmetical sentences which
are equivalent to an arithmetical sentence of first-order PA
obtained from a quantifier-free formula of PA by putting before
the formula the universal quantification  of each number
variable occurring inside the formula. So, a first-order
arithmetical sentence A belongs to 310 when A is equivalent to an
arithmetical sentence of first-order PA
n1nk B
where B is a quantifier-free formula of the formal language of PA.
 Dedekind, Hilbert, Gdel 59

x 6n+10 is the class of all the first-order arithmetical sentences which

are equivalent to an arithmetical sentence of first-order PA
obtained from a quantifier-free formula of the formal language
PA by putting before the formula a quantification ( or ) of each
number variable occurring inside the formula, beginning with a
group of existential quantifiers, followed by a group of universal
quantifiers, and so on for a total of n+1 groups of quantifiers.
x 3n+10 is the class of all the first-order arithmetical sentences
which are equivalent to an arithmetical sentence of first-order PA
obtained from a quantifier-free formula of the formal language
PA by putting before the formula a quantification ( or ) of each
number variable occurring inside the formula, beginning with a
group of universal quantifiers, followed by a group of universal
quantifiers, and so on for a total of n+1 groups of quantifiers.

It is easy to see that a first-order arithmetical sentence belongs to 6n0 if and

only if its negation belongs to 3n0 (and so a logical sentence belongs to
3n0 if and only if its negation belongs to 6n0).
In these hierarchies we find all the first-order arithmetical sentences
which are equivalent to sentences obtained by quantifying all the free
number variables occurring in a quantifier-free formula of first-order PA.
Moreover, a logical sentence belongs to 6n0 if and only if its negation
belongs to 3n0 .
Note that a set X of natural numbers is called 6m0 ( or 3m0) when the
sentence n X belongs to 6m0 (to 3m0, resp.). Specifically, a recursive set
of natural numbers is a set belonging to 600 = 300, and an enumerable set
of natural numbers is a set belonging to 610. There are also comparable
hierarchies of second-order arithmetical sentences, a topic thoroughly
investigated in the textbooks on proof-theory. Further details on these
classifications may be found, e.g., in (Girard 1987) and (Girard 2006-07).

3. Dedekinds Theorem: arithmetical sentences

are equivalent to logical sentences
The very important paper (Dedekind 1888) may be considered the proof of
a theorem establishing the equivalence between arithmetical sentences and
logical sentences, in the following sense:
For every arithmetical sentence A there is a logical sentence B such that
AlB.
This theorem may be called Dedekinds Theorem.
60 Chapter Four

Dedekind (1888) gives a rather trivial definition of the number 0 as a

logical concept (for example, the logical concept of emptyset ) and the
definition of a successor operation on arbitrary sets (and in particular, on
natural numbers) as a logical operation (for example, by defining the
successor of x as s(x) = x {x}), and finally he uses the induction
principle as a way to define the set of natural numbers in a very original
and fruitful way:

the set of all the natural numbers is the smallest set

containing and closed under the successor operation s.

By means of these definitions, one obtains the following equivalence

between the sentence saying that something x is a natural number and a
logical sentence x belongs to every class which contains and is classed
under the successor operation s:
x N X IND(X,x)
where IND(X,x) is an abbreviation of this logical first-order formula:
X y(y X s(y) X) x X
Moreover, by means of the definitions given by Dedekind, every arith-
metical concept may be translated into a logical concept.
Now, we may prove on the basis of this logical definition of the set
of natural numbers, given in (Dedekind 1888) that for every arithmetical
sentence A there is a logical sentence B such that AlB.
Indeed, each arithmetical sentence A is equivalent to a universal sentence
n1nkC(n1,,nk)
i.e.
x1xk(x1 N xk N C(x1,,xk))
i.e.
x1xk(x1 N xk N C(x1,,xk))

or to an existential sentence
n1nmC(n1,,nm)
i.e.
x1xk(x1 N xk N C(x1,,xk))

In the first case, according to Dedekinds definitions, A is equivalent to the

logical sentence
x1xk(X IND(X,x1) X IND(X,xk) C*(x1,,xk))
i.e.
x1xk(X IND(X,x1) X IND(X,xk) C*(x1,,xk))
 Dedekind, Hilbert, Gdel 61

where C* is obtained from C by replacing every arithmetical concept with

the corresponding logical concept; and in the second case, according to
Dedekinds definitions, A is equivalent to the logical sentence
x1xm(X IND(X,x1) X IND(X,xm) C*(x1,,xm))
where C* is obtained from C by replacing every arithmetical concept with
the corresponding logical concept.
Of course, what is established in the paper (Dedekind 1888) may be
considered and has been considered a way of reducing arithmetic to
logic, or of giving arithmetic a logical foundation: on this view, the paper
(Dedekind 1888) shows how it is possible to remove arithmetical
intuition and to build arithmetic by using only logical tools. Against this
reductionistic reading of (Dedekind 1888) we may raise all the criticisms
which are usually presented against the reductionistic approach in the
philosophy of mathematics and the philosophy of science. Moreover, it is
not certain that Dedekinds real aim in his paper (Dedekind 1888) was the
reduction of arithmetic to logic, i.e. to remove arithmetical intuition and to
replace arithmetical intuition with logical intuition.
In any case, it is possible (and we believe that it is better) to consider
the work done in (Dedekind 1888) as a way to better understand what the
set of natural numbers is, and to open new ways for the development of
arithmetic without removing the usual ways based on arithmetical
intuition: the set of natural numbers is better understood when it is defined
in logical terms as in (Dedekind 1888), and we get from (Dedekind 1888)
that there is the possibility of proving or of refuting each arithmentical
sentence A by proving or refuting the logical sentence B which is
equivalent to A. I.e., it is possible (and we prefer) to consider (Dedekind
1888) as a paper which shows the equivalence between arithmetical
sentences and logical sentences, in the sense that each arithmetical
sentence is equivalent to a logical sentence.

4. Weak Dedekinds Theorem: a relation

between arithmetical and logical hierarchies
A weak form of Dedekinds Theorem concerns the relationship between
the aforementioned hierarchies of logical sentences and the afore-
mentioned hierarchies of first-order arithmetical sentences, and in
particular the relationships between the first levels of these hierarchies. We
prefer to call this restriction of Dedekinds Theorem the Weak Dedekinds
Theorem.
The Weak Dedekinds Theorem says that:
62 Chapter Four

a) Every first-order arithmetical sentence belonging to 610 is equivalent to

a logical sentence belonging to 31;
b) Every first-order arithmetical sentence belonging to 310 is equivalent to
a logical sentence belonging to 61.

Indeed, as shown in textbooks on proof theory (e.g. in (Girard, 1987) and

(Girard, 2006-07)), when C is a quantifier-free formula of PA, then
a) the logical sentence equivalent to the arithmetical sentence
n1nmC(n1,,nm)
x1xm(X IND(X,x1) X IND(X,xm) C*(x1,,xm))
is equivalent to a logical sentence which begins with a string of second-
order universal quantifiers followed by a closed first-order formula, i.e. to
a logical sentence belonging to 31;
b) the logical sentence equivalent to the arithmetical sentence
n1nkC(n1,,nk)
x1xk(X IND(X,x1) X IND(X,xk) C*(x1,,xk))
is equivalent to a logical sentence which begins with a string of second-
order existential quantifiers followed by a closed first-order formula, i.e.
to a logical sentence belonging to 61.
So, the Weak Dedekinds Theorem specifies where, for each
arithmetical sentence belonging to the first levels of the aforementioned
hierarchies of arithmetical first-order sentences, we find its equivalent
logical sentence: each existential first-order arithmetical sentence is
equivalent to a universal logical sentence, each universal first-order
arithmetical sentence is equivalent to a universal logical sentence. There is
an interesting exchange of quantifiers when we go from an arithmetical
sentence to its equivalent logical sentence!
Thus, the Weak Dedekinds Theorem may be expressed as follows, in
two parts:

a) 610 31 (first part of the Weak Dedekinds Theorem)

b) 310 61 (second part of the Weak Dedekinds Theorem)
where X Y means that every element of X has an equivalent in Y.

When the Weak Dedekinds Theorem is expressed in this way, it is natural

to pose the questions: does the inverse of the first part of the Weak
Dedekinds Theorem hold true? does the inverse of the second part of the
Weak Dedekinds Theorem hold true? We shall show that the positive
answer to these questions is provided by very important theorems of 20th-
century mathematical logic.
 Dedekind, Hilbert, Gdel 63

5. Gdels Completeness Theorem for first-order logic:

the inverse of the first part of the Weak Dedekinds
Theorem
Gdels Completeness Theorem for first-order logic (Gdel 1930) is
usually presented as a theorem which states (together with the rather trivial
Validity Theorem) the equivalence of the logical provability of first-order
formulas and the logical validity of first-order formulas, i.e. the
equivalence of the syntax and the semantics of first-order logic.
Another way to present Gdels Completeness Theorem for first-order
logic is as a theorem establishing the duality between the logical
provability of a first-order formula A and the existence of countermodels
for formula A (i.e. models for the formula A).
There is another way to present Gdels Completeness Theorem for
first-order logic, as a theorem which makes it possible - by means of the
arithmetization of logical syntax, first performed in (Gdel 1931) - to
establish the equivalence of logical sentences and first-order arithmetical
sentences, in the following sense: Every logical sentence belonging to 31
is equivalent to a first-order arithmetical sentence belonging to 610 , so that
31 610
i.e. as the inverse of the first part of the Weak Dedekinds Theorem.

Let A be a logical sentence belonging to 31: we will find an arithmetical

sentence which belongs to 610 and is equivalent to A. Since A is a logical
sentence belonging to 31, then A is equivalent to the validity of a closed
first-order formula B, i.e. A is equivalent to
X f1fnP1Pq B.
where X, f1fn,P1Pq are all the logical variables occurring in B.
Now, the Completeness Theorem for first-order logic says that the logical
validity of B implies the provability of B, and moreover (by the
correctness lemma) the logical validity of B is equivalent to the provability
of B; i.e. A is equivalent to the logical sentence
there is a logical proof of B.
The possibility of the arithmetization of the logical syntax entails that the
last sentence there is a logical proof of B, and therefore the logical
sentence A, is equivalent to the arithmetical first-order sentence
there is a number n which is the code of a logical proof of B
which can be formulated in the language of first-order PA as
n DER(n, B)
64 Chapter Four

where DER(n,m) (n is the code of a logical proof which ends with a

formula whose code is m) is a quantifier-free formula of first-order PA.
So, n DER(n, B) belongs to 610 and is equivalent to A.

6. Hilberts Thesis: the equivalence between satisfiability

and consistency
With the term Hilberts Thesis we refer to one of the principles which led
Hilbert to the formulation of his Foundational Program, and so to the
beginning of Proof-Theory.
Hilberts Thesis may be formulated as follows: when A is a logical
formula (or a set of logical formulas), obtained from a mathematical axiom
(or from a set of mathematical axioms) by replacing each extra-logical
concept with a corresponding logical variable, the following statements are
equivalent:
x A is consistent, i.e. there is no logical proof of a contradictory
sentence from the hypothesis A (in Hilberts terms, A is not
contradictory)
x A is satisfiable, i.e. there is a model of A (in Hilberts terms,
there exists the mathematical concept defined by A).

It is easy to find this thesis (explicitly or implicitly) in Hilberts papers

devoted to the foundations of mathematics, e.g. in the seminal papers
presented in (Hilbert, 1935) and the book (Hilbert and Bernays, 1934-39).
Note that both the statements considered equivalent in Hilberts Thesis are
logical sentences, since they do not contain extra-logical concepts, but
these logical sentences are deeply different, so that the equivalence stated
in Hilberts thesis is not obvious. Indeed:
x The sentence A is consistent refers to internal objects of logic
(i.e. to the logical proofs from the hypothesis A) and says that
inside the set of the logical proofs from A there is no proof of a
contradiction.
x The sentence A is satisfiable refers to external objects (i.e. to
the models of A) and says that A becomes a true sentence when
we give suitable values to the logical variables occurring in A,
and these values are not always inside the universe of the logical
objects.
x The sentence A is consistent is a first-order sentence (where by
F we mean a contradictory sentence)
x(x is a logical proof of F from A)
 Dedekind, Hilbert, Gdel 65

i.e. x (x is not a logical proof of F from A)

x The sentence A is satisfiable is not a first-order sentence; for
example, if A is a first-order formula and the function variables in
A are f1,,fn and the predicate variables in A are P1,,Pq, then A
is satisfiable is the second-order sentence belonging to 61
X f1fnP1Pq A

Hilberts thesis underlies Hilberts program to prove the existence of

models of an axiom (or the existence of models of an axiom system) i.e.
the sentence A is satisfiable - not through the presentation of a model,
not through reduction ad absurdum, but through a proof of the consistency
of the axiom (the consistency of the axiom system), i.e. a proof of the
sentence A is consistent. To prove A is satisfiable through the
presentation of an extra-logical model is to prove a logical sentence by
means of non-logical tools, i.e. to obtain a proof which does not satisfy the
principle of the purity of methods; whereas to prove A is satisfiable by
showing directly that A is consistent satisfies the principle of the purity
of methods.
But it is natural to ask the question: is Hilberts Thesis true? i.e. for
every logical formula A, are the sentences A is consistent and A is
satisfiable equivalent?
Of course, for every formula A the sentence A is satisfiable implies
the sentence A is consistent: because every proof preserves truth, no
contradictory sentence may be proved from a satisfiable formula.
But: what about the inverse implication, i.e. from A is consistent to
A is satisfiable?

7. Weak Hilberts Thesis: the inverse of the second part

of Weak Dedekinds Theorem
The Weak Hilberts Thesis is the restriction of Hilberts Thesis to first-
order formulas A. The Weak Hilberts Thesis is true and may be proved
from the Completeness Theorem for first-order logic. Let A be a formula
of first-order logic. A is satisfiable is the logical sentence
X f1fnP1Pq A
where f1,,fn are all the function variables occurring in A and P1,,Pq are
all the predicate variables occurring in A, and this logical sentence is the
same as the logical sentence
X f1fnP1Pq A
By the Completeness Theorem for first-order logic, this sentence is
equivalent to the following:
66 Chapter Four

(x(x is a proof of A))

i.e. (when F stands for a contradictory sentence)
(x(x is a proof of F from A))
i.e.
x(x is not a proof of F from A))
i.e. the sentence A is consistent.
By means of the arithmetization of the logical syntax (see (Gdel
1931)), we know that for every first-order logical formula C, the sentence
A is consistent i.e.
x(x is not a proof of F from C))
becomes equivalent to the arithmetical sentence
n DER(n, C , F)
where DER(n,p,m) (n is the code of a logical proof which has as the
hypothesis a formula whose code is p and ends with a formula whose code
is m) is a quantifier-free formula of first-order PA, so that C is
consistent becomes equivalent to an arithmetical sentence belonging to
310.
Therefore, the Weak Hilberts Thesis gives the inverse of the second
part of the Weak Dedekinds Theorem, i.e.: Every logical sentence
belonging to 61 is equivalent to a first-order arithmetical sentence
belonging to 310 , so that
61 310

Indeed, if A is a logical sentence belonging to 61, then by definition A is

equivalent to the logical sentence C is satisfiable for some first-order
formula C. So, by the Weak Hilberts Thesis, A is equivalent to the
sentence C is consistent i.e. to the sentence
x(x is not a proof of F from C))
and then (by the arithmetization of the logical syntax) A is equivalent to
the arithmetical sentence belonging to 310
n DER(n, C , F) .

8. The equivalence between arithmetical sentences

and logical sentences, established by the Weak Dedekinds
Theorem, the Completeness Theorem for first-order logic,
and the Weak Hilberts Thesis
We may summarize the results, described in the previous sections, in the
following general statement.
 Dedekind, Hilbert, Gdel 67

i) Every existential arithmetical first-order sentence belonging to

610 is equivalent to a universal logical sentence belonging to 31,
and vice versa every universal logical sentence belonging to 31 is
equivalent to an existential arithmetical first-order sentence
belonging to 610 .
ii) Every universal arithmetical first-order sentence belonging to 310
is equivalent to an existential logical sentence belonging to 61,
and vice versa every existential logical sentence belonging to 61
is equivalent to a universal arithmetical first-order sentence
belonging to 310.

The proof of a) is given by the first part of the Weak Dedekinds Theorem
and the Completeness Theorem for first-order logic, by using the
arithmetization of the logical syntax.
The proof of b) is given by the second part of the Weak Dedekinds
Theorem and the Weak Hilberts Thesis (provable from the Completeness
Theorem), by using the arithmetization of the logical syntax.

Thus, these results may be expressed as follows:

a) 610 31
b) 310 61
where X Y means that every element of X has an equivalent in Y and
every element of Y has an equivalent in X .

It is difficult to read these results in a reductionistic way. Indeed, when the

fact that every sentence of a discipline S has un equivalent in the set of the
sentences of another discipline T is read as S may be reduced to T, then
these results lead us to say that arithmetic may be reduced to logic, and
also that much of logic may be reduced to first-order arithmetic!
Perhaps a better reading of these results would be the following: the
discovery of strong mutual dependencies between logic and arithmetic,
and the need (in the case we are interested in proposing, to find or to
discover the foundations of logic and/or the foundations of arithmetic) to
seek simultaneous foundations of both logic and arithmetic: i.e. what
Hilbert proposed in (Hilbert, 1905).
These results show that, given this correspondence between logic and
arithmetic, a universal arithmetical quantifier corresponds to an existential
logical quantifier, and an existential arithmetical quantifier corresponds to
a universal logical quantifier.
68 Chapter Four

Perhaps, this exchange of quantifiers when we go from an arithmetical

sentence to a logical sentence, and vice versa - is the most important
discovery provided by these results, surely an unexpected result. We know
in particular, by means of investigations within the development of
linear logic the strong difference between the universal quantifier and
the existential quantifier: the universal quantifier is given through
reversible rules so that it may be considered a negative (i.e. non creative)
operator, whereas the existential quantifier is given through not-reversible
rules so that it may be considered a positive (i.e. a creative) operator. Thus,
the difference between logic and arithmetic may be expressed as follows:
what is creative in logic becomes non-creative in arithmetic, what is non-
creative in logic becomes creative in arithmetic.

9. Completeness of logic: the collapse of the distinction

between universal and existential quantifiers
Completeness of logic is the statement: every logical sentence is
equivalent to its logical provability. This way of expressing Completeness
is better than the usual one: for every logical sentence A, A is true if and
only if A is logically provable. Indeed, by using this formulation,
Completeness looks like something related to the distinction between
semantics (the truth of logical sentences) and syntax (the logical
provability of logical sentences). But, on the other hand, it is evident that
our way of expressing Completeness is equivalent to the usual one.
A weak form of the Completeness of logic is the following: every
logical sentence belonging to 31 is equivalent to its logical provability.
This weak form of the Completeness of logic holds, and is simply the
Completeness Theorem for first-order logic.
Let A be logical sentence belonging to 31, so that (by definition) A is
equivalent to the logical validity of a first-order formula C. By the
Completeness Theorem for first-order logic, the logical validity of C is
equivalent to the logical provability of C and also to the logical provability
of the logical validity of C. Thus, A is equivalent to the logical provability
of A.
We know also that this weak form of the Completeness of logic allows
us to state the interesting correspondence between logical sentences and
arithmetical sentences discussed in the previous section. What happens,
under the hypothesis that the Completeness of logic holds in the general
form?
 Dedekind, Hilbert, Gdel 69

It is interesting to note what happens when we suppose that the

simplest extension of the Completeness Theorem for first-order logic
holds:
every logical sentence belonging to 31 or belonging to 61
is equivalent to its logical provability.

I.e. we consider the consequence to extend to the logical sentences

belonging to 61 what holds for the logical sentences belonging to 31: the
logical sentence is equivalent to its logical provability. Note that each
sentence belonging to 61 is equivalent to the satisfiability of a first-formula
and by the weak Hilberts thesis is equivalent to the consistency of a
first-order formula.
We argue now that under the hypothesis that every logical sentence
belonging to 61 is equivalent to its logical provability - every logical
sentence belonging to 61 is equivalent to an arithmetical sentence
belonging to 601 .
Let A be a logical sentence belonging to 61, so that A must be
equivalent to a logical sentence
X f1fnP1Pq B
where B is a first-order formula, f1,,fn are all the function variables
occurring in B and P1,,Pq are all the predicate variables occurring in B,
so that A must be equivalent (by the weak Hilberts thesis) to the logical
sentence CON(B)
x (x is not a logical proof of F from B)
Therefore, the logical provability of A is also equivalent to the logical
provability of
x (x is not a logical proof of F from B)
i.e. to the sentence
y(y is a logical proof of CON(B))
Thus, from the possibility of the arithmetization of the syntax, we obtain
that A is equivalent to the arithmetical sentence
n DER(n, CON(B))
i.e. to an arithmetical sentence belonging to 601 .

This result may be summarized as:

Completeness of logic implies 61 601

From this result, together with that stated in the previous sections i.e. 610
31 and 310 61, we get:
Completeness of logic implies 61 31
70 Chapter Four

and
Completeness of logic implies 310 601

Therefore, we may conclude that, under the hypothesis of the

Completeness of logic:
a) every existential logical sentence belonging to 61 is equivalent to a
universal logical sentence belonging to 31, i.e. the satisfiability of a first-
order formula is always equivalent to the logical validity of a first-order
formula;
b) every universal arithmetical sentence belonging to 310 is equivalent to
an existential arithmetical sentence belonging to 601, i.e. every number-
theoretic theorem on a quantifier-free formula is equivalent to the
solvability of an equation expressed by a quantifier-free formula.
These consequences seem rather paradoxical, and essentially lead to a
kind of collapse of universal and existential quantifiers both in logic and in
arithmetic.
The Incompleteness Theorem, proved by Gdel in (Gdel, 1931),
avoids these consequences. A nice consequence to express the
Incompleteness Theorem is: not every logical sentence belonging to 61 is
equivalent to its logical provability, i.e.
Not (61 601 )

Gdels Incompleteness Theorem avoids the collapse of quantifiers, in

arithmetic and in logic! So, it is not a disaster for logic or for arithmetic:
rather, it saves logic and arithmetic from a collapse of quantifiers!
Moreover, Gdels Incompleteness Theorem says that the fact that an
arithmetical sentence A is equivalent to a logical sentence B does not
imply that the provability of A is equivalent to the logical provability of B:
it is possible that the logical sentence B has no logical proof.

References
Dedekind R. (1888). Was sind und was sollen die Zahlen? Braunschweig:
Vieweg & Sohn.
Girard J.-Y. (1987), Proof Theory and Logical Complexity, Volume 1,
Napoli: Bibliopolis.
. (2006-07), Le point aveugle. Paris : Hermann.
Gdel K. (1930), Die Vollstndigkeit der Axiome des logischen
Funktionenkalkls, Monatshefte fr Mathematik und Physik, 37: 349-
360.
 Dedekind, Hilbert, Gdel 71

. (1931), ber formal unentscheidbare Stze der Principia Mathematica

und verwandter Systeme I, Monatshefte fr Mathematik und Physik,
38: 173-198
Hilbert D. (1905), ber die Grundlagen der Logik und der Arithmetik,
Verhandlungen der Dritten Mathemstiker-Kongresses, Leipzig:
Teubner: 174-185.
. (1935), Gesammelte Abhandlungen, Dritter Band: Analysis,
Grundlagen der Mathematik, Physik, Verschiedenes, Berlin: Springer.
Hilbert D. and Bernays P. (1934-39), Grundlagen der Mathematik, Berlin:
Springer.
 CHAPTER FIVE

THE STATUS OF MATHEMATICAL

KNOWLEDGE

DAG PRAWITZ

SUMMARY I am interested here in defending the very commonplace view

that mathematical knowledge is essentially obtained by deductive
inference. This view has been challenged by Carlo Cellucci in a number of
books and papers. I argue that the view is less easily defended than is
usually assumed, because when the deductive validity of an inference is
defined in the standard way in terms of logical consequence, it is simply
not true in general that a deductively valid inference applied to premisses
for which conclusive grounds are available yields a conclusive ground for
the conclusion. The notion of deductively valid inference must therefore
be reconsidered. I suggest that it should be explained in terms of
conclusive grounds. But Cellucci argues rightly that this is not enough in
order to defend the challenged view, because it must also be shown that
there are conclusive grounds for the initial premisses of the deductive
proofs used in mathematics; otherwise the deductive method collapses into
what he calls the analytic method. It is suggested by examples how this
requirement may be satisfied so that an argument could be obtained
ultimately for the view that we have conclusive grounds for the
propositions asserted in mathematics.

KEYWORDS deductive inference, validity, mathematics, knowledge,

axiomatic method

In a number of thought-provoking books and papers, Carlo Cellucci has

criticized several well-known views in philosophy of mathematics.
Advocating a fresh start in this discipline, he has outlined a new approach
with radically different views. The main target of his criticism has been
the axiomatic method or, more precisely, the view that it is by this method
74 Chapter Five

that mathematical problems are solved. In its place, Cellucci wants to put
what he calls the analytic method. Examples of other themes that he
discusses are the relation between discovery and justification, the role of
intuition, the nature of mathematical objects, and the character of
mathematical definition. But questions concerning what method is used for
solving problems and the role of proofs are the central ones, and I shall
restrict myself here to these questions.
Certainly, after Gdels incompleteness theorem, one cannot think that
mathematics is rightly characterized by saying that it aims at proving
theorems in given axiomatic systems. However, Celluccis criticism,
although based first of all on the incompleteness theorem and explicitly
directed against identifying mathematics with the axiomatic method, is in
effect more sweeping and hits more entrenched ideas in philosophy of
mathematics, or at least, so it seems. The axiomatic view was surely very
influential before Gdel, but was perhaps not as dominating as Cellucci
suggests. I think for instance that the ideas presented by Bertrand Russell
in his book Introduction to Mathematical Philosophy 1 are representative
of a different view that was already in agreement with several of
Celluccis critical points. Russell speaks of two directions of mathematics.
One starts with propositions whose truth is taken to be already known and
aims at obtaining new knowledge by making deductions from this starting
point. Another starts from the same point but aims instead at deducing
these known truths from more elementary principles. The first direction is
the familiar one. In this ordinary mathematics one is not concerned with
axioms. To move in the other direction is to practice a less familiar form
of mathematics that Russell calls mathematical philosophy. The endeavour
is then to find axioms that are sufficient to deduce what we consider
ourselves to know already, and this is a process that need not have a
definite endpoint, since the axioms that have been obtained at one stage
may be possible to deduce from even more fundamental principles. In
neither case do we start from axioms or practice the axiomatic method as
described by Cellucci.
However, Russell takes for granted that mathematics is concerned in
either case with deductions or deductive proofs. In this way he concurs in
the deeply rooted opinion that mathematics, after an early empirical
beginning, has been essentially a deductive enterprise. I am here mainly
interested in discussing whether this view can be defended against the
general criticism that Cellucci delivers.

1
Russell (1919).
 The Status of Mathematical Knowledge 75

I am aware of the fact that this view may become abandoned to some
extent in our time because, by employing computers in research,
mathematics may begin to rely on propositions that are established
empirically.2 But this question is not discussed by Cellucci and will not be
taken up here either.

1. Celluccis notions of axiomatic and analytic proof

Celluccis discussion of the axiomatic and analytic method is closely
connected with what he calls the notions of axiomatic and analytic proof.
He contrasts the two kinds of proofs by emphasizing three opposite
properties that characterize them: axiomatic proofs are deductive, start
from primitive premisses that are held to be true, and aim at justifying the
conclusion, while analytic proofs are non-deductive, start from problems,
and aim at discovering hypotheses that are capable of solving them. The
analytic method for solving a problem contains two distinct processes.
Firstly, one uses analytic proofs in order to find hypotheses that are
capable of solving the problem, and secondly, one makes a choice among
the hypotheses so obtained on the grounds of their plausibility, which is
made after weighing arguments for and arguments against the different
hypotheses. A chosen hypothesis is not claimed to be true, only to be
plausible in the sense that the arguments for it prevail over those against.
A hypothesis therefore constitutes a new problem and is the starting point
for a new round of the analytic method and so on, ad infinitum.3
According to Cellucci, in the last century most mathematicians have
thought of themselves to be pursuing axiomatic proof. But they
werent4. In fact, although this is not generally recognized, since
antiquity the analytic method has been the basis for problem solving, not
only in mathematics but also in natural science and medicine5.
As Cellucci points out, because of Gdels incompleteness theorem we
cannot expect in general to be able to solve mathematical problems by
deducing answers from some given set of axioms. However, this is not an
argument against the general view that in mathematics one solves
problems by deductive proofs from premisses held to be true. Cellucci
seems to reject such a view for two main reasons: (1) the process by which

2
Prawitz (2008).
3
Cellucci (2008, pp. 2-3).
4
Ibid, p. 13
5
Cellucci (2013, p. 34)
76 Chapter Five

one finds a hypothesis that is capable of solving the problem is a proof

process, although not a deductive one, and it is therefore too restrictive to
describe mathematical problem solving as a deductive enterprise; (2) the
common view that a mathematical problem is solved by deducing a
theorem from known truths is inadequate because generally there is no
rational way of knowing whether primitive premisses are true; the initial
premisses of a deductive proof are only more or less plausible hypotheses.6
Argument (1) is partly, at least, a question of terminology. There is a
course a heuristic phase in the solution of a mathematical problem in
which guesses are made and different strategies are tried. The search for
suitable premisses from which it would be possible to deduce an answer
can to some extent be described as a rule-governed process, and we may
choose to call it proving. But there are good arguments against stretching
the term proof or inference that far. What Cellucci has in mind is in
particular a process where one applies deductive rules back-wards, so to
say. For instance, let the problem be whether B holds, and let us assume
that in trying to answer this problem in the positive, we may note that B
follows deductively from A and A o B. If we have reason to hold A o B
as plausibly true, we can then set up the problem of proving A. Continuing
in this way by applying deductive rules backwards, we arrive at a
construction which, when read in the appropriate direction, represents a
deduction of B from some initial premisses. Provided that they can
reasonably be accepted as hypotheses, that is, hypotheses not just assumed
for the sake of argument but accepted as plausibly true, we may have an
argument for holding B true. Now, Cellucci notes that A follows from the
two premisses B and A o B by the non-deductive rule of abduction.
Admittedly, we sometimes reason non-deductively in this way when we
want to explain why B is true: having reasons for holding that A is a
sufficient condition for B and this being the only sufficient condition for B
that we know of, we venture to draw the conclusion A from the premisses
A o B and B. We may reasonably call this an inference act where we pass
from some beliefs represented by the premisses to a new belief represented
by the conclusion. However, in the case considered by Cellucci, B is a
problem, that is, we do not yet hold B true, and then it does not seem
adequate to speak of an inference. To be sure, the act of setting up the
problem of proving A because of being in a situation where we hold A o
B true and want to prove B may be represented by the schema of
abduction,

6
Cellucci (2008, pp. 3 and 12)
 The Status of Mathematical Knowledge 77

B AoB
A
Though, in this case where B is not held or assumed to be true but is only
something we would like to prove, we do not normally say that the schema
represents an inference. It is not just a matter of convention whether we
read an arrangement of sentences top down or bottom up, as Cellucci
suggests.7 In one case, we may have something that represents a deductive
proof, and in the other case, something that is better called a proof search
actually, a kind of proof search that has for a long time been implemented
on computers within automated deduction.
It seems therefore preferable to call the first process involved in the
analytical method by which hypotheses are found a search process rather
than a proof process. Anyway, regardless of what we think of this, even in
Celluccis parlance, it is essentially a deductive matter that the found
hypothesis or hypotheses are capable of solving the problem, because what
this means, if I understand him rightly, is that an answer to the problem is
deductively derivable from them. The proof process involved in the
solution provided by the analytic method seems therefore to be deductive
just as in the case of the axiomatic method. Nevertheless, there remains an
essential difference between the two methods, because, according to
Cellucci, the axiomatic method is an unjustified truncation of the analytic
method8 it is a truncation because it limits itself to deduction of an
answer to a given problem from some accepted premisses, which is only
one first phase of the analytic method, and this truncation is unjustified,
because the premisses should not be taken as unproblematic truths, but as
new problems to be solved by another round of the analytic method. Thus,
the main argument for the inadequacy of the deductive method seems to
depend on the argument (2) above.

2. Inferences and knowledge

Although, as we have seen, deductive proofs occur as an essential part of
the analytic method, Cellucci plays down their importance and hardly
mentions them in his description of the method. This depends presumably
on his view that it is only thanks to the use of non-deductive inferences
that new knowledge can arise; deductive proofs cannot give rise to this

7
Ibid, p. 4.
8
Ibid p. 3.
78 Chapter Five

they are according to him non-ampliative.9 This view seems to stand in

sharp contrast to the opinion that I am trying to defend, namely, the view
that mathematics, after its deductive turn in ancient Greek, is essentially a
deductive science, which is to say that it is by deductive proofs that
mathematical knowledge is obtained.
To discuss these seemingly opposite views fruitfully we must first look
more carefully at the notions of knowledge and inference. Cellucci stresses
the importance of a global approach to questions about the nature of
mathematics that places them within the general context of knowledge and
biological evolution.10 He says, all organisms survive by making
hypotheses on the environment by a process that is essentially an
application of the analytic method; in particular, our hunting ancestors
solved their survival problems by making hypotheses about the locations
of preys on the basis of hints and similarly mathematicians solve
mathematical problems by making hypotheses for the solution of problems
on the basis of hints provided by them11. I shall try to follow him in such
a global approach.
I think it is right to say that all organisms make inferences and that this
ability is essential for their survival. By an inference I then mean a mental
act in which an agent forms a new belief on the basis of some of her earlier
beliefs, in other words, a transition from some premisses that represent her
current beliefs to a conclusion that represents a new belief; typically, the
premisses are beliefs entertained because of some observations, while the
conclusion is a belief about something of importance for the agent, for
instance, concerning a possible prey or a threatening danger. In the case of
animals and for the most part also in the case of humans, these inferences
are intuitive or automatic, that is, the agent is not aware of her beliefs and
of the transition between them. The beliefs are in other words implicit, and
it is we who ascribe them to the agent in our attempts to explain her
behaviour; it seems reasonable to explain certain sudden actions by
assuming that a new implicit belief has been drawn as a conclusion of
earlier implicit beliefs.
It can be said that an agent answers a question and in this sense solves
a problem by making such inferences, most of which are certainly non-
deductive, and that this resembles applications of what Cellucci calls the
analytic method. The conclusions drawn by the agent could be called

9
Cellucci (2011, p. 128), Cellucci (2013, 34).
10
Ibid, pp. 32-33.
11
Cellucci (2008, p. 15).
 The Status of Mathematical Knowledge 79

hypotheses in the sense that she may be ready to give them up because of
new inferences. But they are not normally taken as new problems. Rather,
they are beliefs firm enough to be the basis for actions. Clearly, it is
essential for survival that they are correct for the most; that is, what is
believed must be true on the whole, if the inferences are to be beneficial
for survival. By true I then mean truth in the usual, unsophisticated
sense, not to be identified with truth as possession of a model, as
consistency, or as convention, which are the three ideas of truth that
Cellucci mentions in his criticism of the axiomatic method.12 I assume that
Cellucci also relies on such a notion when he speaks about the need to
investigate whether the hypotheses obtained by analytic proofs are
plausible; that is, I presume, it is plausible that they are true.
Inferences of this kind are often genetically hardwired and show
themselves in the form of stereotyped actions in response to certain
stimulus. In higher animals they may be more flexible and depend on
several parameters including earlier experience. With humans they
become not only more flexible but sometimes also what I shall call
reflective, that is, the premisses and the conclusion appear as explicit
beliefs, and the agent is aware of the transition that occurs, sometimes
even critically examining the inference step as well as the beliefs that
occur as premisses.
The evolutionary advantage of being able sometimes to hold up an
intuitive or automatic inference and make it reflective is obvious, and we
may be genetically disposed to perform reflective inferences. The point is
of course to increase the chances for beliefs to be correct, that is, for what
we believe to be true, and the same holds for the culturally developed
practice to demand good grounds for our own and others beliefs.
Reflective inferences constitute one major source of our grounds for
beliefs, and we evaluate the acceptability of an inference with respect to
how good ground it yields for the conclusion. We must often base our
actions on beliefs whose grounds are far from conclusive, but we learn to
match the strength or quality required of grounds to the importance and
seriousness of the context in which we form beliefs and make actions. It is
generally held that the ground required to make a mathematical assertion
is of the strongest possible kind a ground that we call conclusive.
The notion of ground is also a key notion for the Platonic idea of
propositional knowledge. In order to have such knowledge, it is not
enough that the proposition in question happens to be true, we must also

12
Ibid, pp. 10-12.
80 Chapter Five

possess a good ground for holding it true, in other words, be justified in

holding it true. Reflective inferences have in this way a justificatory aim,
the aim to find good grounds for beliefs and to provide new knowledge
thereby, and in the light of what has been presented above, we can say,
using a formulation of Cellucci, that reflective inference appears as a
continuation of strategies by which organisms solve their problems13.
Such an aim is essential also to Cellucci, who emphasizes that the
hypotheses that come out from an analytic proof must be examined with
respect to their plausibility. But he is sceptical not only to the idea of
conclusive grounds but also to the more general idea that we can come to
know things, that is acquire grounds for beliefs, by deductive inferences,
which we have to consider next.

3. Deductive inference and knowledge

The distinctions reflective/non-reflective and deductive/non-deductive
applied to inferences cut each other. We certainly often make deductive
inferences without being aware of it. Animals also perform such inferences.
A conceivable example is suggested by Chrysippus (3rd century B.C.): A
dog is running along a road following his master who is out of sight. When
coming to a fork the dog sniffs one of the roads. Finding no trace of the
master, he suddenly sets of, without sniffing, along the other road.
Regardless of what one thinks about Chrysippus dog, humans doubtless
make automatic inferences conforming to disjunctive syllogism (modus
tollendo ponens): without being aware of holding a certain disjunction true
and without noticing that we have made an inference, we start to act as if
we held the second disjunct true upon having made an observation refuting
the truth of the first disjunct. I shall not try to make a detailed explanation
of this kind of behaviour, but it seems reasonable to assume that
experiences indicating that certain disjunctions are true become stored
somehow and that we have some innate ability, Cellucci calls it natural
logic, to use this stored information together with the information that one
disjunct is false. I see no reason to exclude that some animals may have a
similar capacity.
Deductive inference, being a species of inference, must be expected to
have the same general aim as inference, that is, to take the agent from one
state of belief to another one where a new belief is formed. Similarly,
reflective deductive inferences should have the same use as reflective

13
Ibid, p. 21.
 The Status of Mathematical Knowledge 81

inference in general to give a good ground for the conclusion and thereby
provide new knowledge in a Platonic sense. A possible objection could be
that in the case of deductive inference the belief cannot be new because
the agent must have already entertained this belief if she held the
premisses true.
There is a philosophical dictum, which at least used to be fairly common,
saying that the content of the conclusion of a deductive inference is
already contained in the content of the premisses. This has been taken to
stand in opposition to the idea that deductive inference may generate new
knowledge. What was meant was seldom articulated in such a way that a
real opposition emerged. Sometimes the supposed opposition was taken to
constitute a paradox that had to be explained, because it was acknowledged
that it seemed to be a common experience that new knowledge was
generated by deductive inference. Since there was no real argument for the
view that the latter would be impossible, it was never clear what was to be
explained.
What have to be made more precise in this discussion is not only what
one means by the content of the conclusion being contained in the content
of the premisses, but also what one takes knowledge and new knowledge
to mean. It is true that there is no agreement on how the notion of
propositional knowledge is to be analysed, nor on whether the analysis can
be expected to result in a definition of the notion of knowledge in terms of
other notions or only in establishing certain relations between the notion of
knowledge and some other notions.
In the case of a Platonic or justificatory notion of knowledge, one
would say that a person acquires new knowledge, if she gets in possession
of a ground for holding a proposition true and earlier lacked any ground
for such a belief. It does not seem likely that there is a cogent argument for
saying that a deductive inference cannot bring about new knowledge in
that sense.
Some think that knowledge should be analysed in terms of reliable
processes that make us hold a proposition true. Certain deductive inference
procedures are then counted as reliable processes, and are hence seen as
giving us knowledge. It seems out of question that one could argue for the
view that such a process could not make one hold a proposition true that
one did not hold true earlier.
Cellucci seems to concur in the philosophical dictum that we are
discussing, saying that deductive inference are non-ampliative, which is to
say that the conclusion contains no more than was implicitly contained
82 Chapter Five

in the premisses14. However when it comes to explain in more detail what

this means, he does not use to this metaphorical expression but says that an
inference is non-ampliative means that the conclusion either is literally a
part of the premises, or implies nothing that is not already implied by the
premises.15 The second disjunct in this explanation is equivalent,
presupposing some elementary properties of implication, to saying a little
simpler that the premisses imply the conclusion. Whether this property in
turn has the consequence that the inference cannot generate new
knowledge is the issue under debate and this cannot therefore be assumed
here without begging the question. Clearly, the first disjunct, that the
conclusion is a literal (or syntactic) part of one of the premisses, does not
exclude that a new ground is obtained by the inference. For instance, an
inference of the kind considered by Chrysippus, conforming to the
disjunctive syllogism, is counted as non-ampliative since one of the
premisses contains the conclusion as one of its disjuncts. Clearly, a person
can lack a ground for asserting a sentence, although she has grounds for
asserting a number of other sentences one of which contains the first
sentence as a part, embedded in a number of logical constants; for
instance, she may have lacked grounds for asserting B until she infers by
the disjunctive syllogisms that B must be true, having grounds for holding
A and the disjunction A B true.
Therefore, a deductive inference being non-ampliative in Celluccis
sense stands in no way in opposition to the possibility that it gives new
knowledge in the sense intended here.16 Although no argument is known
for the view that deductive inference cannot generate knowledge, I have
said nothing so far that actually shows deductive inference to have the
privileged status with respect to mathematical knowledge that it is
commonly claimed to have. To get anywhere in that direction we must
first of all attend more closely to the notions of deductive inference and
ground.

14
Cellucci (2011, p. 128).
15
Cellucci (2013a, p. 55).
16
Celluccis notion of non-ampliative is really intended for another context than
the present discussion, namely, when asking what inferences allow us to infer a
new hypothesis from current scientific hypotheses or from observation data.
(Private communication).
 The Status of Mathematical Knowledge 83

4. Deductive inference and ground

It should be noted that a deductive inference does not in general preserve
the level of probability or plausibility as Dummett (1973) laconically
remarks, this fact supplies a rational to those who, usually on incoherent
grounds, distrust complicated chains of arguments17. Thus, there is no
guarantee that a person who has good but non conclusive grounds for the
premisses of a deductive inference gets a good ground for the conclusion
by performing the inference.
This fact may make us doubt that deductive inference has a privileged
status in mathematics when it comes to getting grounds for its assertions.
When the grounds for the premisses of a deductive inference are non-
conclusive, it may be appropriate to require in addition arguments for and
arguments against the conclusion. We may require of a good but non-
conclusive ground for a belief that it is also argued that the arguments for
are stronger than the arguments against, which is what Cellucci requires of
a plausible hypothesis18 (of course, we have to be on our guards for a
regress here). Then, if grounds in mathematics were non-conclusive, it
would be equally true in mathematics as it is in natural sciences, that
deductive inference is not enough in order to establish theorems or solve
problems.
What are we then to mean by a deductive inference? The answer is not
at all obvious. In the above I have taken an inference in general to be a
mental act of a certain kind. If we now restrict ourselves to reflective
inferences, as is reasonable when we are discussing the nature of
mathematics, we may focus on the propositions or sentences that represent
our beliefs, and we may count them, instead of beliefs, as premisses and
conclusions of inferences. It is now often said that truth preservation is the
defining characteristic of deductive inference. What this means cannot be
explained by simply saying that if the premisses are true, then so is the
conclusion. If the conditional is meant as a material one, the condition is
of course too weak, and if one tries to specify a stronger conditional, for
instance that the implication is to hold in all possible worlds, one gets
entangled in a number of difficult questions.
The common and well-known explanation of what is meant by truth
preservation in this context is that if the premisses are true then so are the
conclusion under all variations of the non-logical terms involved. This

17
Dummett (1973, p. 27).
18
Cellucci (2013, p. 34).
84 Chapter Five

equates a valid (deductive) inference with the conclusion being a logical

consequence of the premisses, as this latter notion is usually explicated. It
makes the notion of valid inference depend on the notion of logical
constant, which is admitted as a problematic feature even by those who
adhere to the equation. The real problem with this equation is however that
it makes the notion of deductive inference both too narrow and too broad.
It is too narrow because it leaves out typical mathematical inferences such
as complete induction. It is too broad because in general it is not true that
an inference that is valid according to this explanation and that is applied
to premisses for which grounds are available yields a ground for the
conclusion; just think of all valid inferences whose validity can be
established only by arduous proofs or whose validity has not been
established.
The last fact is particularly important. The view that I want to defend is
that mathematical knowledge is obtained by deductive inferences from
truths that are considered to be obvious. But, since inferences that preserve
truth do not generally preserve availability of grounds, if by valid
deductive inferences are meant those that preserve truth, they cannot be
claimed to generate knowledge. Therefore, a deductive proof cannot be
defined as a sequence of inferences that are valid in the usually proposed
sense. In other words, the so-called valid inferences are not in general
legitimate in deductive proofs, even if it is a necessary condition for an
inference to be deductively legitimate that it does not lead from truth to
falsity.
The difficulties that we meet when we try to characterize the deductive
inferences may lead us to think that there is no such special category of
inferences. But this would be to ignore common experience that we have
from many walks of life and science. Let me give two examples. Anyone
who has played with Sudoku and crosswords knows that there is a
principle difference between these two kinds of puzzles. The obvious
difference is that every correct step in the solution of a Sudoku is a
deductive inference from the rules of the puzzle and what has so far been
written into the boxes of the grid, while it is a matter for discussion
whether the solution of a crossword agrees with the clues that are given.
Similarly, there is a striking, principle difference between the discussions
pro and con that is typical when experts try to formulate a policy for
energy productions and the usual form that the solution of a mathematical
problem assumes. In the former case, the delivery of arguments pro and
con is an essential element in the task of the expert, while such arguments
are perspicuously absent from the solution of a mathematical problem.
And this is not just a matter of style. If the mathematician was to present
 The Status of Mathematical Knowledge 85

the decisions that he made when choosing his or her strategy for solving
the problem, there could be arguments pro and con, but they would still be
absent from the solution that has been found.
There is no general agreement about what characterizes a deductive
inference. But one can hardly make any progress in the project I am
considering without proposing such a characterization. As has already
been suggested, it must be required of an inference for being legitimate in
a deductive proof, that it yields a conclusive ground for the conclusion
when there are conclusive grounds for the premisses. I think that this is
also a sufficient condition for being legitimate. One may hope for a more
informative characterization of the legitimate deductive inferences, saying
what it is that gives an inference the epistemic force to yield a conclusive
ground for the conclusion given such grounds for the premisses. It is not
possible in this paper to go deeper into how this idea can be developed19,
but I shall make some comments on the notion of conclusive ground,
which I hope will be sufficient to see how there may be such grounds for
the mathematical propositions we hold true.

5. Conclusive ground
One may have doubts about the actual existence of conclusive grounds,
and if so, the characterization of deductive inference in terms of them may
seem pointless. Such doubts may come from the insight that one can never
exclude the possibility of errors in human affairs. But it is important not to
confuse two different notions involved here. Consider a computation of
two arithmetical terms that are built up by the usual arithmetical
operations. If it gives the same result, it counts as a verification of the
identity statement, asserting that the two terms denote the same number.
The computation deserves to be called a conclusive ground for this
assertion. But of course the existence of such conclusive grounds does not
exclude the possibility of mistakes in calculations that we make.
Embarrassingly enough, one sometimes believes a false identity statement
to be true, having not discovered a certain error in the attempted
computation, in spite of having checked it several times. In this sense, one
can say that one can never absolutely exclude the possibility of errors.
Nevertheless we know what we mean by an arithmetical computation.
There is nothing particularly obscure about it. And it is entirely right to say
that if a computation of two terms gives the same result, then the

19
See further Prawitz (1914).
86 Chapter Five

proposition or sentence that identifies them is true not only plausibly

true.
This should dispel the feeling that the existence of conclusive grounds
is illusory; what is illusory is that we can completely exclude the
possibility that we are mistaken in believing that we are in possession of a
conclusive ground. We may tie the fact that a computation constitutes a
conclusive ground for an arithmetical statement to what we mean by that
statement, saying that to assert the statement is to assert that there is, in a
tenseless sense, a certain computation of the two terms which gives the
same result. If this is what we mean, then the truth of the statement should
amount to the existence of such a computation, and having actually
produced it, we are in possession of a conclusive ground for the statement.
Similarly, one may say that what is meant by natural number is
explained by saying that 0 is a natural number and that the successor to a
natural number is again a natural number. Therefore, anyone who knows
this meaning has a conclusive ground for asserting that 0 is a natural
number and that the successor of a natural number is a natural number.
Let us say that to have a conclusive ground for asserting an open
statement B(x) under the assumption A(x) is to know an effective operation
that, given a term t and a conclusive ground for the assertion of A(t),
transforms the ground to a conclusive ground for the assertion of B(t).
Then it follows that given a conclusive ground for A(0) and for the
assertion of A(sx) under the assumption A(x), where x ranges over the
natural numbers and sx stands for the successor of x, one can find a
conclusive ground for A(n), given any natural number n: one simply takes
the given ground for asserting A(sx) under the assumption A(x), and
applies it successively n times to the given ground for A(0); that is, one
first applies it to the ground for A(0), then applies it again to the result so
obtained, and so on.
This is an example of how, for a deductive inference conforming to the
principle of complete or mathematical induction, there is an operation that
transforms conclusive grounds for the premisses to a conclusive ground
for the conclusion. Cellucci remarks that to appeal to facts of this kind in
order to justify a deductive inference is circular.20 It is true that to show
that the described operation really yields a ground for the conclusion we
need to apply mathematical induction. This is a relevant remark in case we
want to persuade a person doubting the cogency of mathematical induction
of its cogency. But it is not a relevant objection to the claim that the

20
Cellucci (2011, p. 130).
 The Status of Mathematical Knowledge 87

operation does yield a conclusive ground for the conclusion given the
stated presuppositions.
It should thus be clear that there are conclusive grounds for assertions
which we are in possession of when knowing what the assertions mean, as
for instance in the case the assertion that 0 is a natural number or that t = t.
Similarly, we know in other cases how to get in possession of a conclusive
ground by making certain operations, such as carrying out computations to
ground the assertion of an arithmetical identity t = u. Furthermore, the
example of mathematical induction illustrates how given conclusive
grounds for the premisses of a deductive inference are transformed to a
conclusive ground for the conclusion by a certain operation.
To vindicate in this way the general idea that mathematic knowledge is
obtained by deductions from truths already known, one has to show that
for all deductive inferences used in mathematics there are operations that
transform conclusive grounds for the premisses to a conclusive ground for
the conclusion. This has to be seen as a project whose possible realization
I have only given a hint of above.21
However, this is not enough, because as Cellucci remarks, if the
axioms are only plausible propositions, then the notion of axiomatic
proof collapses into that of analytic proof22. A similar collapse threatens
to occur for the view that I am advocating, if it cannot be shown that
accepted deductive proofs can ultimately rest on initial premisses for
which we have conclusive grounds, because if the strength of grounds for
the initial premisses are less than being conclusive, then there is no
guarantee that the deductive proof preserves the level of plausibility of the
initial premisses, and then, as already pointed out, one may reasonably
require something more than a deductive proof in order to get a good
ground for the conclusion. This I see as the most challenging part of
Celluccis argument.
Argument (2) as stated above (section 1) goes clearly too far. The
analytic method as Cellucci describes it, or just careful arguments pro and
con, may function as a method for getting to know that the initial
premisses are true, at least if one uses the terminology developed here; a
fallible method of course, but nevertheless a rational method that may
result in knowledge. What is meant is presumably that there is no way of
finding conclusive grounds for the initial premisses. I have indicated a way
in which one may find such grounds in virtue of what one takes the

21
For more details about such a project, see Prawitz (2014).
22
Cellucci (2008, p. 12).
88 Chapter Five

involved concepts to mean. But it is far from obvious that by this strategy
one succeeds in finding conclusive grounds for all the propositions that are
taken to be obviously true and are therefore allowed as initial premisses in
mathematical proofs. This project is equally essential in order to meet the
challenge posed by Cellucci.

6. Remarks on the deductive turn of mathematics

The virtue of the commonplace view that I am trying to defend is that it
seems to tally with what historians call the deductive turn of mathematics
in ancient Greeks. The Babylonians, whose mathematics was well
developed in many respects, used freely, for instance, the regularity later
known as Pythagoras theorem without trying to back it up deductively, as
far as we know. They certainly had abundant inductive evidence for its
truth. Here I find it quite adequate to use Celluccis way of speaking and
say that they solved various geometrical problems by the help of a
plausible hypothesis. In contrast, the Greeks were able to prove the
theorem deductively from principles that they considered obviously true.
The Babylonians knew Pythagoras theorem, most philosophers would
say, in view of the overwhelming inductive evidence that they had. But the
Greeks considered themselves to know the theorem in a new way, and
their posterity has shared this view. Since the Greeks, most mathematicians
understand themselves as not making mathematical assertions unless they
believe it has been established deductively, which means that there is a
deductive proof of the assertion whose initial premisses are established
deductively or express obvious truths; ultimately all assertions are
consequently seen as resting on deductive proofs whose initial premisses
are assertions of obviously true sentences. Mathematicians are also usually
careful to distinguish between categorical and hypothetical assertions in
case the initial premisses are not considered to be established deductively
or to assert obviously true sentences, they limit themselves to making the
corresponding hypothetical assertion.
Celluccis position is in manifest conflict with this way of understanding
the nature of mathematics. He wants to say for instance that Ribet solved
Fermats problem by showing that the Taniyama-Shimura conjecture or
hypotheses is a sufficient condition for its solution23. But he is aware of
the fact that most mathematicians object to saying so. They say that Ribet
showed that the conjecture implies Fermats Last Theorem, and that Wiles

23
Ibid, pp. 3-4.
 The Status of Mathematical Knowledge 89

solved the problem by proving the conjecture and hence the theorem; not
that Wiles gave an axiomatic proof, but he inferred the theorem
deductively from initial premisses that were agreed by mathematicians to
express known truths.
The deductive turn of mathematics seems to have taken place already
at the time of Plato and Aristotle, thus before the idea of axiomatic
systems. Although we do not know in detail how this occurred, it seems
clear that at this time mathematicians considered themselves to have
proved theorems, not only to have established hypothetical assertions.
The issue is whether they were right in thinking so. Cellucci maintains
that they were not, that there was no principle difference between the
simple initial premisses from which the Greeks proved their theorems and
the hypotheses that Babylonians had used when solving geometrical
problems. It seems reasonable to say the burden of proof rests on those who
claim that there is a principle difference. To say only that the mathematical
proofs start ultimately from obvious truths is not a satisfactory response.
The axiomatic method lay near at hand as an attempt to improve the
response. It made precise which truths one had to rely upon and one could
try to make plausible that they were self-evident.
Mathematicians in general have seldom cared about deducing what
they considered to be already known truths from more fundamental
principles; with Russell we could say that they are usually not interested in
mathematical philosophy. In the case of arithmetic, there was not even a
proposal of an axiom system before the end of the 19th century.
The axiomatic idea was nevertheless so influential philosophically that
it came to be considered the ideal methodology of science in general. This
caused scientists in the 17th century such as Huygens to plead for the view
that in his field one will find a kind of demonstration that does not create
as high a degree of certainty as the geometric proof and that is actually
very different from the method of proof of the mathematicians. For they
prove their propositions from certain and unassailable principles whereas
here the principles are tested by means of their consequences24. Newton
made similar remarks in the direction of the hypothetical-deductive
method. There is of course a similarity between this method and the
analytic method that Cellucci describes, although he does not suggest that
hypotheses are tested as in the hypothetical-deductive method by
comparing some assertions that can be deduced from them with observable
phenomena.

24
C. Huygens, Treatise on Light, Chicago, 1945, p. vi.
90 Chapter Five

In mathematics, the axiomatic idea kept its sway for a longer time. I
would say that the decline and fall of the axiomatic method, to use an
expression of Cellucci, occurs in the 19th and 20th centuries the decline
comes with the idea that the axioms do not express obvious truths but are
arbitrarily chosen, and the fall came with Gdels incompleteness theorem.
The fall of the axiomatic method does not affect the view that
mathematical knowledge is acquired by deductive proofs from obvious
truths, because this view is not tied to the idea that one can specify once
and for all a set of axioms from which all deductive proofs are to start. For
instance, in arithmetic a deductive proof of an assertion can start from
reflective principles that are not given in advance but are formulated in the
context of the assertion in question and are then seen to be obviously true.
The challenge to explain in what way the ultimate starting points for
mathematical proofs are obvious truths remains however. It is an open
question whether the project that I outlined to explain this can be carried
through. I see Celluccis criticism as a stimulating challenge of a common
view about mathematics that has been accepted on too weak grounds.

References
Cellucci, Carlo (2008), Why Proof? What is a proof?, in Deduction,
Computation, Experiment, R. Lupacchini and G. Corsi (eds), Berlin:
Springer, pp. 1-27.
. (2011), Classifying and Justifying Inference Rules, in Logic and
Knowledge, C. Cellucci et al (eds), New Castle upon Tyne: Cambridge
Scholars Publishing, pp. 123142.
. (2013), Philosophy of mathematics: Making a fresh start, Studies in
History and Philosophy of Science 44, pp. 32-42.
. (2013a), Rethinking Logic: Logic in Relation to Mathematics,
Evolution, and Method, Dordrecht: Springer (forthcoming).
Dummett, Michael (1973), The Justification of Deduction, (Proceedings of
the British Academy, vol. LIX), London: Oxford University Press.
Prawitz, Dag (2008), Proofs Verifying Programs and Programs Producing
Proofs: A Conceptual Analysis, in Deduction, Computation,
Experiment, R. Lupacchini and G. Corsi (eds), Berlin: Springer, pp.
81-94.
. (2014), Explaining deductive inference, in Dag Prawitz on Proofs and
Meaning, H. Wansing (ed.), Dordrecht: Springer Verlag (forthcoming).
Russell, Bertrand (1919), Introduction to Mathematical Philosophy,
London: George Allen & Unwin, Ltd.
 CHAPTER SIX

EMPIRICISM AND EXPERIMENTAL

MATHEMATICS

GABRIELE LOLLI

SUMMARY After a brief survey of the recent humanistic philosophy of

mathematics known as neo-empiricism, and its relation with so-called
experimental mathematics, we note the limitations of this philosophy, due
to it considering mathematics only from the point of view of procedures.
We criticise the idea that mathematics can be characterised by its method.
We propose as a model of a philosophy of experimental mathematics the
ideas of Pavel A. Florenskij, a Russian thinker now rediscovered after
years of disgrace. His suggestions are consistent with the great didactic
tradition of the Russian school of mathematics.

KEYWORDS neo-empiricism, experimental mathematics, physics

1. The new empiricism

Towards the end of the twentieth century, several invitations to revive the
philosophy of mathematics led to the appearance of different trends with a
common denominator, that of considering mathematics as a human
activity. Although this theory might appear tautological, it has proved
fruitful. Those who have contributed to spreading these so-called
humanistic philosophies of mathematics include Reuben Hersh, together
with Philip Davis (Hersh 1979, and Davis, Hersh 1980). An anthology of
the more interesting papers on the philosophy of mathematics of the late
twentieth century was edited by Tymoczko (1998). In Davis, Hersh (1980)
many threads are intertwined, from empiricism to social contrstructivism,
uphold by Wilder, Ernest, Bloor and others. We do not intend, however, to
give a survey of these positions (already in Lolli 2006); we wish to dwell
on a particular trend, that of new empiricism, which may have the
92 Chapter Six

strongest ties with mathematical practice, thanks to the pervasive, or

invasive, influence of the computer.
To look at mathematics as an activity in continuity with other
expressions of human industriousness has suggested the theory that the
methods and procedures of mathematics are no different from those of the
natural sciences.
The traditional empiricist philosophy of mathematics aimed to provide
a foundation for and an explanation of numbers and other basic mathematical
concepts in terms of sensible experience. The new empiricism does not
have such a genetic worry; its claim is apparently more humble, but
probably in fact more ambitious: it maintains that the procedures of
mathematics, both of discovery and of validation, are no different from
those of the natural sciences
For example, Hilary Putnam (1975) stated that the methods of
mathematics are quasi-empirical (and by quasi-empirical methods he
meant methods analogous to those of the physical sciences) and differ
from those of the physical sciences only in that the individual statements
that are generalised by induction, and that are used to test the theories are
themselves the result of computations, instead of being observational
statements.
The first to use the term quasi-empirical in our times in the Western
world was Imre Lakatos (in 1976); he took it from Euler, via George
Polya. Polya had the merit of unearthing the heuristic methods of Euler;1
Lakatos made them popular among philosophers.
Later the empiricists also discovered Gauss, who in his notes admitted
that he arrived at mathematical truths through systematic experimentations.
They could have appealed also to the authority of Gdel, who in the 1951
Gibbs Lecture at Brown University, said:2

1
In Polya 1954, pp. 17-22, there are long quotations from Eulers paper of 1756
Specimen de usu observationum in mathesi pura.
2
The context is the discussion of the second incompleteness theorem, which Gdel
concludes with the following alternative:
So the following disjunctive conclusion is inevitable: Either mathematics is
incompletable in this sense, that its evident axioms can never be comprised in a
finite rule, that is to say, the human mind (even within the realm of pure
mathematics) infinitely surpasses the powers of any finite machine, or else there
exist absolutely unsolvable diophantine problems of the type specified (where the
case that both terms of this disjunction are true is not excluded ).
Gdel claims that the second horn seems to imply that mathematical objects
and facts (or at least something in them) exists objectively and independently of
 Empiricism and Experimental Mathematics 93

[] one may conjecture the truth of a universal proposition (for example,

that I shall be able to verify a certain property for any integer given to me)
and at the same time conjecture that no general proof of this fact exists. It
is easy to imagine situations in which both these conjectures would be very
well founded. For the first half of it, this would, for example, be the case if
the proposition in question were some equation F(n) = G(n) of two
number-theoretic functions which could be verified up to very great
numbers n.
Moreover, exactly as in the natural sciences, this inductio per
enumerationem simplicem is by no means the only inductive method
conceivable in mathematics. I admit that every mathematician has an
inborn abhorrence to giving more than heuristic significance to such
inductive arguments. I think, however, that this is due to the prejudice that
mathematical objects somehow have no real existence.
If mathematics describes an objective world just like physics, there is
no reason why inductive methods should not be applied in mathematics
just as they are in physics. The fact is that in mathematics we still have the
same attitude today that in former times was held toward all science;
namely, we try to derive everything by cogent proofs from the definitions
(that is, in ontological terminology, from the essences of things). Perhaps
this method, if it claims a monopoly, is as wrong in mathematics as it was
in physics.
This whole consideration incidentally shows that the philosophical
implications of the mathematical facts explained do not lie entirely on the
side of rationalistic or idealistic philosophy, but that in one respect they
3
favour the empiricist viewpoint.

Hence Gdel envisages the possibility that in the case of the realistic
option Platonism expounds the vision of mathematics as a natural science.
He did not pursue this idea, which would be consistent with his later
analysis of mathematical perception as an analogue of sense perception;
nor did empiricists follow his lead and profit from Gdels prestige.

our mental acts and decisions, that is to say, some form or other of Platonism, or
realism as to the mathematical objects. Then comes the quotation in the text
above.
3
To be more precise, it suggests that the situation in mathematics is not so very
different from that in the natural sciences. As to whether, in the final analysis,
apriorism or empiricism is correct is a different question.
94 Chapter Six

2. Experimental mathematics
At the end of the nineteenth century, Felix Klein classified mathematicians
as logicists, formalists and intuitionists (Klein 1893); at the end of the
twentieth century, a new category was added, that of experimental
mathematicians. They owe their existence and professional status to the
computer.
Experimental mathematics is defined

as the methodology of doing mathematics which includes using computers

to obtain intuitions and ideas, to discover new patterns and relations [] to
test conjectures and above all to falsify them, to work around a possible
result to see whether it is worth looking for a proof, to suggest approaches
to a formal proof, to substitute long derivations by hand with computer
derivations, to obtain analytical validations of numerical results (Borwein,
Bailey 2004).

Appeal to computers is not without contradictions, given that

machines have also inaugurated a particular mathematical field, that of
automated reasoning. The paradox lies in the fact that automated proofs
are the celebration of formal logic embedded in the machines. The
leitmotiv of humanistic philosophies of mathematics, by contrast, is the
attack on logic and proof, assumed as purported a priori knowledge.
To take in automated proofs in the empiricist perspective, Tymoczko
used an elegant argument; after the proof of the Four Colour Theorem he
proposed (in 1979; see also Lolli 1986) that we faced a new type of proof:
a traditional proof is an a priori deduction of a sentence from premises;
the 4CT proof is a proof with a gap, which must be filled with a well-
concocted experiment (thus introducing empirical elements to mathematics);
the theorem is the first mathematical proposition which is known a
posteriori, which means not that it is false or dubious but that it is known
in a very peculiar manner. According to Tymoczko this again raises the
problem of the relationship between mathematics and the natural sciences.
The meaning of proof must change if we no longer require that a
proof give us a priori knowledge. For Putnam (1975), too, the first
obstacle we have to overcome in order to refute the a priori character of
mathematical knowledge is the method of mathematical proof.
For social constructivism, proofs are at best conventional modes of
presenting mathematical results, and logic is simply an ideological
travesty of a social custom. For Lakatos they are only fallible confirmations.
Logic is not required for the proof of theorems; heuristic intuition has a
higher status as a tool for seeking truth.
 Empiricism and Experimental Mathematics 95

After presenting Cauchys proof of Eulers theorem on polyhedra (V +

E F = 2), based on stretching the polyhedron on a plane, Lakatos says
(erroneously) that there is no obvious way of transforming his argument
into a logical proof: what he has done is what he claims is often done in
mathematics, to have intuitively shown that the theorem is true, through
a mixture of rhetorical appeal and persuasive argumentation (Lakatos
1979).
Empiricism, by contrast, insists on inductive verification. According
to Putnam, if Riemanns Hypothesis were verified by computers in a huge
number of cases we would be justified in considering it verified and in
accepting it without reservations (actually this is already the case; as to
verification: how huge must huge be?).
At first sight, experimental mathematicians are unwilling to accept the
empiricist refusal of proof. They are aware of so-called high precision
frauds, namely identities which hold with a high degree of precision, but
are nonetheless not true. Jonathan Borwein is adamant: research does not
replace proof (Borwein. Bailey 2004, p. 7).
However, even mathematicians fall under the spell of the humanistic
fable; they agree with Lakatos that deductivism hides the research, the
struggle, the adventure; they recognise social influences and accept that in
any discipline strong social canons of validity and proof are at work
(Thurston 1994; see also Grabiner 1974).
Thus many mathematicians begin to feel impatient with the
constraints of the prevailing conception of proof (Borwein, Bailey 2004,
p. 245). The absorbing and fascinating activities made possible by the
computer begin to be sufficiently rewarding in themselves to justify the
time and effort devoted to them; someone prophesies that mathematicians
will become increasingly unconcerned with the search for certainty as a
consequence of their full immersion in computer explorations (see e.g.
Zeilberger 1993; and a rejoinder by Andrews 1994).
The opposition between absolute certainty and inductive validation
has become the kernel of discussions which entirely miss the point. The
function of proof is not to provide absolute certainty, as argued by Lolli
2005, while inductions by enumerations are by no means the basis of
natural sciences.
In Baker 2007 two case studies are considered: Goldbachs conjecture
is universally believed true apart from the admittedly rich inductive
evidence; the conjecture of the even perfect numbers has important
experimental support, but opinions are still divided. According to Baker,
empirical explorations are not done to accumulate positive cases but with
96 Chapter Six

other aims: to test programs for example, or to investigate related

questions (e.g. the partition function in the case of Goldbach).
Induction is not a scientific practice, if by induction we mean the mere
enumeration of favourable cases. To insist on explorations and inductions
in mathematics seeking analogies with the natural sciences does no
service to the latter; they are so rich and dependent on theoretical terms.
In the end, empiricists appear to have been insufficiently bold, or too
biased by their philosophical beliefs. Mathematicians are aware that the
computer offers them much more than numerical research: the computer
creates new worlds which would have been inaccessible to the human
imagination, for example hyperbolic spaces. This comes at a price: to
experience a world and to understand it are two different acts; in this
sense, according to Borwein and Bailey (2004), mathematics is becoming
experimental.
The problem is to understand what this means for the mathematics of
the future, if this many-sided experimenting is really a break with the
past. The limitation of the empiricist vision is that it restricts the focus to
the method of validation alone, thus ending in the stale antithesis between
deduction and induction, between logic and experiment.
Even Cellucci, although he is not an empiricist, suffers from this type
of limitation; in many writings (e.g. 1998) he contrasts the method of
logic with different methods, in particular the method of analysis, or that
of open systems. The discussion of the method is inevitably shallow;
knowledge is not supplied by the reasoning method, but by the models of
reality one builds, and the concepts one uses. I am sure Cellucci would
agree.
One should look elsewhere to find the experimental character of
mathematics. Mathematics is an investigation of the world (once only
physical, nowadays also human). Its concepts make it possible to model
phenomena and at the same time to develop techniques for the formal
manipulation of the concepts themselves. To try to separate mathematics
proper from physics or from other disciplines is an impossible task, and is
in any case useless or harmful. It is needless to recall that in all the
periods of the greatest breakthroughs, from Archimedes to Newton to the
golden age of infinitesimal calculus, there was no difference between the
mathematician and the physicist, between mathematics and mathematical
physics.
Mathematics is trying to recover, also in its image, the link with the
world which was the source of its growth, although it has never been
wholly severed. The period of introspection, due to the need to
systematise the many new concepts emerging from this growth, goes
 Empiricism and Experimental Mathematics 97

roughly from Cauchy to Bourbaki; it now appears to have ended. There

are recent examples of physical and cosmological investigations which
have led to new profound mathematical theorems, as well as, thanks to the
computer, examples of mathematical worlds which would have remained
inaccessible to the human imagination. Michael Atiyah has said, with
reference to Ed Witten, that in his hands physics has again become a
source of inspiration and understanding for mathematics.
The natural sciences are characterised by the concepts they use, not by
their logic, which is always natural logic. A mechanic reasons in terms of
forces, energy, potential, equilibrium and so on. These are not concepts of
nave physics, but concepts of mathematical physics, where mathematics
could in no way be separated from an experimental counterpart. Pure
mathematics when dealing with these concepts tends to bring them back
to their definitions, thus loosening the powerful inferential steps which
are embedded in well-known results, such as e.g. conservation principles.
Through this reduction one loses what is called physical intuition, which
however is in no way intuitive, but is none other than basic physical
knowledge. Such are the reflections one should pursue, from which
empiricism is totally alien in its discussion of method.

3. A voice from the past

To find scholars who have discussed the relationship between
mathematics and natural science with an open mind, we must turn to a
different tradition from Anglo-Saxon academe, to a probably unexpected
place. We will comment on a 1932 essay by Pavel A. Florenskij with the
title Physics as maidservant of mathematics. Florenskij (or Florensky)
was a Russian mathematician and philosopher who was long forgotten for
political reasons in his country; his name has now been cleared and
arouses strong interest in Europe.4 In Appendix 2 we provide some
biographical data.
In Floresnkij 1932 one sees traces of the life experiences described in
the appendix. The essay opens with the statement that the object of this
essay is the experimental character of mathematics and the complaint
that the idea that mathematics is deeply rooted in experiment is not yet
common wisdom. This can be seen in the ceaseless efforts to purify

4
See the biography by A. Pyman, 2010, and Bettis essay, 2009. A few of
Florenskijs writings have been translated into Italian in Florenskij 1995 and
Florenskij 2007.
98 Chapter Six

mathematics, to free it from intuitions which may have accidentally

entered and are interpreted as psychologisms, bad habits or plain
blunders. Geometrical drawings are allowed, a cardboard or iron wire
model still swallowed, but electricity, gravity, magnetic sheets and so on
are unbearable, excessively physical.
According to Florenskij, the space concept is based essentially on
mechanical experiments, hence logic too should learn the lesson:

If mechanical intuitions, albeit only a few, are at the basis of mathematics,

this fact opens the door to the entrance of machines. []
When [] a machine tailored to the characteristics of the solid body
becomes more complex, the champions of mathematical purity look at it
with suspicion, [] and judge it alien to mathematics [] If besides
kinematics dynamics is also involved, the majority of mathematicians see
in it a betrayal of pure thought. Common wisdom brushes aside []
machines like harmonic analysers and the many machines able to perform
the operations of analysis [] Even integrators and machines that integrate
differential equations are accepted with a certain distrust.

Florenskij exhaustively enumerates all the machines used at the time for
advanced computations, many of them now forgotten.5 He himself built
three machines in 1922, two for solving algebraic equations of higher
degree, and even transcendental ones, one hydrostatic and the other
electrostatic, and a third for the integration of arbitrary functions.
Florenskij, however, is not interested only in computations.

5
He mentions the slide rule (E. Gunter, 1623, improved by E. Wingate, 1627, and
Seth Partridge, 1657), Ritters machines to compute algebraic expressions with
square roots, Lalannes arithmetical scale, Exners computing machines to solve
algebraic equations up to the seventh degree the scales of C. V. Boys, G. B. Grant
and R. Skutsch (at the end of nineteenth century) for equations oh higher degree,
the wheels of E. Stamm (1863), M. Deprez (1871), F. Guarducci (1890), and A. B.
Kempes machine (1873) to solve trigonometric equations, L. Torress machine
(1895) for the real and complex solutions of algebraic equations and linear
systems, H. Wehages machine (1878), Varignon polygons, Lord Kelvins
machine (1878), with discs and integrating cylinders; he recalls A. Demaner who
in 1898 applied the principle of communicating vessels to solve third degree
equations, G. Meslin who in 1903 built a hydrostatic balance to solve algebraic
equations, A. Emch who in 1901 used the velocity of liquid flows to extract roots,
and Weltmans rocker arms (1884) for linear systems, and F. Lucass electric
solution of algebraic equations (1888).
 Empiricism and Experimental Mathematics 99

For us what matters is not only what the machine shows us, but how we
come to know it, and this how is not something external to the machine,
to the tool of knowledge, but a built-in character. When we draw a
circumference with a compass, we have to know whether the lead has
come back to the starting point, when drawing a segment whether the rule
lies on it or is detached and so on. Usually these requirements are assumed
to be satisfied by themselves, disregarding the cognitive acts involved,
hence the aspect of reality involved in such acts. Usually one decides that a
machine is needed, and it is assumed that all comes free. In other words,
mathematics is assigned the abstract metaphysical property of
omniscience, and of immediate knowledge [] the machine will do its
duty and mathematics will reflect on the result without any concrete and
vital tie to the object of its reflection. In this case, in fact, the necessary
intuitions would belong only to the kinematic realm and mathematics
would need no new intuitions.

According to Florenskij, this detached and aseptic use of mechanics is

mistaken, and even impossible, because the knowledge of what we do is
obtained through a series of physical factors that unfold in time and
space.6 The abstract metaphysical dowry of omniscience is an illusion.

In conclusion, mathematicians either have to appeal to telepathy or have to

be committed to a sort of indirect knowledge, and thereby legally introduce
into mathematics the intuitions always illegally used of the different
elements of nature and their distinctive features. But then the axiomatics of
mathematics has to be completely revised.

Even to think up a mathematical formula requires a construction. A

formula is the embodiment of abstract concepts in concrete equipment:
words, letters, signs: It is a construction, hence it requires an engineerings
hand.
To sum up, in mathematics we have to introduce the models and tools
of physics, possibly of chemistry, the aids of biology and of psychology.7

6
Among western empiricist mathematicians only Davis 1972 dared to express
similar ideas.
7
Florenskij sounds prophetic here. The chains of inferences can be formed by
stitching together independent shreds; exactly in the same way as in a broth of
oligonucleotides longer and longer chains are formed when strings are tied
together by the bond of complementary oligonucleotides. This is more than a
metaphor: in 1995 Leonard Adleman proved the existence of a Hamiltonian circuit
in a graph with a biochemical proof, using DNA polymerase. See Cipra 1996.
100 Chapter Six

Do the veins and annual lines of logs, which are a system of isopotential
force lines, not teach us anything? This burst of imagination is an
example of what Florenskij meant when he said: All the scientific ideas
that I cherish have always been aroused in me by the sense of mystery.

4. Physics as a servant of mathematics

Florenskij is not such a maverick as one might think; his ideas resonate in
contemporary Russian mathematics, unless both belong to a still older
tradition, which unfortunately we do not know; Russian mathematics has
peculiar features both in research and in education. One representative of
this school is Mark Levi who (in 2009) gathered a rich body of examples
to refute the common belief that mathematics is the servant of physics; in
almost the same words as Florenskij, in this book physics is put to work
for mathematics, proving to be a very efficient servant (p. 2). The title
reads: Using physical reasoning to solve problems.
Mark Levi studied in the Soviet Union in the seventies and he recalls
that already in high school he had met and absorbed this approach, mainly
from a book by Uspenski, translated into English as Uspenski 1961; he
also refers to Kogan 1974 and Balk and Boltyanskii 1987.
A new type of proof is added to those already known: the physical
proof. To solve a problem by physical reasoning, the first step is the
definition of a physical incarnation of the problem, thus reversing the
usual direction from physical problem to mathematical model. The
mathematical model is usually given by differential equations; in physical
terms, these can often be replaced by algebraic vector equations relative
to physical notions. Typically, a solution is given by the equilibrium of a
system, hence by the vanishing of the potential energy. Computations are
minimal or nil, pre-calculus often suffices, and the solution gives an
intuitive reason for the why (see the example in Appendix 1
Levi takes a minimalist position, according to which the physical
argument can be a tool of research and intuitive explanation, as
Archimedes himself to whom Levi obviously refers - officially
maintained. However the examples proposed promise more: even Levi
admits that in the translation into mathematical terms and the solution by
mathematical techniques something is lost. Mechanics, from which most
of the examples are drawn, is a basic attribute of our intellect, it is
geometry with the emphasis on motion and touch, which gives us an
extra dimension of perception (p. 4).
 Empiricism and Experimental Mathematics 101

Appendix 1
To give an idea of the arguments treated by M. Levi, we present his first
and simplest example (p. 6): wanted is a proof of the fact that, given three
points A, B, C in a plane, point X of the plane such that the sum of the
distances XA + XB + XC is minimal is that for which the three angles
AXB, AXC, BXC are equal, and each is equal to 120; one ties three
strings together calling the node X, then one slips each string through one
of the holes drilled at points A, B, C of a fictitious table, hanging equals
weights (by convention, 1) under the table; the potential energy of the
first string is then XA, since to drag X from A to its position one has to
raise the unit weight by distance XA; similarly for B and C.

Then the sum of the distances XA + XB + XC is the potential energy of

the system; if it is minimal, the system is in equilibrium; the three forces
of tension acting on X add up to zero, hence as vectors they form a
triangle if placed head-to-tail; the triangle is equilateral since the weights
are equal, hence the angles are 120.
The mathematical solution requires minimising the sum

S(x) = |x a| + |x b| + |x c|

putting to zero the gradient S = (wS/wx, wS/wy) of S.

The derivative w |x a|/wx is (x a)/|x-a|, and similarly for b and c,
and the gradient is the sum of the unitary vectors ea + eb + ec which form
an equilateral triangle.
102 Chapter Six

Appendix 2
Pavel Aleksandrovic Florenskij (1882-1937) was born in what is now
Azerbaijan, graduated in mathematics in 1904, in Moscow, with a
dissertation written under the supervision of Nikolaj Vasil'evic Bugaev
(1837-1903). Bugaev was the founder of the Moscow Mathematical
Society, which was a sort of mathematical-philosophical school whose
best students included D. F. Egorov and N. N. Luzin.
Bugaev had been impressed, in the development of analysis in the
second half of the nineteenth century, by the emergence and prominence
of discontinous functions, which are in fact the majority of real functions.
From a philosophical point of view, Bugaev and his followers saw the
evolution of the universe as a constant struggle of logos against chaos,
and assumed that mathematics was necessary for the development of a
general conception of the world. Florenskij was strongly influenced by
and contributed to this vision with a principle of discontinuity, by which
he found a justification for a conception of numbers as forms, which in
the continuity of change would not be possible
Besides mathematics, Florenskij had many other interests, philosophical,
theological, religious and artistic. Even in his papers devoted to art
criticism however, such as Florenskij 1995, one perceives his deep
geometrical culture, as shown by Betti 2009: the conception of the world
is the conception of the space, and the spiritual character of each epoch
determines the idea of space and its representation; the space of the arts is
not the Euclidean and Kantian space of classical physics, but a variable
curvature space which is modelled around the sensitivity and the interests
of the artist.
At the heart of his various interests Florenskij always held to a
mathematical conception of the world; in 1900 he wrote to his father that
in mathematics the philosophy of nature knits together with ethics and
aesthetics; even religion finds its proper place in the overall picture (Betti
2009).
After studying Cantors set theory, Florenskij adopted the theory of
the transfinite as a symbol of the ontological and logical relation between
the world of the absolute and that of the relative: human beings are
carriers of the transfinite, not the finite as opposed to the infinite divinity.
When he built a model of the complex plane, in the essay The
imaginaries and geometry (1920, partially translated in Florenskij 2007,
pp. 278-89), he conceived of two worlds connected by a frontier, each
able to leave a trace of itself in the other: it is not one plane, as in the
Argand-Gauss model, but it is made up of two planes, the first the real
 Empiricism and Experimental Mathematics 103

one and the other that of pure imaginaries; in between are the point with
complex coordinates a + ib (in the same essay he describes the space of
Dantes Comedy as an elliptic geometry).
The mathematical models Florenskij used in philosophy were not to
him mere analogies or similes; they bring out substantial affinities.
Mathematics is the preferred tool for knowledge because it has to do with
the necessary structure of thought, which corresponds to ontological
structures.
After graduating, instead of embarking on an academic career Florenskij
dedicated himself to theology, and became an orthodox priest. He
continued however to cultivate science, also for practical reasons. After
the revolution he survived by doing different jobs for which he was suited
thanks to his scientific training, and these new experiences gave him a
deeper appreciation of scientific knowledge, much more mature and
serious than his early mystical philosophy. Florenskij among other things
had a job in a plastic factory, taught in the Arts and Technics
Laboratories, was the manager of the Electrotechnical Institute K. A.
Krug, contributed to the electrification venture of the USSR and studied
electro-insulating materials; in Siberia, where he had been transported, his
researches concerned anti-freeze liquids, permafrost and electronics. His
last essay, commented on in the text, bears the mark of these scientific
and professional experiences.

References
Andrews G. E. (1994), The Death of Proof? Semi-Rigorous Mathematics?
Youve got to Be Kidding, The Mathematical Intelligencer, 1994, vol.
16, n. 4: 8-16.
Baker A. (2007), Is there a problem of induction for mathematics?, in
Leng, Paseau e Potter 2007: 59-73.
Balk M. B., V. G. Boltyanskii (1987), Geometriya mass (in Russian),
Bibliotechka Kvant, 61, Nauka, Moscow, 1987.
Betti R. (2009), La matematica come abitudine del pensiero. Le idee
scientifiche di Pavel Florenskij, Centro Pristem Eleusi, Universit
Bocconi, Milano, 2009.
Bloor D. (1976), Knowledge and Social Imagery, The Univ. of Chicago
Press, Chicago, 1976, 19912.
Borwein J. M., Bailey D. (2004), Mathematics by Experiment. Plausible
Reasoning in the 21th Century, A K Peters, Natick, MA, 2004.
Cellucci C. (1998), Le ragioni della logica, Laterza, Roma-Bari, 1998.
104 Chapter Six

Cipra B. (1996), Computer Science Discovers DNA, in P. Zorn (ed.),

What's Happening in the Mathematical Sciences, vol. 3, AMS,
Providence, R. I., 1996: 27-37.
Davis Ph. J. (1972), Fidelity in Mathematical Discourse: Is One and One
Really Two?, Amer. Math. Monthly, 1972, vol. 79, n. 3: 252-63;
reprinted in Tymoczo 1998: 163-75.
Davis Ph. J., R. Hersh (1980), The Mathematical Experience, Birkuser,
Basel, 1980.
Ernest P. (1998), Social Constructivism as a Philosophy of Mathematics,
New York, NY State Univ. Press, 1998.
Florenskij P. A. (1932), Fizika na sluzbe matematiki, Socialisticevskaja
rekonstrukcija i razvitie, 4 (1932): 43-63; partial Italian translation in
Florenskij 2007: 290-9.
. (1995), Lo spazio e il tempo nellarte (N. Misler ed.), Adelphi, Milano,
1995.
. (2007), Il simbolo e la forma, Bollati Boringhieri, Torino, 2007.
Gdel K. (1951), Some basic theorems on the foundations of mathematics
and their implications, Collected Works, Vol. III, Oxford Univ. Press,
New York, 1995: 304-23.
Grabiner J. V, (1974), Is Mathematical Truth Time-Dependent?, Amer.
Math. Monthly, 1974, vol. 81, n. 4: 354-65; reprinted in Tymoczko
1998: 201-13.
Hersh R. (1979), Some Proposals for Reviving the Philosophy of
Mathematics, in Advances in Mathematics, 1979, vol. 31: 31-50;
reprinted in Tymoczko 1998: 9-28.
. (1997), What is Mathematics, Really, Oxford Univ. Press, Oxford,
1997.
Kitcher P. (1984), The Nature of Mathematical Knowledge, Oxford Univ.
Press, New York, 1984.
Klein F. (1893), Lectures on Mathematics (1893), AMS, Providence,
2000.
Kogan B. Yu. (1974), The Applications of Mechanics to Geometry, Univ.
of Chicago Press, Chicago, 1974.
Lakatos I. (1976), Proofs and Refutations, Cambridge Univ. Press,
Cambridge, 1976.
. (1979), What does a Mathematical Proof Prove?, Mathematics,
Science and Epistemology, Cambridge Univ. Press, Cambridge, 1979:
540-51; reprinted in Tymoczko 1998: 153-62.
Leng M., A. Paseau, M. Potter (2007), Mathematical Knowledge, Oxford
Univ. Press, Oxford, 2007.
 Empiricism and Experimental Mathematics 105

Levi M. (2009), The Mathematical Mechanic. Using physical reasoning to

solve problems, Princeton Univ. Press, Princeton, 2009.
Lolli G. (1986), La Macchina e le dimostrazioni, Il Mulino, Bologna,
1986.
. (2005), QED. Fenomenologia della dimostrazione, Bollati Boringhieri,
Torino, 2005.
. (2006), Mathematics as a Human Endeavour, SNS Pisa, 2006, in
http://homepage.sns.it/lolli/articoli/Pisa.pdf
Polya G. (1954), Mathematics and Plausible Reasoning, vol. 1: Induction
and Analogy in Mathematics, Princeton, Princeton Univ. Press, 1954.
Putnam H. (1975), What is Mathematical Truth, in Philosophical Papers, 2
voll., Cambridge Univ. Press, Cambridge, 1975; reprinted in
Tymoczko 1998: 49-65.
Pyman A. (2010), Pavel Florensky: A Quiet Genius, Companion, N.Y.,
2010.
Thurston W. P. (1994), On Proof and Progress in Mathematics, Bulletin
AMS, 1994, vol. 30, n. 2: 161-77.
Tymoczko Th. (1979), The Four-Color Problem and Its Philosophical
Significance, The Journal of Philosophy, vol. 76, 1979: 57-83;
reprinted in Tymoczo 1998: 243-66.
. (1998) (ed.), New Directions in the Philosophy of Mathematics,
Princeton Univ. Press, Princeton, 1998.
Uspenski V. A. (1961), Some Applications of Mechanics to Mathematics,
Pergamon Press, New York, 1961.
Wilder R. L. (1981), Mathematics as a Cultural System, Pergamon Press,
Oxford, 1981.
Zeilberger D. (1993), Theorems for a Price: Tomorrows Semi-rigorous
Mathematical Culture, Notices AMS, 1993, vol. 40: 978- 81.
 CHAPTER SEVEN

IS TRUTH A CHIMERA?

CESARE COZZO

SUMMARY In his book Perch ancora la filosofia Carlo Cellucci argues

that truth does not play any role in (modern) science: truth is only a
chimera that prevents us from adequately understanding the character of
knowledge and therefore must be disposed of. I summarize Celluccis
evidence for his contention that truth is a chimera. I then raise four
objections to Celluccis views on truth. My conclusion is that, Celluccis
arguments notwithstanding, a notion of truth is necessary for the human
activity of problem solving and therefore for an adequate understanding of
the phenomenon of knowledge.

KEYWORDS: truth, problems, hypotheses, assertions, intellectual virtues.

1. The thesis that truth is a chimera

Carlo Cellucci proposes a heuristic conception of philosophy. In his book
Perch ancora la filosofia he states that philosophy can still be fruitful
only if it is an inquiry into the world (Cellucci, 2008, p.10). Philosophy,
or rather good philosophy, is or ought to be an activity which aims above
all at knowledge a knowledge that differs from scientific knowledge in
no essential respect, and is not restricted to any field (ibidem). There is
one difference, however, between philosophy and the sciences:
philosophy deals with questions that are beyond the sciences of the
present that is, questions the latter are unable to handle and tackles
them, trying out unexplored routes. In so doing, when successful, it may
give rise to new sciences (ibidem). This is the heuristic view,
according to which the aim of philosophy is to seek new knowledge and
new procedures for discovery (p. 34).
Since an essential ingredient of the heuristic view is the thesis that
philosophy aims at knowledge, one might think that an obvious
108 Chapter Seven

consequence is that philosophy aims at truth, and that the latter thesis
should also be part of the heuristic view. Yet this inference would be
wrong. Cellucci claims that truth is a chimera: throughout its long
history, philosophy has obstinately, though fruitlessly, chased after some
chimeras that is, fantasies or illusions concerning knowledge, which
have led philosophy into directions that have prevented it from adequately
understanding the character of knowledge. Such chimeras must be
disposed of if we wish to develop a fruitful philosophy (pp. 77-8). The
list of chimeras to be banished includes many illustrious philosophical
notions: truth, objectivity, certainty, intuition, deduction, rigour, mind
(p.78). This paper deals with the first item on the list. Why does Cellucci
think that truth is a chimera, a fantasy or an illusion? Are the reasons he
provides for his claim convincing?
Cellucci begins by arguing that some attempts at characterizing the
concept of truth are inadequate. The first target of his critique is the
correspondence theory of truth, which is commonly attributed to Aristotle.
Celluccis interpretation of Aristotle runs contrary to this widespread
attribution. In his view (p. 79), Aristotle thinks that the concept of truth as
correspondence should be discarded. Perhaps the main objection to the
correspondence theory that Cellucci sees in Aristotle and which can later
be found in new guises in Kant, Frege and other philosophers (pp. 85-6)
can be summed up in a single remark: correspondentists fail to specify a
genuine relation between truth-bearers and reality. The correspondence
theory is based on the idea that truth consists in a relation of
correspondence between thoughts (or other truth-bearers) and an objective
reality that is independent of our thoughts. But the idea that such a relation
exists the critic argues is an illusion: if we endeavour to specify the
nature of the relation of correspondence, we end up with a relation
between a thought and another thought, a representation and another
representation, or a sentence and another sentence. Cellucci (pp. 85-6)
endorses this objection and concludes that the concept of truth as
correspondence cannot be applied to modern science. Indeed he seems to
suggest that this concept is essentially misconceived and cannot be applied
to any item of knowledge.
The theory of truth that Cellucci ascribes to Aristotle is a theory of
truth as intuition of the essence. This theory of truth is adequate for
Aristotelian essentialist epistemology. According to an essentialist
epistemology, science discovers the real essences of things. But
essentialism was abandoned in the seventeenth century. For Cellucci,
Galileos scientific revolution is a move from an essentialist to a non-
essentialist epistemology. In the footsteps of Galileo, modern scientists
 Is Truth a Chimera? 109

renounce all claims to knowledge of the essences and limit the scope of
science to some properties that can be dealt with mathematically. A
straightforward consequence is that the theory of truth as intuition of the
essence cannot be applied to modern science. One might ask, however,
if another concept of truth could be applied to modern science. The answer
is in the negative, because all the alternative concepts that have been
proposed are inadequate (p. 86). To support the latter claim Cellucci
examines the weak points of the concept of truth as consistency (David
Hilbert), as systematic coherence (Harold Henry Joachim), and as
provability (Dag Prawitz). Cellucci concludes that all alternative concepts
of truth that have been proposed are inadequate: this implies that no
known concept of truth applies to modern science. In short: philosophers
have devised various concepts of truth, but whichever philosophical
concept of truth we choose, the thesis that truth is a characteristic feature
of scientific knowledge always turns out to be wrong, though for different
reasons. In this sense truth is an illusion.
Illusions, however, can be useful, and sometimes necessary. We might
grant that truth is an illusion and that science does not deliver truths, whilst
still admitting that such an illusion is necessary for science, perhaps
because it plays the role of a regulative idea. Cellucci rejects this. He
claims that truth does not play any role in (modern) science. Truth is only
a chimera that prevents us from adequately understanding the character
of knowledge and therefore must be disposed of. The first step towards
this conclusion is the claim that no known concept of truth applies to
modern science. The second step is the remark that since, despite this,
science has developed, its development did not depend on reference to any
concept of truth (p. 89). Therefore, a philosophy in accordance with the
heuristic conception must abandon the notion of truth.
An analytical epistemologist might object: x knows that p implies p
is true; therefore, if philosophy aims at knowledge, it aims at truth as
well. But Cellucci rejects the connection between knowledge and truth:
knowledge does not aim at truth, it aims at plausibility (p. 177).
Plausibility is a key notion in the heuristic view. Knowledge consists of
plausible hypotheses. A hypothesis is plausible if, and only if, it is
compatible with the existing data (p. 177). By compatibility with the
existing data Cellucci means that if we compare the arguments for and
against the hypothesis based on the existing data, the arguments in favour
of the hypothesis outweigh the arguments against it (pp. 177-8). The
heuristic view is centred upon the problem of knowledge, but the task of a
philosophical investigation concerning knowledge according to the
heuristic view is entirely different from the task that most analytical
110 Chapter Seven

epistemologists undertake. Most analytical epistemologists believe the

primary task of epistemology to be an analysis of the concept of
knowledge aimed at answering the question what is knowledge?.
According to Cellucci this question should be discarded: the question
what is knowledge? should be replaced [...] by the question what is the
role of knowledge in nature? (p. 74). His answer is that the role of
knowledge in nature is to solve problems: above all, knowledge [...]
serves to solve the problem of survival (p. 193) of the individual
organism and of the species (p. 199). But the problem of survival is only
a particular instance of the general fact that all knowledge is the solution
to problems, from the basic problem of survival to other less basic
problems (p. 213). In conclusion, the aim of philosophy is to seek new
knowledge, but for Cellucci this does not imply that philosophy seeks
truth, because knowledge does not imply truth. The concept of truth is a
remnant of Aristotelian science (p. 90), the vestige of an obsolete
philosophy.

2. My objections in short
I have tried to summarize Celluccis evidence for his contention that truth
is a chimera. Perhaps I failed to notice some important ingredient of his
arguments. If so, I hope that Cellucci will reply and that his reply will
highlight the crucial ingredient. But if my summary is correct and nothing
important has escaped me, then I do not think that the arguments provided
constitute convincing grounds for disposing of the notion of truth.
Moreover, it seems to me that the tenet that truth is a chimera is not
essential to Celluccis heuristic conception of philosophy and knowledge.
Indeed, I would say that, Celluccis arguments notwithstanding, a notion
of truth is necessary for the human activity of problem solving and
therefore for an adequate understanding of the phenomenon of knowledge.
More specifically, I have four objections to Celluccis views on truth: 1)
the concept of truth was not invented by philosophers: a basic notion of
truth plays a key role in everyday life, for non-philosophers and non-
scientists as well; 2) the role played by the basic notion of truth in
everyday life is inextricably intertwined with the practice of seeking
knowledge and solving problems; 3) there are many statements that are
true, therefore this notion is not an empty concept; 4) many of the
statements to which the basic notion of truth applies are statements made
by the modern sciences.
 Is Truth a Chimera? 111

3. The problem of truth

The details of Celluccis criticisms of the traditional philosophical theories
of truth (correspondence, coherence, provability, etc.) are in my opinion
not always cogent. But my concerns about these details are of little
importance, because I agree with Celluccis main point that the best
known philosophical attempts at analysing the notion of truth are
inadequate in some way or another. Philosophers have tried to provide a
definition or a reductive analysis of the notion of truth. A reductive
analysis is an answer to the question what is truth? framed as an
equivalence of the form: z is true if, and only if, z has property X, where
property X is characterized in terms of independent concepts that are
already clear and do not presuppose the notion of truth (in order to avoid
circularity). Correspondence, coherence, provability have been proposed
as key-concepts for specifying property X. Difficulties of various kinds
beset not only these proposals, but also other attempts at providing a
definition or a reductive analysis of truth. It is perhaps reasonable to desist
from these attempts. Granted that Cellucci is right on this point, however,
it does not follow that we must abandon the concept of truth. The failure
of these philosophical attempts only suggests that reductive analysis or
definition, one of the possible strategies for solving the problem of truth, is
unpromising. But if you try one way and reach a dead end, it is still
possible to try other ways.
The problem of truth concerns a notion which we all possess,
philosophers and non-philosophers alike. That all speakers possess a
notion of truth is evident from the existence of corresponding words in all
known natural languages and from many practices and judgments
indicating that the notion expressed by these words is deemed important
and valuable, albeit controversial. Controversies also arise because we
realize that we do not fully understand our notion of truth. We thus feel in
need of clarification. Can we gain a better understanding of the notion of
truth that we already possess? This is the problem of truth. A reductive
analysis or a definition might be one way of satisfying the need for
clarification, a way of solving the problem. The reductive strategy fails.
But it is wrong to conclude that the problem of truth is unsolvable from
the mere fact that one specific solution strategy fails. Cellucci should agree
that a reductive strategy is not the only way of dealing with the problem of
truth, because his own approach to the problem of knowledge is based on a
similar idea. As we have seen, he rejects the reductive strategy of
analytical epistemologists who endeavour to analyse the concept of
knowledge by answering the question what is knowledge?. Cellucci
112 Chapter Seven

favours an alternative philosophical strategy aimed at investigating the

role of knowledge: the question what is knowledge? according to
Cellucci should be replaced by the question what is the role of
knowledge?. Similarly, philosophers concerned with the problem of truth
could say that the question what is truth? should be replaced by the
question: what is the role of truth in our life? and try to answer this
question.
The notion of truth is not a philosophical construct. Parmenides, Plato
and Aristotle began to reflect upon a notion, aletheia, which already
appears in Homers Iliad and Odyssey and was used in ancient Greek
before it became the subject of philosophical investigation. The ubiquitous
usage of words like true and truth in English and other languages is
part of everyday life and the linguistic records left by many past
civilizations indicate that in those past civilizations, too, corresponding
words were used. Our philosophical task is to investigate the role of the
notion expressed by these words. The investigation of the role of truth will
show that the notion is important, although no reductive analysis or
definition is available. We can start by pinpointing certain basic facts.

4. The equivalence property

A significant fact concerning the role of phrases like ...is true or it is
true that ... in everyday linguistic usage is that they have the following
property: the result of applying these phrases to a sentence p is treated as
equivalent to p. Let us call this property the equivalence property. Paul
Horwich and other deflationists about truth think that the equivalence
property is all we need to know about truth and thus that the concept of
truth is superficial and trivially explainable (Horwich 2010, p. 4). My
opinion, by contrast, is that the equivalence property, together with other
facts about the role of truth, reveals the profound and pervasive
importance of this notion. The equivalence property in particular has
consequences for the way in which we understand the nature of some
problems.
Consider the following situation. Yesterday my friends and I roamed
the streets of Rome. We drove, we walked and went to some pubs in
different parts of the city. It was rather late when I parked my car. I was
sleepy and tipsy. After I parked, we moved to other districts in someone
elses car. This morning I cannot remember where I parked my car. The
problem is: Did I park my car in Via Bellini?. On its solution depends
whether I must go to Via Bellini to recover my car, rather than to Via dei
Gracchi or Piazza Dante or somewhere else. Location-problems of this
 Is Truth a Chimera? 113

kind often occur in everyday life: we wonder whether we left the keys in
the other jacket or whether the book was in the bag. Now consider the
equivalence property. In virtue of the equivalence property we can
reformulate the question concerning my car as the question: Is it true that
I parked my car in Via Bellini?. This is the problem. This is what I am
interested in knowing. Whether it is true that I parked in Via Bellini. If I
want to avoid the word true, I can say: I am interested in knowing
whether I parked in Via Bellini. But the problem is one and the same.
Admittedly, the best (albeit fallible) way to solve the problem is to
investigate whether or not the hypothesis that I parked there is plausible.
But I am interested in plausibility only to the extent that it is an epistemic
means of establishing the truth (albeit fallibly).
So this is my first point: we can make the practice of problem solving
intelligible to ourselves only if we understand what the problem that we
want to solve is. The problem of primary importance to me is not whether
it is plausible that I parked in Via Bellini, but whether it is true. An answer
that is merely plausible may fail to be a solution and the real solution may
fail to be plausible. Consider the following continuation of the story. My
friends and I try to reconstruct our lively and eventful night by
reassembling the pieces of fragmented memories. Then we compare the
evidence for and against the hypothesis that I parked in Via Bellini. The
evidence in favour prevails. Yet, alas, I did not park in Via Bellini. This
scenario illustrates the possibility of a hypothesis being compatible with
the existing data without being true. Plausibility does not imply truth. On
the other hand, another continuation of the story is as follows: we decide
to go to Via Bellini and find no car. The plausibility of the hypothesis I
parked in Via Bellini is reduced to a minimum. Now it is very
implausible that I parked in Via Bellini, since the car is not there.
Nevertheless, it may still be true: perhaps the car is not there now because
it was stolen during the night. Truth does not imply plausibility. In any
case, if I am interested in recovering my car, what matters is which of the
two statements I parked in Via Bellini or I did not park in Via Bellini
is true, not which of the two is plausible. In virtue of the equivalence
property, the following two sentences are equivalent.

1) Cesare parked in Via Bellini.

2) It is true that Cesare parked in Via Bellini.

But (1) and (2) are not equivalent to

3) It is plausible that Cesare parked in Via Bellini.

114 Chapter Seven

Thus the location-problem I described is a problem concerning truth, and

not simply a problem regarding plausibility. Obviously, location-problems
are only one example. We could give a great variety of other examples
from everyday life or from the sciences. Medical problems: Does the
patient have a kidney stone?. What the physician and the patient want to
know is whether it is true that the patient has a kidney stone, not simply
whether it is plausible. Scientists working for NASA want to know
whether the statement There is life on Mars is true, not only if it is
plausible. In September 2011 the Opera team at the Gran Sasso
underground laboratory wanted to know whether it is true that neutrinos
are faster than light, though they were already well aware that it was not
plausible. There is a very large and important class of problems that
consists of questions regarding the truth of a corresponding statement. A
statement is a meaningful declarative sentence p in a given context of
utterance. Let us term problems concerning the truth of a statement
propositional problems. The solution to a propositional problem
specified by p can be found in many ways, but there are two kinds of
solutions: a positive solution to the effect that p is true, or a negative
solution to the effect that p is not true. My first point is that we cannot
fully understand the activity of problem solving, at least as far as
propositional problems are concerned, without resorting to the notion of
truth.
We try to solve propositional problems by weighing the pros and cons
of rival hypotheses, that is by establishing their plausibility. Plausibility is
our criterion for choosing a solution. However, in our inquiry we are
always aware that we could be in one of three kinds of epistemic
situations. In some epistemic situations we succeed in choosing a
hypothesis: the hypothesis that turns out to be the most plausible and will
remain such. In other epistemic situations we select a hypothesis as the
most plausible, but though we are provisionally satisfied, the future course
of inquiry will show that the hypothesis is mistaken. In yet other epistemic
situations the available arguments are insufficient, and we are unable to
select one particular hypothesis from among the many candidates. Thus
we remain in ignorance of a solution. Nevertheless, we do not rest content
with our present lack of knowledge and continue to seek a solution. The
force that drives us beyond the present epistemic situation to correct our
mistakes and repair our ignorance is our concern for truth. This aspect of
the role of truth can be summarized in a motto: true is that with respect to
which we can always be mistaken or ignorant.
 Is Truth a Chimera? 115

5. Problems, hypotheses and assertions

When we try to solve propositional problems, we start from hypotheses.
Hypotheses can be described as sets of interconnected sentences (sometimes
accompanied by diagrams or pictures). Their role is tentative: we hope
that, if accepted, they will lead to an answer to our question. They are
treated as candidates for a solution, not as if they were already a solution.
Making a hypothesis is a first tentative step towards a solution, but in most
cases many other steps are necessary before we can confidently state that
we have a solution. Different hypothetical reconstructions of our
movements last night lead to different answers to the question Did I park
my car in Via Bellini?. In order to be taken into consideration, a
hypothesis should be relevant and promising. But before concluding that
the hypothesis really provides a solution, we should scrutinize it and
compare it with other rival hypotheses. In the end, we sometimes pass
from the level of hope to the level of confidence: we become convinced
that our reasoning has led us to a satisfactory conclusion. Only then do we
select one of the hypotheses and put forward the solution resulting from it.
When we publicly advance a solution to a problem regarding the truth of a
statement, we make an assertion. We make an assertion to the effect that
the statement in question is true (the solution is positive) or that it is not
true (the solution is negative). My second point concerning the connection
between truth and problem solving is that to account for the practice of
problem solving we must understand the difference between these two
kinds of acts: making a hypothesis and making an assertion. I claim that in
order to understand the difference we must resort to the notion of truth.
We hope that a hypothesis will give us a solution. When we are
confident that we have the solution, we make an assertion. The different
attitudes that distinguish hypothesis and assertion are hope and confidence.
But confidence is not certainty. The difference between presenting a
sentence as a hypothesis and asserting it is not that the former act is
fallible and the latter is infallible, nor that the asserted sentence is true and
the hypothesis is not. Both acts are fallible, in both acts the sentences may
be true or untrue. The difference is that when we make an assertion we
attach a special force to the asserted sentence that we do not attach to a
hypothesis: we undertake a commitment. To undertake a commitment is
also to accept a rule: one ought to honour the commitment. By making an
assertion we implicitly submit to the rule prohibiting the assertion of
sentences that are not true in the context of utterance: if the asserted
sentence is not true, our act is objectively wrong. When we make an
assertion we undertake the commitment that the asserted sentence is true.
116 Chapter Seven

Subjectively, we may be unaware that the sentence we assert is in fact not

true and that we are thus failing to honour the truth-commitment. But if we
discover that the sentence is not true, our commitment binds us to
withdraw the assertion and admit that it was wrong. A commitment to
retraction or withdrawal derives from the basic commitment to the truth of
the asserted sentence. From this, we can derive a further two commitments:
a commitment to justification and a commitment to responsibility. If our
assertion is challenged, we are committed to provide a justification, i.e. an
argument supporting the truth of the asserted sentence. And we are
committed to being held responsible if somebody acts on the basis of the
sentence we asserted, but it turns out to have been untrue. Peirce (1902b,
C.P. 5.543, p. 384) wrote: to assert a proposition is to make oneself
responsible for its truth. When we make a hypothesis we do not
undertake any such commitment: it is only an interesting hypothesis! we
may reply if someone objects. If someone objected to our assertion, by
contrast, we would not say: it is only an assertion. Either we would
retract it, thereby admitting error, or we would say something amounting
to This is what I claim and I can demonstrate that your objection is
wrong. As a consequence, our act of assertion is more vulnerable to the
risk of error than the act of making a mere hypothesis (whose legitimacy
requires only that the hypothesis be interesting).
The difference between hypothesis and assertion manifests itself
through a different attitude towards incompatible alternatives. Tolerance
of incompatible alternatives is a characteristic feature of the level of
hypothesis. By making a hypothesis we open up a path of investigation. At
the start in the space of hypotheses different alternative paths of
investigation are often opened up. Making hypotheses involves a special
freedom and plurality, which can be very fruitful. A hypothesized sentence
ought to have certain properties: the hypothesis should be relevant,
promising etc. Obviously, we cannot entertain as a hypothesis a sentence
to whose falsity we are already committed. But there is nothing wrong
with considering an interesting hypothesis whilst at the same time not
having discarded other interesting rival hypotheses that are incompatible
with it. Some of the hypothesized sentences will fail to be true, though we
do not know which. But when we are at the level of hypotheses we are free
to legitimately entertain incompatible ideas and to treat each of them as a
candidate for the solution.
Freedom and plurality are nice. If we want a solution to the problem,
however, we must choose one of the alternatives. If the activity of problem
solving simply remained at the stage of making hypotheses, we would
never become convinced that a problem is solved. Are we always in such a
 Is Truth a Chimera? 117

state of suspended judgment? Far from it! We often become satisfied that
we have found the solution to our problem. We express this conviction in
language by moving to the level of assertions. Admittedly, we are aware of
the possibility of error. Sometimes it turns out that our belief that we
possessed a solution was mistaken. In this case the old problem is
reopened and we must go back to the stage of making hypotheses. But the
fallibility of our conviction does not change the fact that selecting a
solution and making our choice public through the speech act of assertion
is a different stage of our activity of problem solving. The act of
linguistically expressing our choice of a solution is an assertoric act. We
choose one of the hypotheses and assert the corresponding solution. This
act would become pointless if we also asserted a contrary potential
solution resulting from one of the other hypotheses. So the level of
assertion is characterized by an intolerance of incompatible alternatives.
We are free to make various incompatible hypotheses, but we are not free
to make incompatible assertions. We advance the selected sentence as the
solution and commit ourselves to its truth, which rules out the truth of a
rival potential solution.
If this is the right explanation of the difference between hypothesis and
assertion, it seems clear that a notion of truth is needed to make the
difference intelligible. We need a notion of truth in order to grasp the
nature of the commitment that characterizes the act of assertion: the
commitment that the asserted sentence is true. Truth plays a crucial role
and without this notion a significant part of our activity of problem solving
would not be possible. As a consequence, when Cellucci recommends that
we get rid of the notion of truth, what he is actually advocating is a
profound change to our epistemic and linguistic practices, a change whose
impact is difficult to imagine. This was not his intention. He wanted to
adequately investigate the character of human knowledge as it really is. He
did not want to replace human knowledge with something profoundly
different. Thus the exhortation to dispose of the notion of truth clashes
with Celluccis heuristic view. The thesis that truth is a chimera should be
rejected.
On Celluccis behalf it might be objected that when we select one of
the hypotheses that were candidates for a solution and finally present it as
the solution to our problem, we are simply choosing that hypothesis as the
most plausible: doing so does not require reference to the notion of truth.
This objection, however, neglects the equivalence property. Perhaps I
choose the hypothesis I parked in Via Bellini because it is the most
plausible alternative. But a positive solution to the problem Did I park my
car in Via Bellini? cannot be it is plausible that I parked in Via Bellini,
118 Chapter Seven

but only I parked in Via Bellini, which is equivalent to It is true that I

parked in Via Bellini. A negative solution cannot be it is not plausible
that I parked in Via Bellini, but only I did not park in Via Bellini,
which is equivalent to It is not true that I parked in Via Bellini.
Therefore, as I have argued above, though plausibility is a criterion for
selecting a solution, the linguistic expression of our choice of a solution
amounts to a claim that a statement is true or not true, not simply plausible
or implausible.
Two lines of thought converge in the above considerations. The first is
that a meaningful declarative sentence corresponds to a certain kind of
problem (a propositional problem) and that asserting the sentence
corresponds to the act of publicly proposing a positive solution to the
problem. The second line of thought is that a meaningful declarative
sentence p is equivalent to a sentence obtained by applying to p the
phrases ....is true or it is true that ... and that asserting p is the act of
committing oneself to the truth of p. The sources of the two lines of
thought are diverse. The idea that propositions correspond to problems
was familiar to ancient Greek geometers and was more recently proposed
by Kolmogorov (1932). The equivalence property had already been
pinpointed by Aristotle in Categories (14b 14-18, 1963, pp. 39-40), but it
was probably after Frege (1918) and Tarski (1935) that it became the
starting point for many philosophical investigations of truth. The idea that
assertion is an act by which we commit ourselves to the truth of the
asserted sentence was advanced by Peirce (1902b, C.P. 5.543, p. 384) and
is now the basic idea in a widespread conception of assertion (cf.
MacFarlane, 2011). The origins of these ideas are diverse, but they are
harmoniously combined in the above considerations, which indicate that
the notion of truth is a necessary ingredient of our practice of problem
solving.
The equivalence property of phrases like ...is true or it is true that
... is a first fact concerning the role of truth in our everyday use of
language; the connection between the notion of truth and our acts of
assertion is a second fact, which Crispin Wright (cf. 1992, pp. 15-17)
expresses in the motto that truth is a norm of assertoric practice. A
consequence of this connection is that the notion of truth underlies the
assertoric use of sentences even if those sentences do not contain the word
true. The truth-commitment of assertion manifests itself in the strongest
criticism that we can level at an act of assertion. We can criticize an
assertion in various ways. We can criticize it as insincere. We can criticize
it as implausible or unjustified. But even if at the time when an assertion is
made the assertion is sincere and plausible and justified, there is another
 Is Truth a Chimera? 119

crucial sense in which we agree that it can be wrong and can be criticized:
because it is not true. Suppose that Enrico sincerely asserts that I parked in
Via Bellini. He sincerely believes that he remembers that I parked there.
Since we lack any grounds for not thinking things to be as Enrico sincerely
says that he remembers them, we rightly consider his assertion justified
(though defeasible) and we deem it plausible that I parked in Via Bellini.
So the assertion was sincere, justified and plausible (it agreed with all the
available data). But it may later turn out that it was not true. If we now
discover that my car was parked elsewhere, we will say that Enricos
assertion was wrong and Enrico himself will withdraw it. Not only is the
assertion wrong now, it was also wrong in this crucial sense when it was
initially made, because the asserter made a commitment and failed to
honour it.
A critical reader will object that I am painting far too idealized a
picture of assertoric practice. Our assertions are only sometimes made on
the basis of a conviction that we have reached a satisfactory solution to the
corresponding problem. Very often we make an assertion only because we
intend to act, or to make someone else act, in a certain way. Enrico may
assert You parked in Via Bellini without any recollection of where I
parked last night and without having made any effort to solve the problem.
He may make this assertion simply because he knows that, if his speech
act is accepted, we will go to Via Bellini and because, for some reason, he
wishes us to go there. Speakers often make assertions without knowing the
truth of the asserted sentence and without even having tried to solve the
corresponding problem. Moreover, speakers are often unwilling to retract
an assertion they have made, even if their interlocutors raise plausible
objections; on the contrary, the asserters try to hide and play down any
evidence against the assertion and to persuade other speakers to accept the
assertion by all means, quite regardless of truth. In most cases the motive
for this conduct is that the asserter intends to influence the others actions.
It is indeed essential to assertoric practice that we speakers are aware
that we and our fellow speakers act on the accepted assertions (cf.
Dummett 1981, pp. 302, 355). An assertion linguistically expresses a
commitment to the existence of the solution to a propositional problem.
We learn the practice of assertion in certain basic circumstances. In a basic
assertoric circumstance the asserter is in a better position to solve the
propositional problem than the hearer. The asserter can thus convey the
information that the problem is solved to the hearer. Basic assertoric
circumstances also present themselves in the everyday practice of
competent speakers: Gino can remember that I parked in Via Bellini,
whereas I have forgotten; by asserting You parked in Via Bellini, he
120 Chapter Seven

solves the problem for me and gives me the solution. The hearer can take
advantage of the assertoric act, benefiting from a piece of information that
she or he would not otherwise have acquired. The hearer can thus orient
her or his actions by relying upon the speakers assertion. Assertion is a
key tool for the intersubjective coordination of actions through language.
It is a plain fact, though, that the tool of assertion is often misused.
Since all speakers are aware of its functioning, the asserter, without
genuine concern for the solution to the corresponding propositional
problem, can use assertion in order to make the hearer do and believe what
appears to be expedient. This instrumental use of assertion, however, is
intelligible, and thus possible for us, only against the background of a
basic tacit agreement among speakers that an asserter is committed to the
truth of the asserted sentence. Without such a background the hearer
would not have the basic reason to rely upon the asserters utterance, and
the utterance would have no influence on the hearers actions. Therefore,
though the instrumental use of assertions is very frequent and widespread,
it does not refute the connection between assertion and truth; in fact, it
confirms the tenet that there is such a connection: the view that assertion
involves a truth-commitment explains how the instrumental use is
possible. The speech act of assertion is governed by an implicit norm of
truth: in making assertions and deciding whether to accept assertions we
speakers are aware that we ought to seek truth, though we often do not
care.

6. Intellectual virtues
What is it to seek truth? It is a way of acting, a practice. To answer the
question what is it to seek truth? the best we can do is to describe the
practice of seeking the solutions to propositional problems. We may call
this practice inquiry. Inquiry demands certain behavioural traits or
dispositions that go by the name of epistemic virtues or intellectual
virtues. As Peirce wrote: in order to reason well [...] it is absolutely
necessary to possess [...] such virtues as intellectual honesty and sincerity
and a real love of truth (1902a, p.43, C.P. 2.82, cf. Haack 1996).
Plato already highlights some intellectual virtues in his polemic against
the sophists. For example, he insists on the willingness to take objections
seriously and to retract ones assertions, if these objections stand up to
honest counterarguments (cf. Gorgias 458a, in Plato 1997, p. 802). But the
concept of intellectual virtue is more explicitly formulated by Aristotle in
the Nicomachean Ethics (1139b). Today virtue epistemology is one of the
most influential philosophical approaches to the problem of knowledge (to
 Is Truth a Chimera? 121

get an idea the reader can look at Axtell 1997, 2000, Fairweather and
Zagzebski 2001, Roberts and Wood 2007, Baehr 2011). An incomplete list
of intellectual virtues includes curiosity, reflectiveness, attentiveness,
carefulness, sensitivity to detail, critical attitude, self-scrutiny, intellectual
honesty, fairness, flexibility, open-mindedness, intellectual non-
conformism, tenacity, intellectual courage. The list shows that intellectual
virtues are extremely diverse. Sometimes they come into conflict with one
another. A special wisdom (which is another virtue) may be necessary to
find the right dosage of virtues appropriate to a particular epistemic
context. But we are somehow aware that, in spite of their diversity, all
these behavioural traits contribute to a common goal. One can say that all
the intellectual virtues radiate (Code 1984, p. 34) from a central core: a
concern for truth. The intellectually virtuous agent seeks truths that are
relevant to the problem she, or he, confronts. The intellectual virtues are
ways of acting whose conscious motivation is the search for relevant
truths. This is one of the differences between intellectual virtues and
intellectual skills (cf. Zagzebski 1996, pp. 106-16). An intellectual skill,
such as the ability to think up analogies, is a capacity that can be used both
to hide truth and to discover truth. An analogy can serve to deceive an
interlocutor by deliberately misrepresenting reality or to solve a problem
through an enlightening connection. Unlike intellectual virtues, the
exercise of intellectual skills is not necessarily motivated by a search for
truth. But since intellectual virtues are motivated by our quest for truth,
intellectual virtues would not exist without some grasp of the notion of
truth. It is obvious that we are not always intellectually virtuous.
Intellectually virtuous acts often involve effort and sacrifice. Nevertheless,
rare as they may be, no one would deny that these ways of acting are
important features of our epistemic practices and a valuable ingredient of
life. Without a notion of truth, we would lack them.
Cellucci might say that intellectual virtues are not aimed at truth, that
they do not pursue truth, but plausibility. This suggestion is refuted by
many examples of intellectual courage. Perhaps the most famous is
Copernicus proposal of his heliocentric astronomic theory. This proposal
was an act of intellectual courage, not only for the well known religious
and political reasons, but also because in 1543 the Copernican system was
extremely implausible. Cellucci says that plausibility is compatibility
with the existing data, meaning that if we compare the evidence for and
against the hypothesis based on the existing data, the evidence in favour of
the hypothesis prevails over the evidence against it (cf. Cellucci 2008, pp.
177-8). But in 1543 the evidence against the Copernican system prevailed
over the evidence in favour of it. According to Herbert Butterfield (1965,
122 Chapter Seven

p. 41): if you grant Copernicus a certain advantage in respect of

geometrical simplicity, the sacrifice that had to be made for the sake of
this was nothing less than tremendous. You lost the whole cosmology
associated with Aristotelianism [...] you had to throw overboard the very
framework of existing science, and it was here that Copernicus clearly
failed to discover a satisfactory alternative. But Copernicus sought truth,
not plausibility.

7. True statements
How would human beings live without a notion of truth? In Cozzo 2012
the reader can find an attempt to describe imaginary beings who do not
possess the notion of truth, but are otherwise similar to us. I believe that
this difference with regard to truth would have a wealth of far-reaching
consequences. If the above considerations are correct, the notion of truth
plays a very significant role in our life. Without some grasp of the notion
of truth we might perhaps have cognitive faculties (vision, hearing,
memory, etc.) and some skills, but we would not have intellectual virtues,
nor assertions and propositional problems, and thus we would not have a
language like that we are familiar with. What remained of our practice of
problem solving would be extremely primitive. Therefore I think I have
provided grounds to support the first two objections to Celluccis view that
I have listed in section 2: the concept of truth plays a key role in everyday
life, for non-philosophers and non-scientists as well, and this role is
strictly intertwined with the practice of seeking knowledge.
The other two objections were that there are many statements which
are true and that many of these statements are statements made by the
modern sciences. Yes. I claim that there are many true statements
belonging to the sciences and to everyday life. I do not say that they are
certain or incontestable. We can imagine how we might criticize them. For
example we might find mistakes if we investigate how we came to accept
them. Or we might realize that they contradict other statements whose
truth is better established. Criticism of a statement always takes shape
against the background of many other statements accepted as true. None of
them is immune to criticism. Nevertheless, many of them are true. This is
not a wild claim. After all, I think Cellucci would agree. Cellucci knows
that many statements are true. It is true that a cube cannot be the sum of
two cubes. It is true that our sun is one star among billions in the Milky
Way alone. It is true that two hydrogen atoms combine with one oxygen
atom to form a water molecule. It is true that dogs wag their tails and
mosquitoes sting. It is true that Carlo Cellucci lives in Rome and wrote a
 Is Truth a Chimera? 123

book entitled Perch ancora la filosofia. I think he will not deny that these
statements are true. And, though he might deny it, it is true that I am one
of his grateful pupils.

References
Aristotle (1963). Categories and De Interpretatione: Translated with
Notes, ed. by J.L. Ackrill. Oxford: Oxford University Press.
. (2002). Nicomachean Ethics, translated by C. Rowe, philosophical
introduction and commentary by S. Broadie. Oxford: Oxford
University Press.
Axtell G. (1997). Recent Work on Virtue Epistemology. American
Philosophical Quarterly, Vol. 34, No.1: 1-26.
Axtell G. ed. (2000). Knowledge, Belief and Character. Lanham-Boulder-
New York-Oxford: Rowman & Littlefield.
Baehr J. (2011). The Inquiring Mind. Oxford: Oxford University Press.
Butterfield H. (1965) The Origins of Modern Science (revised edition).
New York: The Free Press.
Cellucci C. (2008). Perch ancora la filosofia. Roma-Bari: Laterza.
Code L. (1984). Toward a Responsibilist Epistemology. Philosophy and
Phenomenological Research. Vol. 45, No. 1: 29-50.
Cozzo C. (2012). Gulliver, Truth and Virtue. Topoi, vol. 31, No. 1: 59-66.
Dummett M. (1981). Frege. Philosophy of Language (second edition).
London: Duckworth.
Fairweather A. and Zagzebski L. eds. (2001). Virtue Epistemology: Essays
in Epistemic Virtue and Responsibility. Oxford: Oxford University
Press.
Frege G. (1918). Der Gedanke. Beitrge zur Philosophie des Deutschen
Idealismus, Band 1, Heft 2 : 58-77, rep. in Frege (1966).
. (1966). Logische Untersuchungen. Gttingen: Vandenhoeck &
Ruprecht.
Haack S. (1996) Concern for truth: What it Means, Why it Matters. In: P.
R. Gross, N. Levitt, and M. W. Lewis (eds), The Flight from Science
and Reason, New York: New York Academy of Sciences: pp 57-63.
Horwich P. (2010). Truth-Meaning-Reality. Oxford: Clarendon Press.
Kolmogorov A. N. (1932). Zur Deutung der intuitionistischen Logik.
Mathematische Zeitschrift, 35: 58-65.
MacFarlane J.(2011). What is Assertion? In: Brown J. and Cappelen H.
(eds), Assertion. Oxford: Oxford University Press: 79-96.
Peirce C. S. (1902a). Partial Synopsis of a Proposed Work in Logic (from
Minute Logic). In: Peirce (1960a).
124 Chapter Seven

. (1902b). Belief and Judgment (from Reasons Rules). In: Peirce

(1960b).
. (1960a). Collected Papers. Volume II, ed. by Hartshorne C. and Weiss
P. Cambridge, Mass.: Harvard University Press.
. (1960b). Collected Papers. Volume V, ed. by Hartshorne C. and Weiss
P. Cambridge, Mass.: Harvard University Press.
Plato (1997) Complete Works, ed. with an introduction by John M.
Cooper. Indianapolis/Cambridge: Hackett.
Roberts C. R. and Wood W. J. (2007). Intellectual Virtues. An Essay in
Regulative Epistemology. Oxford: Clarendon Press.
Tarski A. (1935). Der Wahrheitsbegriff in den formalisierten Sprachen,
Studia Philosophica 1: 261-405.
Wright C. (1992). Truth and Objectivity. Cambridge, Mass.: Harvard
University Press.
Zagzebski L. (1996) Virtues of the mind. Cambridge, UK: Cambridge
University Press.
 SECTION II

THE PROJECT FOR A LOGIC OF DISCOVERY

 CHAPTER EIGHT

TO ESTABLISH NEW MATHEMATICS,

WE USE OUR MENTAL MODELS AND BUILD
ON ESTABLISHED MATHEMATICS

REUBEN HERSH

SUMMARY Mathematicians proof doesnt start from a pre-ordained set of

axioms. It starts from relevant pieces of established mathematics out of
which some new mathematical result can be derived and incorporated into
established mathematics. Our reasoning is not syntactic but semantic. We
use our mental models of mathematical entities, which are culturally
controlled to be mutually congruent within the research community. These
socially controlled mental models provide the much-desired semantics
of mathematical reasoning.

KEYWORDS proof, mental models, semantics, mathematical practice,

warranted assertibility, truth.

1. Introduction
At last, in the 21st century, the maverick topic of mathematical practice
arrived as a legitimate theme of philosophical investigation. Carlo Cellucci
attended to mathematical practice, and challenged outworn philosophical
clichs. Paolo Mancosu edited a collection entitled The Philosophy of
Mathematical Practice. An Association for the Philosophy of
Mathematical Practice was organized, and published two issues of the
journal Erkenntnis.
But philosophical writing on mathematical practice still struggles to
get a grip. Before one philosophizes on mathematical practice, one might
wish to find out, What do real mathematicians really do? This article
reports on the mathematical practice of actual mathematicians. It focuses
on proof--the front side of mathematics. But in the course of our report,
128 Chapter Eight

we must also look at the back sideheuristics, or the analytic method

(Hersh, 1988; Cellucci, 2008).
I will quote from Andrew Wiles proof of Fermats Last Theorem
(FLT), but I start off with Vaughan Jones (of the Jones polynomial
renowned in knot theory) and Bill Thurston (who cracked open four-
dimensional topology, by grounding it in three-dimensional non-Euclidean
hyperbolic geometry). At a conference in Sicily in 1995 on Truth in
Mathematics, Jones told of the wonderful properties of the Fourier
transform, and listed a few of the many fields in mathematics and physics
where it is essential. He said, To doubt the truth of the Fourier
transform, however the word truth be interpreted, would be mathematical
lunacy.The mathematician is as certain of his faith in mathematics as he
is in the fact that a ball will drop if held above the ground and released
more sure than that the sun will rise the next morning. (Dales 203, 205)
Jones backed up his claims with interesting examples from braid theory.
At that same conference the illustrious number theorist and algebraic
geometer Yuri Manin (author of a beautiful textbook on logic) quoted the
following confessions of Bill Thurston:

When I started as a graduate student at Berkeley, I had trouble imagining

how I could prove a new and interesting mathematical theorem. I didnt
really understand what a proof was. By going to seminars, reading
papers, and talking to other graduate students, I gradually began to catch
on. Within any field, there are certain theorems and certain techniques that
are generally known and generally accepted. When you write a paper, you
refer to them without proof. You look at other papers in the field, and you
see what facts they quote without proof, and what they cite in their
bibliography. You learn from other people some idea of their proofs. Then
youre free to quote the same theorem and cite the same citations. (Dales,
152.)

Some philosophers may find these testimonies strange, but they will strike
mathematicians as commonplace. In order to explicate them, I will spell
out the concept of mathematicians proof: proof as it is understood by
mathematicians.

2. Wiles proof of FLT isnt an axiomatic proof,

its a Mathematicians Proof
In 1637, in the margin of a copy of Diophantus Arithmetica, Pierre de
Fermat wrote in Latin: It is impossible for a cube to be written as a sum
of two cubes or a fourth power to be written as the sum of two fourth
 Establishing New Mathematics 129

powers, or, in general, for any number which is a power greater than the
second to be written as a sum of two like powers. I have a truly marvelous
demonstration of this proposition which this margin is too narrow to
contain. This is Fermats last theorem (FLT). Fermats proof of it was
never found, but it was verified and checked for all powers of n up to
4,000,000. For over three and a half centuries, it was the most famous
open problem in mathematics. Around 1964, a young Japanese
mathematician, Goro Shimura, published a remarkable conjecture, which
became known as the Taniyama-Shimura conjecture (TS): Every elliptic
curve defined over the rational number field is uniformized by modular
functions. This very bold and startling conjecture has no apparent relation
to FLT. It asserts an intimate connection between two seemingly unrelated
subjects: algebraic geometry, where elliptic curves are a major topic, and
function theory, where the modular functions are central and classical. But
then, in the 1970s, the German number theorist Gerhard Frey found
reason to surmise that from TS one could prove FLT! An outline of a
possible proof of Freys surmise was offered in 1985 by the famous
mathematician Jean-Paul Serre. And in August of the following year, Ken
Ribet of Berkeley proved the lemmas in Serres proposal! Thus it was
established: FLT would indeed be proved, if only TS were proved. That
final step was made by a Princeton professor who had been obsessed with
FLT since the age of 10. Andrew Wiles realized that to obtain FLT, the
full Taniyama-Shimura conjecture is unnecessary, it is enough to prove TS
just for semistable elliptic curves. And that is what Andrew Wiles did,
with help from his student Richard Taylor. (See Mozzochi and Gowers)
Wiles paper uses a great variety of sophisticated mathematical ideas.
Nevertheless, an outsider can read the introduction, to glimpse what he did
and how he did it.1
In Wiles proof, as in all mathematical research, two aspects are
intricately intertwined: the heuristic or problem-solving, and the rigorous

1
Carlo Cellucci (2008) raised an interesting objection. It wasnt Wiles, he argued,
it was Ribet who proved FLT, by deriving FLT from TS. Wiles merely proved TS,
not FLT itself. Ken Ribet himself certainly never would claim any such credit, for
merely deriving FLT from an unproved conjecture. Yet, strangely enough, if the
order of events had been different, if Wiles had first proved TS, and Ribet had then
derived FLT from TS, it would indeed have been Ribet who had the glory of
proving FLT! That is how mathematicians assign credit. The first mountaineer to
stand atop Mount Everest gets the glory, even though everyone knows that his
climb was merely the last in a long sequence of preparations. Sad to say, criticism
from a philosopher will not change the way mathematicians assign credit.
130 Chapter Eight

or deductive. The researcher trying to find a rigorous proof uses plausible

or heuristic reasoning in order to find it, so the heuristic and rigorous
aspects of mathematical research are inseparable. (George Polyas deep
and entertaining accounts of mathematical heuristics (1925, 1945, 1954,
1980) unfortunately have been ignored by philosophers, because they are
presented as pedagogy, not philosophy. For more on Polya, see my article
(2011) and Frank (2004)).
If you believe that a mathematical proof is supposed to start with some
unproved axioms, you might be puzzled and frustrated by Wiles work.
No axioms in sight! (The same is true of most mathematical research
publication, all the way back to the Geometrie of Rene Descartes.) Some
sentences from Wiles paper will convey a sense of how it reads.

The key development in the proof is a new and surprising link between two
strong but distinct traditions in number theory, the relationship between
Galois representations and modular forms on the one hand and the
interpretation of special values of L-functions on the other. The former
tradition is of course more recentThe second tradition goes back to the
famous analytic class number formula of Dirichlet, but owes its modern
revival to the conjecture of Birch and Swinnerton-Dyer. In practice,
however, it is the ideas of Iwasawa in this field on which we attempt to
draw, and which to a large extent we have to replace..The turning point
in this and indeed in the whole proof came in the spring of 1991. In
searching for a clue from commutative algebra I had been particularly
struck some years earlier by a paper of Kunz [62]It was only on reading
Section 6 of [121] that I learned that it followed from Tates account of
Grothendieck duality theory for complete intersections that these two
invariants were equal for such rings. The impact of this result on the main
problem was enormous.Then in August, 1991 I learned of a new
construction of Flach [35].Then, unexpectedly in May, 1993, on reading
of a construction of twisted forms of modular curves in a paper of Mazur
[71] I made a crucial and surprising breakthrough. ..Believing now that the
proof was complete, I sketched the whole theory in three lectures in
Cambridge, England (Wiles 1995)

One doesnt have to be familiar with L-functions and so on to see that this
proof depends on a vast acquaintance and deep understanding of the
literaturethe relevant parts of established mathematics.

3. Established mathematics
Established mathematics is the body of mathematics that is accepted as the
basis for mathematicians proofs. It includes proved statements in the
 Establishing New Mathematics 131

literature, and also some simpler statements that are so well accepted that
no literature reference is expected. The central core of established
mathematics includes not only arithmetic, geometry, linear and polynomial
algebra, and calculus, but also the elements of function theory, group
theory, topology, measure theory, Banach and Hilbert spaces, and
differential equations--the usual first two years of graduate study. And
then to create new mathematics, one must also master major segments of
the established knowledge in some special area.
Every mathematician has mastered a substantial portion of established
mathematics, and has complete confidence in it. He/she could not be
qualified, accepted or recognized as a mathematician without attaining
such confidence and mastery. Confidence in established mathematics is for
a mathematician as indispensable as confidence in the mechanics of a
piano for a piano virtuoso, or confidence in the properties of baseballs and
bats for a big league baseball player. If youre worried about that, you
arent even in the game.
Established mathematics is an intricately interconnected web of
mutually supporting concepts, which are connected both by plausible and
by deductive reasoning. Starting from numbers and elementary geometry,
we have built a fantastically elaborated and ramified collection of
concepts, algorithms, theories, axiom systems, examples, conjectures, and
open problems. It provides most of the models or applications that are
in daily use, in the marketplace, the bank, and the inter-planetary rocket
control center. Deductive proof, mutually supporting interconnections, and
close interaction with social life (commerce, technology, education) all
serve to warrant the assertions of established mathematics. Deductive
proof is the principal and most important warrant. Publication of a
research result means adding something new to the body of established
mathematics.
How is the established core established? We mathematicians do
remember once having seen the Fourier transform proved, by means of
some more elementary parts of algebra and calculus, but, as Vaughan
Jones proclaims, thats not the most important part of why we believe in
the Fourier transform. And where does our belief in elementary algebra
and calculus come from? Not from axiomatic set theory (Zermelo and
Fraenkel), nor from Dedekind and Peanos axioms of the natural numbers.
For centuries before Dedekind, Peano, Zermelo or Fraenkel were born, the
practice of arithmetic, algebra and calculus had been firmly established. In
actual mathematical practice, mathematicians start from a given, a basis,
which we accept as firmly as we accept the reality of the physical and
social worlds. As Jones said, to question the Fourier transform, or any
132 Chapter Eight

other standard mathematical tool, would be mathematical lunacy.

Established mathematics is warranted by common consent based on
shared experience, and reinforced by its logical and intuitive connection to
basic arithmetic.
The body of established mathematics is not a fixed or static set of
statements. The new and recent part is in transition. A piece of
mathematics first arises in the course of a mathematicians work as new
knowledge. It gains status as he/she communicates it to his closest
community. If it has any interest to others, it spreads by word of mouth
and electronic media. It is written up for publication, spread around the
internet, and after scrutiny by referees, may be published, in print or on
line. After that it may fall into obscurity, or it may became firmly
established, generally known and used. What is firmly established varies
both in time and in space, because people in New York, Paris or Moscow
may choose different tools and ideas from established mathematics.
What justifies established mathematics? Doesnt it have to have a
foundation? No, it doesnt. Providing a foundation for established
mathematics has fascinated many people, but it is not a necessity.
Although arithmetic can be derived from set theory, arithmetic does not
rely on set theory as a foundation. Arithmetic is a well-established ancient
practice, closely linked to visual and tactile interpretations, and essential
for government, commerce, and industry. Accepting it is a standard
criterion of sanity. To a lesser degree, similar things can be said about
ordinary plane and solid geometry and ordinary school algebra. For
mathematicians, the essentials for competence go on to calculus, real and
complex functions, groups, linear algebra, differential equations, measure
theory, and Banach and Hilbert spaces.
Established mathematics is established on the basis of history, social
practice, and internal coherence. The more recent parts were established
on the basis of rigorous proofs based on the older parts. It doesnt need a
foundation, it IS the foundation for what the mathematician is trying to do:
to build on it.
What has been published remains subject to criticism or correction.
The status of established mathematics is not absolute truth, but rather,
warranted assertibility. In the higher reaches of this structure, (for example
the nonlinear partial differential equations of evolving continuous media)
the situation is rather different from that in elementary arithmetic, and
becomes to some extent analogous to the situation in empirical science. In
a mathematical model for physics or engineering, while of course it is
hoped that mathematical reasoning will enlighten the physical problem, it
also happens that physical reasoning enlightens a mathematical problem.
 Establishing New Mathematics 133

At the advanced research level, where sometimes only a handful of experts

have the ability and the motivation to check a proof or a derivation,
mathematics does not have quite the same status of virtual certainty as in
elementary arithmetic. We strive to understand strange structures for
which our intuition fails. We check until we feel we have checked as well
as possible, and have found no loopholes. A notorious example is the
classification of finite simple groups. The leading organizer of that
research community compared their achievement--a complete list of the
finite simple groups--to flying over Antarctica, trying to make sure you
havent missed any mountain tops (Gorenstein). Its quite unlikely that
you would have overlooked one. A kind of warranted assertibility.

4. Mathematicians proof vs. axiomatic proof

Let me display as sharply as possible the difference between this kind of
proof, the kind produced by Andrew Wiles, which I am calling
mathematicians proof, and the kind of proof, going back to Aristotle,
called axiomatic proof. An axiomatic proof is supposed to transmit or
transport truth value from the axioms to the conclusions. In the pre-
modern understanding of Euclidean geometry, one knew that the axioms
were true (perhaps because they were self-evident) and therefore one
knew that the theorems were true. One can say that modern mathematics
was born when non-Euclidean geometry was discovered--when the
believed self-evidence of Euclids axioms evaporated. Nothing claims to
be self-evident any more.
Nowadays it is occasionally claimed or hoped that the set-theoretic
axioms of Zermelo and Fraenkel are a basis for all of standard
mathematics (including the works, cited by Wiles, of Mazur, Tate, Kunz
and so on). My personal favorite is number 5, the Replacement Schema.
Here it is [Jech]:

u1uk[x!y(x,y,)
wvr(rv s(sw & (x,y,[s,r,]))]

(The image of a set under a definable map is itself a set.)

Of course it isnt faith in Zermelo-Fraenkel (or any alternative set-

theoretic axioms) that causes mathematicians to accept Wiles proof. It is
the status of established mathematics as a whole, and the status of the
results he quotes, as accepted parts of established mathematics. This is
characteristic of most contemporary mathematical proofto start from
134 Chapter Eight

established mathematics, which is taken as given, known, reliable. Any

proposed axiomatic foundations cannot be as credible and reliable as the
established mathematics they are supposed to support.
By starting from established mathematics, this mathematicians
proof establishes new mathematics. Axiomatic proof, on the other hand,
cannot establish anything, because axioms are not established, they are
simply postulated. Cellucci argues convincingly that any attempt to
establish them leads either to an infinite regress, or else to a more or less
arbitrary, subjective decision to adopt them provisionally--by fiat. He
writes:

It is widely believed that the axiomatic method guarantees the truth of a

mathematical assertion (Rota). This belief is unfounded because it depends
on the assumption that proofs are deductive derivations of propositions
from primitive premisses that are true in some sense of `true'. Now, as we
will presently see, generally there is no rational way of knowing whether
primitive premisses are true. Thus either primitive premisses are false, so
the proof is invalid, or primitive premisses are true but there is no rational
way of knowing that they are true, then we will be unable to see whether
something is a proof, so we will be unable to distinguish proofs from non-
proofs. In both cases, the claim that the axiomatic method guarantees the
truth of a mathematical assertion is untenable. (Cellucci 2008).

5. Mathematicians proof is semantic, not syntactic

In addition to the absence of axioms, and the total reliance on the body of
established mathematics, there is another very important, striking feature
of Wiles discourse to be noted by philosophers, and again, one that is
found in nearly all mathematical publication. There is no syntactic
argument, no syllogistics or modus ponens! In fact, it has often been
remarked that mathematicians who are not logicians, know very little
logic, and never mention logic in our work. That does not mean that we
are illogical or irrational! It means (if I may use a technical term from
logic) that our reasoning is semantic rather than syntactic (Rav). Wiles
talks about all sorts of esoteric mathematical entities with complete
confidence, he knows what he means, he knows the relevant facts or
properties. His argument simply uses the facts and properties of
mathematical entities. He can do that because the number theorists who
will understand and evaluate his work have congruent or matching
understandings of those objects or entities. Of course at times he does look
into a book on his shelf, but certainly not at every line of his paper. He
knows what a Galois representation is, he knows what a semistable elliptic
 Establishing New Mathematics 135

curve isnot just because he has memorized a formal definition, he

knows their properties and capabilities well enough to use them effectively
in his proof. That is what I mean when I say he has mental models of these
objects--mental models which were acquired and developed in the course
of his mathematical practice, which shaped them and molded them to be
congruent or matching to the models possessed by other experts in his
field of number theory.
For more explanation and justification of mental models of mathematical
entities, see my [2011], p. 312 ff.. These socially regulated mental
models of mathematical objects are candidates for the new semantics
sought in footnote 24 of Buldt et al. (an important task for the philosophy
of mathematics is, then, to work out the details of a new semantics for the
platonese we still prefer to speak.) [Warning! The term mental model is
used here with a different meaning than in the works of cognitive scientist
Philip N. Johnson-Laird.]
Mathematical reasoning, in the construction of a proof, is based on
possession of shared mental models of mathematical entities, and on using
ones knowledge of these mental models in ones possession. Indeed,
something similar was already expressed by John Locke in 1690. if we
consider, we shall find that [the knowledge we have of mathematical
truths] is only of our own ideas. The mathematician considers the truth and
properties belonging to a rectangle or circle only as they are an idea in his
own mind. (An Essay concerning human understanding). In 1739 David
Hume enriched this viewpoint by contextualizing it socially: There is no
Algebraist nor Mathematician so expert in his science, as to place entire
confidence in any truth immediately upon his discovery of it, or regard it
as any thing, but a mere probability. Every time he runs over his proofs,
his confidence encreases; but still more by the approbation of his friends;
and is raisd to its utmost perfection by the universal assent and applauses
of the learned world. (Treatise of Human Nature, p. 231)
Both in heuristics and in deductive proof, the mathematician works
with his/her socially sanctioned mental model. Arguments or steps in a
deductive proof are convincing when they are clearly seen, mentally
perceived as unquestionable. A familiar example is the commutative law
of multiplication in elementary arithmetic. To do 7 times 9 you imagine
seven rows of soldiers, and in each row, nine soldiers. But look at it from
the side instead of from the front, and you see nine rows, with seven
soldiers in each row. So 7 times 9 is the same as 9 times 7. And this
insight could just as well have been about 5 and 11 as about 7 and 9. So
you clearly see, it doesnt matter which two numbers you multiply, the
result will always be the same if you reverse the order. This is a
136 Chapter Eight

completely convincing, correct proof--not just a guess or a conjecture.

Still, graduate students must also learn the tedious proof in Landaus
classical text, which follows from the Dedekind-Peano axioms by
mathematical induction. The first proof is intuitive and explanatory, the
second calculational and mechanical.
In some other branches of mathematics, the role of established
mathematics is not so heavy. In combinatorics one can sometimes prove a
counting formula for finite sets by simply presenting an ingenious way to
combine those sets from smaller sets. Proof ideas are sometimes even
presented without wordsas diagrams that speak for themselves (Nelsen
1993 and 2000). That kind of proof is quite different from Wiles kind of
proof. On the other hand, it is even more clearly not axiomatic or
syntactic. One can call it visual or diagrammatic.
That raises the question, What is a proof, really? The simplest
truthful answer was given long ago by the Ideal Mathematician: A proof
is just an argument that convinces a skeptical expert. (Davis and Hersh,
1981). But what does it take to convince a skeptical expert? That is not so
easy to say. At a minimum, the proof should explicitly avoid all the
familiar, well-known fallacies, booby traps, and oversights that workers in
the field are required to know about. A world-renowned probabilist said in
my hearing, You check the most delicate points in the paper, and if that
part is OK, you figure the rest is probably correct.
The Fregean notion of formal proof, using formal derivation in a
formal language, is not relevant to proof as it is known in nearly all
published math papers. However, an active group of computer programmers
and mathematicians are formalizing mathematicians proofs of some
interesting theorems (see Hales and Harrison). The Flyspeck Project of
Thomas Hales proposes, over a period of years and with the assistance of a
considerable group of specialists, to produce a complete formalization of
Hales proof of Keplers conjecture on the packing of space by spheres.
Such a project requires replacing any semantic steps in the proof by
syntactic ones. If successfully completed, the credibility of Hales proof
would be raised from, say, 99% to 99.99%. As Hales is the first to admit,
and as any computer user well knows, 100% is not attainable.
A formal proof is either a formalized axiomatic proof or formalized
from a mathematicians proof. If it is a formal axiomatic proof, the
shortcomings of axiomatic proof, as a source of truth, apply just as well to
a formalized version. If it is a formalization of a mathematicians proof, it
is still based on some parts of established mathematics, but the semantic
reasoning of the mathematician has been replaced by the syntactic
reasoning of the logician. We would expect this formalized proof to be
 Establishing New Mathematics 137

more reliable than the mathematicians semantic proof. Such a formal

proof is in fact a long, complicated computer program. Are long,
complicated computer programs absolutely reliable, with certainty? No
one in 2012 USA believes such a thing. Indeed, in his impressive,
convincing article on formalizing his Kepler conjecture proof, Hales starts
out by disarming criticism. He offers quite an extensive account of the
unreliability of very long, complicated computer programs. They can be
more reliable than human semantic proofs, they cannot be totally error-
free.

6. Established mathematics is fallible

But then, mathematicians proof does not guarantee truth, whatever one
might mean by truth. On pages 43--47 of my (1997) is a litany of
famous and not-so-famous published mistakes. To this list can be added
the erroneous prize-winning publication by Henri Poincare on celestial
mechanics, which forced Gosta Mittag-Leffler at great expense to retrieve
and shred all the copies of the Acta Mathematica that had been mailed out,
so that a completely new article by Poincare could be published. (See
Barrow-Green.)
Mistakes continue to be found and corrected, after publication. The gap
or the error is repaired, corrected, fixed up, without bringing down the
whole body of established mathematics. Hilberts Grundlagen der
Geometrie is devoted to correcting, fixing up, the 2,000-year-old Elements
of Euclid. Sometimes the word correct is used instead of true. After
the Grundlagen had been published, some defects were noticed in it (and
of course, fixed up). What counts as rigorous proof has evolved historically,
and has often been controversial. Some wise philosopher (was it
Wittgenstein?) has defined mathematics as the subject where it is possible
to make mistakes. One is often quite certain that something is false or
incorrect. One is not often absolutely sure what is true or correct.
Questions of truth versus fiction are irrelevant to practice. Established
mathematics is assertiblethat is, available for use in mathematical
proof. In fact, it is warrantedly assertible. (The unappealing expression
warranted assertibility goes back to John Deweys pragmatism.) What is
a warrant? A warrant is a justification to act on an assertion, a
justification based on lived experience. In mathematics, two important
warrants are deductive proof, and successful applicationapplication both
within mathematics itself, and in the real world. There are also weaker
warrantsanalogy and induction--which only grant plausibility, not
established status. Analogy and induction serve to justify conjectures, to
138 Chapter Eight

justify a search for deductive proof. Neither analogy nor induction nor
deductive proof can establish the truth of a mathematical statement, for
Truth in the sense of perfect certainty is unattainable.
Experience! That is really what is behind it all. Experience never
guarantees truth. It can provide warranted assertibility. Deductive proof is
the strongest warrant we know for the assertibility of some proposition.
The strongest possible warrant, yes. Absolute certainty, no. Absolute
certainty is what many yearn for in childhood, but learn to live without in
adult life, including in mathematics.
Disregarding Vaughan Jones use of the word true, and contrary to
the picture often presented in logic text books, deductive proof in
mathematical research publication does not establish anything as true.
Deductive proof connects some proposed result to the body of established
mathematics. Once the proposed theorem is accepted or established, one is
permitted to use it in other proofs. Jones did well to keep the word truth
in scare quotes. Trying to explain what one means by truth in
mathematics is a hopeless quicksand. Jones instead adduces the many
irresistible warrants for the Fourier transform. One of those warrants is the
essential existence proof. This proof makes the Fourier transform as
reliable as basic arithmetic or algebra. If it was somehow found that the
Zermelo-Fraenkel or Dedekind-Peano axioms are contradictory, the
Fourier transform would not be abandoned, any more than the rest of
established mathematics would be abandoned. They are established more
firmly than the axioms that are presented as their foundation.
I must add that warrantedness comes in degrees. The strength of the
warrant varies from virtually unassailable, for standard arithmetic,
algebra, geometry, calculus, linear algebra and so on, down to very solid
for much-studied well known results, both old and new, including the
Wiles-Taylor FLT, and further down to reasonably reliable but needing
some care, which would apply to more recent and less well-known
publications. As more convincing arguments for a mathematical statement
are discovered, it becomes more strongly warranted. A deductive proof
makes it part of established mathematics, but thats not always the end of
the story. If the statement is widely studied, analyzed and used, if it
becomes closely connected, both plausibly and rigorously to other
established mathematics, then its warrant becomes stronger and stronger.
It can even become, like the Fourier transform, so firmly embedded in
established mathematics that it is inconceivable to exclude it.
Plausible (non-rigorous) reasoning is a warrant for making a conjecture,
even for establishing it as a plausible, well-founded conjecture (like
FLT, before Wiles.) And then plausible reasoning (problem-solving) is
 Establishing New Mathematics 139

likely to be essential in finding the rigorous proof. But plausible reasoning

(analogy, induction) is not accepted as a substitute for rigorous proof.
Rigorous proof is the method by which established mathematics becomes
established. But of course, problem solving uses not only plausible
reasoning, it also uses established mathematics, which was previously
established by rigorous proof! The front of mathematics (rigorous proof)
and the back (problem-solving by plausible reasoning) are not opposed or
competing, they are intricately interconnected, they are inseparable. One
attains rigorous proof by a problem-solving process. In the process of
proving a conjecture, deficiencies, weaknesses or imprecisions are often
revealed and corrected.
When people speak of mathematical certainty, they ordinarily mean
very strongly warranted assertibility, not total, absolute certainty.
Mathematics is human, and nothing human can be absolutely certain.
Well- established core mathematics is warranted as strongly as anything
we know. Warranted assertibility is all that can be attained, either in
empirical science or in mathematics. Science and mathematics are different,
because they use different kinds of warrants to confer assertibility.

7. Published vs. private, rigorous vs. plausible

Mathematical publication is not identical with mathematical knowledge.
All that is published is not correct knowledge, mistakes sometimes do get
published. And more importantly, all correct knowledge does not become
published. Some things are not submitted for publication even though
everybody knows they are truein fact, BECAUSE everybody knows
they are true. (Novelty is normally a prerequisite for publication in a
research journal.) Some things are not submitted for publication even
though everybody knows they are true, because they have never been
proved. Much practical down-to-earth how-to-do-it knowledge in
scientific computing falls under this heading (Hersh, 2011).
If you have made a publishable discovery, the process by which you
found out your result probably wont be included in your article.
Discovering the interesting result was probably the outcome of a heuristic
investigation, of the kind that Polya and Cellucci describe. But you will
probably omit that story from your article, if only to save yourself extra
trouble. If you choose to include it, you risk a rejection note from the
editor: We dont have space for all the good papers we are receiving,
even without irrelevant heuristics. This policy is very unfortunate. The
final polished deductive proof may conceal a key insight, that could have
made the mathematical result more meaningful and accessible. Such
140 Chapter Eight

insights may then be available only by personal contact between selected

individuals. Its important to advocate publication of the heuristic side of
mathematical discovery, along with the deductive proof.
Deductive proof is intended to compel agreement. It serves to convince
and to explain, sometimes one more than the other. It legitimates a result
as established. By contrast, a plausible derivation does not establish the
result, even if, as in the example of FLT, it has been verified in millions of
individual cases. Thats where Polyas respect for rigorous deduction
comes in. But Polya is mistaken when he says that deductive proof renders
a statement absolutely certain. Cellucci rightly denies this. Deductive
proof is the standard for acceptance of ones findings into the body of
established mathematics. That is just saying their assertion is warranted.
Not absolutely guaranteed to be free of error.
Sometimes a rigorous deductive proof is simply not available, at a time
when a practical decision must be made. In the absence of deductive proof,
there can still be practical certainty, which can justify decisions affecting
billions of dollars, and even human life (Hersh 2011). Nevertheless, the
distinction between a proved theorem and an unproved conjecture is the
central, characteristic feature of mathematics, as practiced from Euclid to
this day. To underestimate or ignore it is out of the question.
In textbook writing, the axiomatic method has advantages of economy
and clarityand the disadvantage of possibly obscuring goals and
motivations. ZF set theory is a branch of mathematics, not a foundation
for all the rest of mathematics. But questions of logical structure are
intrinsically interesting, and there is some interest in the logical foundations
of Wiles proof of FLT. As part of modern algebraic geometry, his work is
connected to certain very large category-theoretic constructions called
Grothendieck universes (see [McLarty]).

8. Established mathematics is not controversial

Since mathematicians proofs grow out of and are based on established
mathematics, we may want to check, is established mathematics doubted
or rejected by a significant number of dissenters? Is it in danger of
overthrow? Established mathematics is a historically evolved construct.
There have been several episodes that could be construed as major
challenges to it. We will see that in every case, the challenge was made
from within established mathematics, not against it. The episodes we will
briefly consider are the constructivist-intuitionist critique of Kronecker,
Brouwer and Bishop; the revival of infinitesimal methods by Abraham
Robinson; the theoretical mathematics proposal of Quinn and Jaffe: and
 Establishing New Mathematics 141

the introduction of computers into mathematical research in various ways

by several contemporary mathematicians.
The sharpest criticism came from the constructivists Leopold Kronecker,
Luitzen Brouwer and Errett Bishop. This controversy has a long history
and a huge literature. Unlike Kronecker and Brouwer, Bishop actually
aims to affirm the structure and content of established mathematics. His
goal is to reconstruct it, as closely as possible, without the law of the
excluded middle. His monumental effort was directed to either establish
the contents of classical analysis constructively, or else to provide the
closest possible constructive substitute. What is most remarkable is how
well he succeeded in recreating or re-justifying it, with appropriate tweaks
and precautions. This achievement of his is not a repudiation of
established mathematics, but an attempt to strengthen it.
An utterly different re-conceptualizing of classical mathematics was
made possible by Abraham Robinsons introduction of nonstandard
analysis, legitimizing the once illegal infinitesimals.
Nonstandard analysis may appear to be a radical challenge to
established mathematics, since it violates the well-established banishment
of infinitesimal quantities. But Robinson by no means claimed that it
could stand apart from and independent of established mathematics. On
the contrary, he devoted the first chapters of his book to establishing it on
standard foundations, both in logic and in set theory. (An example where
foundational research made a major contribution to mainstream
mathematics.) Of course, he did not claim that his results were true,
since no one claims that the basic statements of either logic or set theory
are known with certainty to be true. All he did was the usual normal thing,
to prove his new results by deriving them from established mathematics.
The upshot was that new results and new methods were added to the body
of established mathematics.
The proposal by Quinn and Jaffe was intended to provide some kind of
license for mathematical publication without complete proof. They
perceived a problem, that such results were appearing in the literature
without being labeled as provisional or incomplete, but they didnt want to
prevent such publications, which they recognized as having a useful role to
play. They merely wanted to label them as such. Their proposal stimulated
a long discussion and controversy. It received a great deal of criticism and
opposition, and very little support.
Finally, there has been some foundational discussion related to
computers. We have already described the work of Thomas Hales and his
collaborators on formalized proof. Gonthier reports that the entire proof of
the four-color theorem has now been formalized or computerized. The
142 Chapter Eight

original version, by Appel and Haken, combined sections of ordinary

mathematicians proof with substantial sections of computer calculations.
This original version of the proof prompted a much-discussed article by
Thomas Tymoczko, who argued that by incorporating a computer
calculation into their proof, the authors were changing the standard notion
of mathematical proof. The following discussion dealt with two aspects of
proof. Does it establish certainty? Does it provide understanding? As to
the first, it is clear that computer proof is fallible, for several reasons. On
the other hand, proof by human calculation is also fallible, and there is a
powerful argument that for calculations that are very long, complex, and
tedious, computers are more reliable than humans. On the second point,
Paul Halmos was particularly vehement in rejecting the computer proof of
the four-color theorem, because we learn nothing from it. To this, there
are two answers. First of all, it may turn out that we cant find any other
proof, so we can either accept the theorem as proved, or reject it, on the
basis of what we have, namely the computer proof. Of course, the
computer proof must be checked by the best standards of computer
verification. The mathematical community already has clearly accepted
the four-color theorem as proved. In addition, it seems that some hand-
made proofs also dont provide much insight.
Doron Zeilberger has proposed in all seriousness that proofs be graded
according to the degree of certainty that is claimed. This suggestion has
not received any support that I am aware of. Finally, it is important to
notice the new field of experimental mathematics as advocated by Jon
Borwein and his collaborators, in several books and a journal of that name.
Experimental mathematics amounts to systematic and persistent use of
computers to make mathematical discoveries. It is avowed and understood
that such discoveries are not accepted as established until a traditional
deductive proof is given. The experimental mathematics of Borwein and
Bailey is an elaboration or modernization of the heuristics of Polya or the
analytic method of Cellucci. It simply brings the speed and memory of the
computer in a sophisticated way to make heuristics much more powerful.
Computers will be used more and more, both in heuristic, or problem
solving, and in actual proving, using formalized reasoning. These
important additions to the mathematicians repertoire will only strengthen
the two sides of mathematical workheuristics and rigor.
We see that all of these critical and radical proposals completely accept
the legitimacy of established mathematics! One may expect it to absorb
and incorporate within itself any future challenges
In practical decision making, both in empirical science and in daily
life, all that we mean by true is well justified or firmly established.
 Establishing New Mathematics 143

As John Dewey famously phrased it, warrantedly assertible. The credibility

of established mathematics is based on experiencemany peoples
experienceand on its connection and application to practical life,
including commerce, science and technology. The purpose of mathematical
proof is to endow a new result with that strong credibility. This is just the
mathematics version of John Deweys view of logic and knowledge.

Acknowledgments
Thanks to Carlo Cellucci, Martin Davis, David Edwards, Sol Feferman
and Robert Thomas for helpful suggestions.

References
Aberdein, A. (2006) The Informal Logic of Mathematical Proof, IN:
Hersh, R. (2006) 56-70.
Azzouni, J. (2005) Is there still a Sense in which Mathematics can have
Foundations? IN: Sica, G. ed. Essays on the Foundations of
Mathematics and Logic, Monza, Italy, Polimetrica International
Scientific Publisher., 9-47
Barrow-Green, J. (1953) Poincare and the three-body problem, London
Mathematical Society
Bishop, E. (1967) Foundations of Constructive Analysis, New York,
McGraw Hill
Borwein, J., Bailey, D. and Girgensohn, R. (2004) Experimentation in
Mathematics, Natick, A K Peters
Borwein, J. and Bailey, D. (2004) Mathematics by Experiment, Natick, A
K Peters
Buldt B., Lowe B. & Muller T. (2008) Towards a New Epistemology of
Mathematics, Erkenntnis
Cellucci, C. (2011a) Philosophy of Mathematics: Making a Fresh Start,
paper presented at the 14th Congress of Logic, Methodology and
Philosophy of Science, Nancy, July 19-26
. (2011b) Classifying and Justifying Inference Rules, IN: Cellucci C.,
Grosholz E. & Ippoliti E., eds. Logic and Knowledge, Newcastle upon
Tyne, Cambridge Scholars Publishing, 123-148
. Why Proof? What is a Proof? IN: Lupacchini R. & Corsi G., eds.
Deduction, Computation, Experiment, Exploring the Effectiveness of
Proof, Berlin, Springer, 1-27.
. (2006) Introduction to Filosofia matematica, IN: Hersh, R. (2006),
17-36.
144 Chapter Eight

. (2005) Mathematical Discourse vs. Mathematical Intuition, IN:

Cellucci, C. & Gillies D., Mathematical Reasoning and Heuristics,
London, Kings College Publications, 137-165.
Dales, H. G. and Oliveri, G. (1998). Truth in Mathematics, Oxford,
Clarendon Press.
Davis, P. & Hersh, R. (1981) The Mathematical Experience, Boston,
Birkhauser.
Devlin, K. (2008) A Mathematician Reflects on the Useful and Reliable
Illusion of Reality in Mathematics, Erkenntnis
Dewey J. (1929). The Quest for Certainty. New York: Minton, Balch and
Company.
. (1934). Art as Experience. New York: G. P. Putnam.
. (1938). Logic, The Theory of Inquiry. New York: Henry Holt.
Frank T. (2004) George Polya and the Heuristic Tradition, Revista
Brasileira de Histria da Matemtica, 4 (7) :19 - 36
Godel K. (1947). What is Cantors Continuum Problem? American
Mathematical Monthly, 54: 515-525.
Gonthier, G, (2008) Formal ProofThe Four-Color Theorem, Notices of
the American Mathematical Society, 55 (11)
Gorenstein, D. (1979) The Classification of Finite Simple Groups, Bulletin
of the American Mathematical Society, BN. S. 1, 43-199
Gowers, T. (2008) The Princeton Companion to Mathematics, Princeton
University Press
Grosholz E.R. (2007). Representation and Productive Ambiguity in
Mathematics and the Sciences. New York: Oxford University Press..
Hadamard, J. (1949) The Psychology of Invention in the Mathematical
Field, Princeton University Press.
Hales, T. (2008) Formal Proof, Notices of the American Mathematical
Society 55 (11)
Harrison, J. (2008) Formal ProofTheory and Practice, Notices of the
American Mathematical Society, 55 (11), 1495-1506
Hersh, R. (1988) Mathematics has a front and a back, Eureka , also in
Synthese 88(2) August 1991
. (1997a) ProveOnce More and Again, Philosophia Mathematica 5,
153-165.
. (1997b) What is Mathematics, Really?, New York, Oxford University
Press.
. (2006) 18 Unconventional Essays on the Nature of Mathematics, New
York, Springer.
 Establishing New Mathematics 145

. (2011) Mathematical Intuition: Poincare, Polya, Dewey, IN: Cellucci

C., Grosholz E. & Ippoliti E., eds. Logic and Knowledge, Newcastle
upon Tyne, Cambridge Scholars Publishing, 297-323.
Hume, D. (1739) A treatise of human nature, New York, Penguin, 1969.
Jaffe, A. and Quinn, F. (1993) Theoretical mathematics: toward a cultural
synthesis of mathematics and theoretical physics. Bull. American
Math. Soc., 29, 1-13.
Jech, Thomas, "Set Theory", The Stanford Encyclopedia of Philosophy
(Winter 2011 Edition), Edward N. Zalta (ed.), forthcoming URL =
http://plato.stanford.edu/archives/win2011/entries/set-theory/
Johnson-Laird P.N. (1983). Mental Models: Towards a Cognitive Science
of Language, Inference, and Consciousness. Cambridge, MA: Harvard
University Press.
Jones, V. F. R., (1998) A Credo of Sorts, in Dales and Oliveri, 203-214
Kitcher, P. (1983) The Nature of Mathematical Knowledge, Oxford
University Press, New YorkLandau, E. (1960) Foundations of analysis,
New York, Chelsea Publishing Company
Locke, J. (1690) An essay concerning human understanding, In The
Empiricists, Garden City, N.Y. Dolphin, 1961.
Lowe, B. and Muller, T. (2010) PhiMSAMP Philosophy of Mathematics:
Sociological Aspects and Mathematical Practice, College Publications,
London
Lynch, J. and Singh, S. (1996) Fermats Last Theorem, a motion picture
on line.
Mancosu, P. (2008) The Philosophy of Mathematical Practice, New York:
Oxford University Press
McLarty, C. (2010) What does it take to prove Fermats last theorem?
Grothendieck and the logic of number theory, The Bulletin of
Symbolic Logic 16 (3)
McLellan J. A. and Dewey J. (1897). The Psychology of Number and Its
Applications to Methods of Teaching Arithmetic. New York: D.
Appleton and Company.
Mozzochi, D. J. (2000) The Fermat Diary, Providence, the American
Mathematical Society.
Nelsen, R. (1993) Proofs without words: exercises in visual thinking,
Washington, Mathematical Association of America
. (2000) Proofs without words II: more exercises in visual thinking,
Washington, Mathematical Association of America
Odlyzko, A. (2011), Review of Experimental Mathematics in Action by D.
H. Bailey et al., American Mathematical Monthly 118, 946-951
146 Chapter Eight

Poincare, H. (1913) Mathematical Creation, IN: The Foundations of

Science, New York, The Science Press.
Polya G. (1945). How To Solve It. Princeton: Princeton University Press.
. (1954). Mathematics and Plausible Reasoning. Princeton: Princeton
University Press.
. (1980). Mathematical Discovery. New York: Wiley.
Polya G., and Szego G. (1925, 1970). Aufgaben und Lehrsatze aus der
Analysis. Berlin: Springer Verlag.
Ratner S. (1992). John Dewey, Empiricism and Experimentalism in the
Recent Philosophy of Mathematics. Journal of the History of Ideas, 53,
(3): 467-479.
Rav, Y. (1999) Why Do We Prove Theorems? Philosophia Mathematica 7,
5-41
. (2007) A Critique of a Formalist-Mechanist Version of the
Justification of Arguments in Mathematicians Proof Practices,
Philosophia Mathematica 15, 291-320
Rescher N. (2001). Cognitive Pragmatism. Pittsburgh: University of
Pittsburgh Press.
Robinson, A. (1966) Non-standard Analysis, Amsterdam, North-Holland
Publishing Company
Sfard, A. (2008) Thinking as Communicating, New York, Cambridge
University Press
Thurston, W. (1994) On Proof and Progress in Mathematics, Bulletin of
the American Mathematical Society 30, 161-177 and in Hersh, R.
(2006) , 37-55
Wiles, A. (1995) Modular elliptic curves and Fermats last theorem,
Annals of Mathematics 142, 443-551
 CHAPTER NINE

FERMATS LAST THEOREM

AND THE LOGICIANS

EMILY GROSHOLZ

SUMMARY The task of reference and the task of analysis (the search for
conditions of solvability of problems) in mathematics are often distinct
and produce disparate discourses, which must then be brought into relation
in a proof or textbook. Wiles proof of Fermats Last Theorem, and the
work of logicians to reformulate the proof in order to reduce its logical
complexity, provide instructive examples of various strategies of
integration, and provide evidence that the philosophy of mathematics must
pay more attention to processes of discovery as well as justification.

KEYWORDS number theory, logic, discovery, reference, analysis, reduction

In this essay on the reception of Andrew Wiles proof of Fermats Last

Theorem (Wiles 1995) among both number theorists and logicians, I
address an important epistemological topic to which, in my opinion, both
mathematicians and philosophers of science have not given sufficient
attention. My central claim is that productive mathematical discourse must
carry out two distinct tasks in tandem, analysis and reference. More
abstract discourse that promotes analysis, and more concrete discourse
(often involving computation or iconic representations) that enables
reference, are typically not the same. The resultant composite text
characteristic of successful mathematical research will thus be
heterogeneous and multivalent, a fact that has been missed by
philosophers who begin from the point of view of logic, where rationality
is often equated with strict discursive homogeneity. It has also been
missed by mathematicians who, using idioms other than logic, are
nonetheless in search of a single unified theory. My ongoing philosophical
project is to examine various strategies of integration that bring disparate
148 Chapter Nine

discourses into rational relation and thereby advance the growth of

knowledge.
I will examine the work of logicians like Angus Macintyre, Colin
McLarty, and Harvey Friedman who analyze and try to reduce the logical
complexity of the proof, in order to showcase the importance of their
strategies of integration for the growth of mathematical knowledge, even
when aspects of the proof may elude their methods. In their project of re-
writing the proof using discourse of lower logical complexity, certain
kinds of abstract structures, used explicitly in Wiles original proof, may
be suppressed, and so aspects of its original organization may be obscured
or complicated. Conversely, what was left unremarked or tacit in the
original, like the foundational justification of some of his sources as well
as certain sources, are brought to light by the logicians attempts to re-
write the proof; and the methods of approximation they use, as well as
their ability to highlight the most combinatoric and arithmetic aspects of
the proof, may turn out to be mathematically suggestive to the number
theorists. Their articulations can thus be considered as extensions of the
original text, where what plays the role of instrument of analysis for the
number theorists becomes an object of reference for the logicians. That is,
the logicians reformulations, while they are sometimes intended to
replace the original, can more fruitfully be considered as superpositions;
the result is then a combination of number theoretical and logical
approaches, rationally integrated, so the information available for the
solution of various novel problems in both number theory and
mathematical logic is increased.

1. Reference and Analysis

I have come to believe that reasoning in mathematics often generates
internally differentiated texts because thinking requires us to carry out two
distinct though closely related tasks in tandem. This is an epistemological
claim about human reasoning in general. The simplest illustration is the
categorical proposition: when we assert S is P, the referential work of
identifying S differs in significant ways from the analytic work of
choosing and determining P. This is true whether the area of research is
rich in objects but poor in relations of interest (like mathematics), or rich
in relations and poor in objects (like Newtonian mechanics applied to the
solar system), or rich in both (like modern biology). (See Dirk Schlimm
2012). Analysis requires us to engage in the abstract, more discursive
project of theorizing, what Leibniz called analysis or the search for
conditions of intelligibility (of problematic objects) or solvability (of
 Fermats Last Theorem and the Logicians 149

objective problems). (Grosholz 2007, Ch. 2) Reference requires the choice

of good predicates, durable taxonomy, and useful notation and icons:
more concretely realized strategies for achieving the clear and public
indication of what we are talking about.
Concepts of rationality and method have been increasingly central to
philosophy of mathematics and epistemology since Descartes wrote his
Discourse on Method and Meditations at the beginning of the seventeenth
century. During the last century, however, rational analysis has been
equated to the deployment of deductive logic (usually first- and second-
order predicate logic), with attempts to assimilate induction and abduction
as closely as possible to deduction, and method to the organization of
knowledge into axiomatic systems expressed in a formalized language.
While deductive argument is important (since its forms guarantee the
transmission of truth from premises to conclusion) as a guide to effective
mathematical and scientific reasoning, it does not exhaust them; and an
unswerving focus on logic diverts attention from other forms of rationality
and demonstration. Indeed, as Carlo Cellucci has argued in a series of
books, this trend has left philosophers with little to say about processes of
discovery in mathematics. Mathematical analysis takes place in the open
systems of ongoing mathematical investigation, while deductive proof
requires a closed axiomatized system, which freezes the current state of
knowledge as if it had been determined once and for all. (Cellucci 1998,
Ch. 6-9; Cellucci 2002, Part 3) The current ways of discussing the two
disparate tasks I noted above in Anglo-American philosophical circles
render the strategies of integration that interest me almost invisible.
Analysis is usually characterized as theory-construction; the mathematician
searches inductively and by abstraction for appropriate predicates and
relations that characterize the objects in question, which are somehow
already given, and abductively for rules and higher level principles that
govern their behavior. Theories are or ought to be axiomatized systems in
a formal language, in terms of which true claims about the objects can be
predicted or explained.
Current discussion of reference in the philosophy of mathematics is a
bit harder to characterize. Some philosophers pretend that structuralism
solves the problem of reference; but structuralism only denies the problem,
which still remains, without addressing it. Sometimes the problem of
model construction is substituted for the problem of reference; this move
is favored by anti-realist philosophers. Thus theories are about models,
though this leaves open the issue of how models refer, or if they do; and
models are not individuated in the way that mathematical things typically
are individuated. A model of first-order Peano Arithmetic is very different
150 Chapter Nine

in kind from a whole number, or a specific Riemann surface; it serves as a

referent for mathematical logic but not for number theory, or even for
algebraic geometry which underwrites so much of modern number theory.
Bertrand Russell argued a century ago that the reference of a name is fixed
by a proper definite description, an extensionally correct description which
picks out precisely the person or natural kind intended, and W. V. O.
Quine argued half a century ago that the ontological commitment of a first
order theory is expressed as its universe of discourse. But first order
theories do not capture the mathematical objects they are about (numbers,
figures, functions) categorically, and the ontological commitments of
higher order theories are unclear. Saul Kripke insisted that we need the
notion of an initial baptism (given in causal terms), followed by an
appropriate causal chain that links successive acts of reference to the
initial act, for only in this case would the name be a rigid designator
across all possible worlds; a rigid designator picks out the correct person
or natural kind not only in this world but in all possible worlds where the
person or kind might occur. Hilary Putnam argued that the ability to
correctly identify people and natural kinds across possible worlds is not
possessed by individuals but rather by a society where epistemic roles are
shared; and Paul Benacerraf argued in a famous essay that linking
reference to causal chains makes an explanation of how mathematicians
refer seem futile.
In sum, it is not generally true that what we know about a mathematical
domain can be adequately expressed by an axiomatized theory in a formal
language, nor that the objects of a mathematical domain can be mustered
in a philosophical courtyard, assigned labels, and treated as a universe of
discourse. What troubles me most about this rather logicist picture is that
the difficulty of integrating or reconciling the two tasks of analysis and
reference (as well as the epistemological interest of such integration) is not
apparent, since it is covered over by the common logical notions of
instantiation and satisfaction.
The assumption seems to be that all we need to do is assign objects and
sets of objects from the universe of discourse (available as a nonempty set,
like the natural numbers) to expressions of the theory. If we carry out the
assignment carefully and correctly, the truth or falsity of propositions of
the theory, vis--vis a structure defined in terms of a certain universe of
discourse, will be clear. In a standard logic textbook, the universe of
discourse is the set of individuals invoked by the general statements in a
discourse; they are simply available. And predicates and relations are
treated extensionally as if they were ordered sets of such individuals. In
real mathematics, however, the discovery, identification, classification and
 Fermats Last Theorem and the Logicians 151

epistemic stability of objects is problematic; objects themselves are

enigmatic. It takes hard work to establish certain items (and not others) as
canonical, and to exhibit their importance. Thus reference is not
straightforward. Moreover, neither is analysis; the search for useful
predicates and relations in mathematics typically proceeds in terms of
intensions, not extensions, in tandem with the search for useful methods
and procedures. Analysis is both the search for conditions of intelligibility
of things and for conditions of solvability of the problems in which they
figure. We investigate things and problems in mathematics because we
understand some of the issues they raise but not others; they exist at the
boundary of the known and unknown. So too, what plays the role of
referent in one mathematical context may appear as an instrument of
analysis in another, and vice versa.
The philosophical challenge is this: how can a philosopher account for
the organization of mathematical discourse, both as it is used in research
and as it is used in textbooks? I would like to argue that organization takes
place at many levels. It is inscribed in many notations and iconic
conventions, in procedures and algorithms, and in the methods that
generalize them. It is expressed in the canonization of objects and
problems, and the collecting of problems into families. And it is precipitated
in iconic as well as symbolic fashion, in diagrams and arrays, as well as in
the enunciation of theorems and principles; moreover, specialized
mathematical language must be explained in natural language. Between
the principles that govern analysis, and the mathematical things to which
we refer, the organization created by the mathematician is quite
multifarious. My claim that mathematical objects are problematic (and so
in a sense historical and in another sense strongly related to the practices
of researchers and teachers) need not lead to skepticism or to dogmatism
about the furniture of the mathematical universe; rather, it should lead us
to examine the strategies of integration that organize mathematical
discourse. We can claim that discourse represents things well without
becoming dogmatic, if we leave behind the over-simplified picture of the
matching up of reference and analysis as the satisfaction of propositions in
a theory by a structure.

2. Wiles Proof of Fermats Last Theorem

Let us look more closely at some aspects of Wiles proof. Fermats Last
Theorem (1630) states that the equation xn + yn = zn, where xyz 0, has no
integer solutions when n is greater than or equal to 3. Fermat himself
proved the theorem for exponent 4, which also reduces the problem to
152 Chapter Nine

proving the cases where n is an odd prime. Euler produced an (apparently

flawed) proof for the case where n = 3 (1753), Dirichlet and Legendre
simultaneously proved the case where n = 5 (1825), and Lam the case
where n = 7 (1839). Sophie Germaine and Ernst Eduard Kummer produced
more general, and generalizable, results in the 19th century, relating the
theorem to what would become class field theory in the 20th century.
The striking feature of Wiles proof, to people who are not number
theorists, is that it does not seem to be about integers! Here is the opening
paragraph of his more than 100 page paper in the Annals of Mathematics.

An elliptic curve over Q is said to be modular if it has a finite covering by

a modular curve of the form X0(N). Any such elliptic curve has the
property that its Hasse-Weil zeta function has an analytic continuation and
satisfies a functional equation of the standard type. If an elliptic curve over
Q with a given j-invariant is modular then it is easy to see that all elliptic
curves with the same j-invariant are modular A well-known conjecture
which grew out of the work of Shimura and Taniyama in the 1950s and
1960s asserts that every elliptic curve over Q is modular In 1985 Frey
made the remarkable observation that this conjecture should imply
Fermats Last Theorem. The precise mechanism relating the two was
formulated by Serre as the -conjecture and this was then proved by Ribet
in the summer of 1986. Ribets result only requires one to prove the
conjecture for semistable elliptic curves in order to deduce Fermats Last
Theorem (Wiles 1995, p. 443).

This apparent change of referents is explained by the fact that the proof
hinges on a problem reduction: the truth of Fermats Last Theorem is
implied by the truth of the Taniyama-Shimura conjecture, that every
elliptic curve over Q is modular. (The converse claim that every modular
form corresponds to a certain elliptic curve had already been proved by
Eichler and Shimura: Fermats Last Theorem follows from the two-way
correspondence.) The condition of modularity is important because then
the elliptic curves L-function will have an analytic continuation on the
whole complex plane, which makes Wiles proof the first great result of
the Langlands Program, and a harbinger of further results. Important
problem-reductions combine, juxtapose and even superpose certain kinds
of objects (and the procedures, methods and problems typical of them) on
other kinds. Wiles proof is not only about the integers and rational
numbers; it is at the same time concerned with much more abstract and
indeed somewhat ambiguous and polyvalent objects, elliptic curves and
modular forms. So for example at the culmination of Wiles proof, where
analysis has invoked cohomology theory, L-theory, representation theory,
and the machinery of deformation theory, we find the mathematician also
 Fermats Last Theorem and the Logicians 153

involved in quite a bit of down-to-earth number-crunching. (Wiles 1995)

(Besides the text of the proof, I also rely on didactic expositions like Ribet
(1995), Darmon et al. (1997), and Li (2001 / 2012, class notes), as well as
Cornell et al. (1997).)
Thus Wiles proof of Fermats Last Theorem can be understood in
terms of two stages. The first stage was already done for him: it is the
result of Eichler-Shimura, which shows that given a certain kind of
modular form f, we can always find a corresponding elliptic curve Ef. The
second stage Wiles had to carry out himself, proving the Taniyama-
Shimura conjecture, that given a certain kind of elliptic curve E, we can
always find a certain kind of modular form that renders it modular. Frey
conjectured (1985) and Ribet proved (1990) that Fermats Last Theorem
follows from this correspondence, qualified by a restriction to semi-stable
elliptic curves. My speculation, to be developed in another essay, is that in
the first stage, modular forms are investigated as the objects of reference,
and treated geometrically as holomorphic differentials on a certain
Riemann surface, while elliptic curves are treated as instruments of
analysis; and conversely in the second stage, Wiles proof, elliptic curves
serve initially as objects of reference (smooth, projective non-singular
without cusps or self-intersectionscurves of genus 1 over a field k, with
a distinguished point O), while modular forms become the instruments of
analysis.
Let us go back to Descartes Geometry for a simpler illustration of the
point of my argument-sketch here. In the problem-solving situations that
constitute the heart of that book, we find algebraic equations, proportions,
carefully labeled (very un-Euclidean) geometrical diagrams, and
schematized mechanical instruments side by side on the page. One
consequence of this novel juxtaposition is the study of polynomials as
such: they become objects of study for algebraists in the 18th and early 19th
centuries, though of course they never lose their attachment to some field
or ring over which their variables and constants range. Notation that
Descartes used as a means of analysis becomes an object of reference for
later algebraists. Conversely, in the 19th c. geometrization of the complex
numbers via the model of the complex plane, the plane as well as the circle
are used as instruments of analysis to organize results that were originally
arrived at via a purely algebraic treatment of imaginary complex
numbers. (The plane was never an object of study for Descartes; his
objects are line segments first and foremost and then, subordinately, planar
algebraic curves. But the possibility of the plane, or more generally 2-
dimensional surfaces as well as n-dimensional Euclidean space, becoming
an object of study in the 19th c. arises as a direct consequence of
154 Chapter Nine

Descartes algebraization of geometry.) Thus, in one context an algebraic

form is an instrument of analysis to solve problems, while in another,
having generated its own distinctive questions, it becomes an object of
study. In one context a geometrical item is the object of study, while in
another it is used abstractly to re-organize a body of results. In order to
decide whether part of a mathematical investigation is concerned with
reference or with analysis, one must look carefully at the context of use.
The central strategies throughout Wiles proof include the use of L-
functions (generalizations of the Riemann zeta function, and Dirichlet
series), as well as representation theory, where p-adic families of Galois
representations figure centrally in the proof of the Taniyama-Shimura
conjecture. Given a certain kind of elliptic curve E, one investigates p-adic
representations in order to construct a corresponding, suitably qualified,
modular form f. Wiles puts the problem in a very general setting: he finds
a universal representation that takes G , the group of automorphisms of
the algebraic completion of the rationals that leaves the rationals
unchanged, to GL2(R), the set of all 2x2 matrices with non-zero
determinant with entries in the universal deformation ring R defined with
respect to a certain finite set of primes . Meanwhile, he also constructs
another universal representation: this one takes G to GL2(T) where the
ring T is the completion of a classical ring of Hecke operators acting on a
certain space of modular forms. (Thus, in both cases, there is a hierarchy
of representations to which all the other, more finitary representations can
be lifted under the right conditions, according to deformation theory.)
Then Wiles shows that T and R are isomorphic! This part of the proof,
the postulation of universal deformations, might seem, from the point of
view of logicians (in particular definability oriented model theorists),
rather extravagant; but in fact this is not where the problem lies for them.
Rather, it is Wiles use of Grothendieck duality about twenty pages earlier
(Wiles 1995, pp. 486f.), where the functor categories use universes, a point
I return to in the next section.
As I observed at the beginning of this paper, mathematical analysis is
the solution of a problem that leads it back to the solution of other
problems. Andrew Wiless fascination with Fermats Last Theorem began
when he was 10 years old, and culminated on the morning of September
19, 1994, when he finally put the last piece of the grand puzzle in place.
To prove the isomorphism, Wiles had to prove a certain numerical
inequality, which turned out to be harder than hed expected. Re-
examining his use of what is called the Kolyvagin-Flach method, while
trying to explain to himself why it didnt seem to be working, he realized
that if he combined it with an earlier approach hed set aside, the problem
 Fermats Last Theorem and the Logicians 155

would be solved. On that morning, something happened that was radically

unforeseeable (even by Wiles, who was very discouraged and did not
believe it would happen), and yet, once it actually took place, presented
the kind of necessity that mathematical results present. It disrupted
mathematics by changing its transversal relations, for now modular forms
were proved to be correlated in a thoroughgoing way with elliptical
equations, and at the same time established a new order. The
unforeseeability was not merely psychological, subjective, and merely
human; the disruption lay in the mathematical objects as well as in the
mind of the mathematician.
Here are two simpler illustrations of my point, that the consideration of
reference must be part of a philosophical account of mathematical
discovery. Once Leibniz articulated the notion of a transcendental curve,
and once the sine and cosine functions were established as canonical
examples of transcendental curves, they were understood as functions
constructed in terms of the circle. Every college student knows that the
circle contains the sine and cosine functions; in a trivial sense they are
inscribed in the circle. Euclid could never have foreseen this disruption of
his object, the circle: but the containment follows necessarily from the
precise definition of the circle, once mathematics develops enough to
distinguish and characterize certain transcendental curves. In the
nineteenth century geometrization of complex analysis, the circle
understood in terms of the equation zn = 1, comes to contain the nth
roots of unity and so to become the emblem of the arithmetic of
cyclotomic fields, now central to number theory and indeed Wiles proof.
Who knows what else the circle will be discovered to contain?

3. The Logicians: McLarty and Friedman

A notable feature of Andrew Wiles proof of Fermats Last Theorem is
that it invokes cohomology theory (inter alia) and thus Grothendiecks
notion of successive universes, which from the point of view of set theory
become very large; and yet the detail of the proof stays on relatively low
levels of that vast hierarchy. In a recent essay, Colin McLarty offers
foundations for the cohomology employed in Wiles proof at the level of
finite order arithmetic; he uses Mac Lane set theory, which has the proof
theoretic strength of finite order arithmetic, and Mac Lane type theory, a
conservative extension of the latter. (McLarty 2012) Angus Macintyre is
re-working aspects of the proof (bounding specific uses of induction and
comprehension) to bring it within a conservative n-th order extension of
Peano Arithmetic (Macintyre 2009) and Harvey Friedman has informally
156 Chapter Nine

speculated that it could be further reduced to Exponential Function

Arithmetic.
Meanwhile, the significant re-working and extension of the proof by
number theorists proceeds independently of logic, in the sense that number
theorists dont seem particularly concerned about the logical complexity of
their methods. (For example, Breuil et al. 2001, and two recent articles by
Mark Kisin, 2009a, 2009b.) On the one hand, we see number theorists
choosing logically extravagent methods that usefully organize their
investigations into relations among numbers, as well as elliptic curves,
modular forms, and their L-functions, inter alia, and make crucial
computations visible and possible. On the other hand, we see logicians
analyzing the discourse of the number theorists, with the aim of reducing
its logical complexity. Should number theorists care whether their abstract
structures entail the existence of a series of strongly inaccessible cardinals?
Serre and Deligne, for example, do sometimes seem to be concerned about
the logical complexity of their methods. (Macintyre 2009, p. 10) Will the
activity of logicians produce useful results for number theorists, or is it
enough if they answer questions of interest to other logicians, such as
whether in fact Fermats Last Theorem lies beyond the expressive strength
of Peano Arithmetic (and thus might be a historical and not merely
artificially constructed example of a Gdel sentence)?
As I have argued above, mathematical discourse must carry out two
distinct tasks in tandem, analysis and reference. In the case of number
theory, the referents are integers and rational numbers in one sense and
additionally, in a broader sense given the problem reduction at the heart of
Wiles proof, modular forms and elliptic curves. For logic, the referents
are propositions and sets (and perhaps also formal proofs), or, if we
include the broader range of category theory as part of logic, categories
(and perhaps also functors). Thus what is an aspect of analysis for the
number theorist may become an aspect of reference for the logician.
Moreover, techniques of calculation that preoccupy the number theorist
remain tacit for the logician because they directly involve numbers, and
considerations of logical complexity that concern the logician remain tacit
for the number theorist because they are not conditions of solvability for
problems about numbers. This disparity is inescapable, but it is also
positive for the advance of mathematics. For when what remains tacit in
one domain must be made explicit in another in order to bring the domains
into rational relation, novel strategies of integration must be devised when,
for example, number theory and logic are brought into working relation.
I use as an illustration of this disparity and the possibility of productive
integration the work of Angus Macintyre (a model theorist) and Colin
 Fermats Last Theorem and the Logicians 157

McLarty (a category theorist) on Wiles proof of Fermats Last Theorem.

At issue is Wiles use of Grothendieck cohomology, as set forth in various
writings and editions of Elments de Gomtrie Algbrique over the third
quarter of the twentieth century. Colin McLarty writes that Grothendieck
pre-empted many set theoretic issues in cohomology by positing a
universe: a set large enough that the habitual operations of set theory do
not go outside it. His universes prove that ZFC is consistent, so ZFC
cannot prove they exist. (McLarty 2012, pp. 359-61) Wiles invokes
Grothendieck cohomology and by implication the vast universes it
involves around page 486 of Wiles (1995), where he uses Grothendieck
duality and parts of Mazurs work, specifically Mazur (1977) and the
textbook An Introduction to Grothendieck Duality by Altman and Kleiman
(1970). The path through these books leads back to Grothendiecks
Elments de Gomtrie Algbrique and functor categories that use
universes.
As McLarty points out, the odd thing is that these rather oblique and
vague references are all that Wiles offers the logician-reader interested in
tracing back his assumptions to their origins; indeed, Wiles never offers an
explicit definition of cohomology. McLarty speculates that Wiles may be
assuming that the Anglophone reader will consult the standard textbooks,
Hartshornes Algebraic Geometry (1977) or Freyds Abelian Categories
(1964); but these books are not included in the extensive references at the
end of Wiles (1995). In any case, both Hartshorne and Freyd treat
questions of proof and foundations in a rather cavalier manner. McLarty
writes that Hartshorne quantifies over functors between categories which
are not well defined in ZF, and he does not prove the basic results he uses.
He cites Freyds Abelian Categories for proofs and also sketches several
other strategies one could use. Freyd in turn waves off the question of
foundations by claiming he could use some theory like Morse-Kelley set
theory, a non-conservative extension of ZF. And that is true of his chief
results (though at least one of his exercises goes beyond that). In general,
from the point of view of the logician, Wiles proves no theorems from the
ground up. (McLarty 2012, pp. 367-68)
Wiles makes use of cohomology theory, and the deformation theory
allied with it, because it helps him to organize the results he needs for his
proof; but it is not what his proof is about. The logical strength of the
theory does not really concern him, so he lets it remain for the most part
tacit and unanalyzed in his exposition. For logicians concerned with model
theory, or with the meaning of Gdels incompleteness theorems,
however, the logical strength of Wiles proof of the Taniyama-Shimura
conjecture, or of other proofs still to be discovered that are now emerging
158 Chapter Nine

from it, is paramount. It must be made explicit, in order to explore the

possibility of proofs of the same result but with lower logical complexity.
One way of posing this question, however, leads us back to the discussion
of the double nature of the definitions in Wiles proof, discussed in
Section 2. Can a proof in number theory really do without geometry? This
is a central question because, even if one succeeds in wresting large parts
of cohomology theory into first or second order arithmetic, even second
order arithmetic will not give any uncountable fields like the reals or
complex numbers or p-adic numbers. An appropriate formalization will
apply to them if one assumes they exist, but will not prove they exist. So
we are dealing not only with the disparity between number theory and
logic, but also with the disparity between number theory and geometry.

4. The Logicians: Macintyre

In a recent essay, The Impact of Gdels Incompleteness Theorems on
Mathematics, Angus Macintyre begins by noting the positive
contributions of logicians to research in various branches of mathematics
(apart from mathematical logic itself). He cites Presburgers work on the
ordered Abelian group , which underlies much of p-adic model theory;
Tarskis work on real closed fields; the uses of Ramseys Theorem in
harmonic analysis; and Herbrand and Skolems contributions to number
theory: Herbrand to ramification theory and cohomology, and Skolem to
p-adic analytic proofs of Finiteness Theorems for diophantine equations.
(Macintyre 2009, pp. 2-3) He then summarizes the reactions of number
theorists to Gdels Incompleteness Theorems. In the last thirty-five
years, number theory has made sensational progress, and the Goedel
phenomenon has surely seemed irrelevant, even though number theorists
are sensitive to the effectivity or logical complexity of their results. On the
one hand, artificially constructed statements that are formally undecidable
seem to be mathematically uninteresting: the equations whose
unsolvability is equivalent (after Gdel decoding) to consistency
statements have no visible structure, and thus no special interest. On the
other hand, the really important results seem mostly to be decidable, at
least in principle: there is not the slightest shred of evidence of some
deep-rooted ineffectivity. (Ibid., pp. 4-5)
Macintyre observes further that while logic is sometimes a good idiom
for recasting mathematical research (as in the cases given above),
sometimes it uncovers results that are of interest to logicians, but not to
geometers or number theorists. What model theory reveals, generally
speaking, are the natural logical or arithmetic-combinatorial features of
 Fermats Last Theorem and the Logicians 159

a subject matter or problem context. Even when the subject matter is

squarely geometrical or topological, these features may be important; but
we cannot expect them to tell the whole story. Logic seems more apt for
the work of analysis than the work of reference in other mathematical
domains. For example, discussing the work of C. L. Siegel, Macintyre
writes, A propos the decision procedure for curves, the natural logical
parameters of the problem, such as number of variables, and degree of
polynomials involved, obscure the geometrical notions that have proved
indispensable to much research since 1929 If ones formalism
obscures key ideas of the subject, one can hardly expect logic alone to
contribute much. (Ibid., p. 6)
In the Appendix to the paper, Macintyre sketches his conception of
what a re-writing of Wiles proof might look like, carried out by a logician
who wanted to show that there is no need for strong second-order axioms
with existential quantifiers involved; in other words, that the proof could
be confined within first-order Peano Arithmetic (PA). Macintyres
conjectured re-writing breaks the proof up into a series of local issues,
giving arithmetic interpretations of specific parts of real, complex or p-
adic analysis or topology. He points out that zeta functions, L-series and
modular forms are all directly related to arithmetic: There is little
difficulty in developing the basics of complex analysis for these functions,
on an arithmetic basis, sufficient for classical arithmetical applications
nothing would be gained by working in a second-order formalism, in a
very weak system. At best such systems codify elementary arguments of
general applicability. (Ibid, p. 7) Thus for the number theorist interested
in advancing research by generalization, the re-writing of the logician
would not be of immediate interest; but for the logician, the re-writing is
of central importance.
However, since number theorists are in fact concerned about the
logical complexity of their methods, in retrospect they would be motivated
at least to study a first-order version of the proof, even if from the
perspective of their immediate interests it appears over-detailed and oddly
arranged. Colin McLarty observes, Macintyre points out that analytic or
topological structures such as the p-adic, real and complex numbers enter
Wiless proof precisely as completions of structures such as the ring of
integers, or the field of rational numbers, which are interpretable in PA.
Macintyre outlines how to replace many uses of completions in the proof
by finite approximations within PA. He shows how substantial known
results in arithmetic and model theory yield approximations suited to some
cases. He specifies other cases that will need numerical bounds which are
not yet known. Theorems of this kind can be very hard. He notes that even
160 Chapter Nine

routine cases can be so extensive that it would be useful to have some

metatheorems. (McLarty 2012, p. 363)
From the point of view of number theory, this re-writing would
damage the organization and perspicuity of the original proof. Thus, the
logically extravagent general methods and theorems seem to be needed
to organize the practice of number theorists. However, the messy and
piecemeal methods of the logician reveal aspects of the proof (its
reducible logical complexity) central to the research practice of logicians.
An analyst or topologist need not be interested in replacing R or C by
finite approximations within Peano Arithmetic, but a model theorist is
highly motivated to spend quite a bit of time and ink in the attempt.
Mathematicians like Wiles, Ribet and Mazur posit the big structures to set
their problem in the best conceptual framework possible, so they can see
how to solve the problem and then how to generalize the result; model
theorists like Macintyre break the big structures into smaller approximations,
in order to solve different kinds of problems. Neither one thinks that a
finite approximation is identical to the original object it approximates; but
for different reasons, and for specified purposes, the number theorist and
the model theorist are both willing to entertain the conjectural equivalence.
Macintyre is concerned about the way many people misunderstand the
import of Gdels incompleteness results, and overstate the inability of
logic to capture the content of important theorems. (It is useful to weigh
his arguments against those of Carlo Cellucci in Cellucci 2007, Section
V.) So at least part of what he is trying to do in the Appendix is to show
that the logical, that is, the arithmetic-combinatorial, aspects of e.g.
Wiles proof loom very large, and can be captured and re-stated
perspicuously by logicians. I would observe that the canonical objects of
geometry and topology can typically be treated by arithmetical-
combinatorial approaches, even if those approaches do not allow us to
capture the canonical objects categorically, or to prove their existence. The
work of logic in other mathematical domains is not reference but analysis.
Macintyre also points out that the monsters, the sentences, functions, or
set-theoretical objects that seem to be squarely beyond the realm of the
effective, seem (so far) not very interesting to mathematicians working in
other areas. One can point to them, but there doesnt seem to be much to
say about them. Like inaccessible ordinals, their very inaccessibility
makes them mathematically inert and unrelated to the items and methods
that currently drive research. Thus logicians may have a great deal to teach
number theorists (and geometers) about the tacit assumptions that guide
their choices about what things, procedures, and methods to canonize; and
 Fermats Last Theorem and the Logicians 161

the interaction between logic and number theory, for example, may give
rise to novel objects, procedures and methods still to be discovered.

Acknowledgements
I would like to thank Colin McLarty, Wen Ching (Winnie) Li, Dirk
Schlimm, Jessica Carter, Norbert Schappacher, Herbert Breger, Donald
Gillies, Karine Chemla, Ivahn Smadja, Angus Macintyre, Chiara
Ambrosio, Franois De Gandt, Dominique Bernardi, Emiliano Ippoliti,
and Carlo Cellucci for their various contributions to the evolution of this
paper. I also thank the Department of Philosophy and the College of the
Liberal Arts at the Pennsylvania State University, as well as the Ville de
Paris, the research group REHSEIS / SPHERE, and the Cit Internationale
Universitaire de Paris, for supporting my sabbatical research (2011-12),
which includes this essay.

References
Altman, A. and Kleiman, S. (1970). An Introduction to Grothendieck
Duality Theory. New York: Springer.
Breuil, C., Conrad, B., Diamond, F., Taylor, R. (2001). On the modularity
of elliptic curves over . Journal of the American Mathematical
Society, 14: 843-939.
Cellucci, Carlo (1998). Le regioni della Logica. Rome: Editori Laterza.
. (2002). Filosofia e matematica.Rome: Editori Laterza.
. (2007). La filosofia della matematica del Novecento. Rome: Editori
Laterza.
Cornell, G., Silverman, J., Stevens, G. (1997). Modular Forms and
Fermats Last Theorem. New York: Springer.
Darmon, H., Diamond, F., Taylor, R. (1997). Fermats Last Theorem.
Conference on Elliptic Curves and Modular Forms, Dec. 18-21, 1993
(Hong Kong: International Press): 1-140.
Freyd, P. (1964). Abelian Categories. New York: Springer.
Grosholz, Emily (2007). Representation and Productive Ambiguity in
Mathematics and the Sciences. Oxford: Oxford University Press.
Hartshorne, Robin. (1977). Algebraic Closed Fields. New York: Springer.
Kisin, M. (2009a). Modularity of 2-adic Barsotti-Tate representations.
Inventiones Mathematicae, Vol. 178, No. 3: 587-634.
. (2009b). Moduli of finite flat group schemes, and modularity. Annals
of Mathematics, 170(3): 1085-1180.
Li, Wen-Ching (2001 / 2012). Class Notes, Department of Mathematics,
162 Chapter Nine

The Pennsylvania State University.

Macintyre, A. (2006). The Impact of Gdels Incompleteness Theorems on
Mathematics. Kurt Gdel and the Foundations of Mathematics:
Horizons of Truth (Proceedings of Gdel Centenary), Vienna): 3-25.
Mazur, B. (1977). Modular Curves and the Eisenstein ideal. Publications
Mathmatiques, Institut des Hautes tudes Scientifiques, Vol. 47: 133-
186.
McLarty, C. (2012). A Finite Order Arithmetic Foundation for Cohomology.
Forthcoming.
. (2012). What Does It Take to Prove Fermats Last Theorem?
Grothendieck and the Logic of Number Theory. The Bulletin of
Symbolic Logic, Vol. 16, No. 3: 359-377.
Ribet, K. (1995). Galois Representations and Modular Forms. Bulletin of
the American Mathematical Society, Vol. 32, No. 4, Oct. 1995: 375-
402.
Schlimm, D. (2012). Analyzing Analogies in Mathematical Domains.
Forthcoming.
Wiles, A. (1995). Modular elliptic curves and Fermats Last Theorem.
Annals of Mathematics, Vol. 142: 443-551.
 CHAPTER TEN

CHRISTIAAN HUYGENSS
ON RECKONING IN GAMES OF CHANCE:
A CASE STUDY ON CELLUCCIS HEURISTIC
CONCEPTION OF MATHEMATICS

DANIEL G. CAMPOS

SUMMARY Carlo Cellucci has argued in favor of a philosophical

conception of mathematics as an open-ended, heuristic practice and
against the foundationalist view of mathematics as a closed-ended body
of knowledge that is completely determined by self-evident axioms. In
this essay, I examine Celluccis heuristic conception of mathematics by
way of a case study, namely, Christiaan Huygenss On Reckoning in
Games of Chance (1657). I argue that the heuristic conception provides an
insightful way to understand Huygenss approach and methods of
hypothesis-making to solve mathematical problems in games of chance.
Moreover, this case provides some corroboration for Celluccis heuristic
conception of mathematics. However, I also appeal to a brief example
from the history of mathematics to raise some questions about Celluccis
view that axioms never have a heuristic function.

KEYWORDS analytic method, axioms, heuristics, mathematical

discovery, mathematical hypotheses.

Carlo Cellucci has argued in favor of a philosophical conception of

mathematics as an open-ended, heuristic practice and against what he calls
the foundationalist view of mathematics as a closed-ended body of
knowledge that is completely determined by self-evident axioms (Cellucci
2000 and 2002). In his essay The Growth of Mathematical Knowledge:
164 Chapter Ten

An Open World View, Cellucci rejects the closed world viewadvanced

by Kant, Frege, and Hilbertaccording to which the method of mathematics
is the axiomatic method and, in order to rise to the level of science,
mathematics must be organized as an axiomatic system. Accordingly, there
are only two ways in which mathematical knowledge can grow, namely,
by deriving new results from given axioms or by introducing new axioms
(Cellucci 2000, p. 155-156). At the heart of this view there is the urge to
guarantee that mathematical theories be absolutely certain (Ibid., p. 157).
As characterized by Cellucci this tradition in the philosophy of
mathematics indeed fits within what John Dewey called the quest for
certainty with respect to the broader modern epistemological tradition. In
order to ensure absolute certainty for mathematics, Kant, Frege and Hilbert
present it as a closed system, depending on ultimate principles that are
either immediately evident or can be justified by indisputable means
(Ibid., p. 158).
Cellucci, however, rejects the closed world view both by pointing out
the consequences of Gdels incompleteness results for it and by arguing
that this view is contradicted by actual mathematical practice (Ibid., p.
158-162). With regard to the latter criticism, he points out that the closed
world view disregards all aspects of mathematical knowledge but one,
namely, the fact that it consists of propositions that can be viewed as
related by the consequence relation (Ibid., p. 162). He then concludes:
Since such relation is static, atemporal, and acontextual, it is no wonder
that the closed world view provides an essentially meagre picture of
mathematical knowledge which, while partially (though not entirely)
useful for pedagogical purposes, does not adequately account for the
richness of mathematical experience (Ibid.).
It is precisely Celluccis philosophical endeavor to account for
mathematical practice, mathematical experience, and especially
mathematical discovery that has drawn me to his work. To this end, in the
same essay Cellucci begins to sketch an open world view with regard to
mathematics. He argues that the analytic method better accounts for the
mathematical method, the dynamics of mathematical work, and the growth
of mathematical knowledge (Ibid.). Here he initially characterizes the
method as solving a problem by reducing it to another one, which is
provisionally assumed as a hypothesis and shown to be adequate to solve
the problem. Such a hypothesis in turn generates a problem to be solved in
much the same way, i.e., introducing a new hypothesis, and so on (Ibid.).
He briefly discusses three distinct forms of the analytic method from its
ancient origins: (i) Seeing the solution to a given problem by means of a
diagram; (ii) reductio ad impossibile where one proves A by assuming ~A
 Christiaan Huygenss On Reckoning in Games of Chance 165

and deriving B and ~B; and (iii) reduction of one problem to another one,
i.e. anagog (Ibid., p. 163-164). According to the open world view,
moreover, mathematical systems are open systems with the following
distinctive features: (i) they are based on the analytic method so that proof-
search begins with a given problem, not with an axiom; (ii) they do not
depend on permanent axioms but on provisional hypotheses; (iii) they are
dialoguing systems so that one cannot assume that solving a problem in a
particular mathematical field requires only concepts and methods of that
field; (iv) they are dynamic systems capable of dealing with changeable
states of affairs; and (v) they involve proofs intended as objects that are
evolutionary, plastic, and modular (Ibid., p. 162-163). In order to
understand the nature of mathematical practice, then, it is crucial to
characterize as carefully as possible the analytic method. In fact, while the
closed world view does not provide a rational account of the process
through which definitions, axioms, and proofs are found and in fact
assumes that no such account can be given, the open world view identifies
in the analytic method a general framework for understanding the process
of mathematical discovery (Ibid., p. 171-172).
Cellucci undertakes the task of accounting for mathematical discovery
via the analytic method in his book Filosofia e matematica (2002).1 Here
he develops a thorough critique of the closed, foundationalist conception
of mathematics in favor of the open, heuristic conception. He again argues
that the heuristic view reveals that mathematics is primarily problem-
solving rather than theorem-proving and, correspondingly, that the actual
method of mathematical inquiry is analytic instead of axiomatic.
Actual mathematical inquiry does not proceed by way of mechanical
deduction from self-evident principles and axioms. Some mathematical
theories might exhibit an axiomatic structure once they are developed and
mathematicians decide to order knowledge in that way, but at that point
they are dead, so to speakestablished, axiomatized theories are no
longer an actual, living matter of inquiry. Mathematical inquiry rather
proceeds by way of analytical problem-solving. According to Cellucci,
the analytic method is the procedure according to which one analyzes a
problem [that is, brakes it into constituent problems, or reduces it to
another problem, and so on] in order to solve it and, on the basis of such
analysis, one formulates a hypothesis. The hypothesis constitutes a
sufficient condition for the solution of the problem, but it is itself a
problem that must be resolved. In order to resolve it, one proceeds in the

1
An English translation of the introduction is available in Cellucci 2005.
166 Chapter Ten

same way, that is, one analyzes it and, on the basis of such analysis,
formulates a new hypothesis. [Thus, analysis] is a potentially infinite
process (Cellucci 2002, p. 174).2 Under this view, therefore, the search
for an absolute foundation to mathematical knowledge is vain. To cast
mathematical axioms as self-evident truths that serve as absolute
foundations for mathematical knowledge is to curtail the actual process of
analytical inquiry. Moreover, in as much as the analytic passage from the
given problem to a hypothesis that constitutes a sufficient condition for its
solution is configured as a reduction from one problem to another, the
analytic method is also called the method of reduction (Ibid., p. 175).
And in as much as the analytic method requires formulation of a
hypothesis for the solution of a problem, it is also called the method of
hypothesis (Ibid., p. 177). Analysis, then, consists in reasoning processes
that we might very broadly conceive as reduction and hypothesis-making.
Cellucci, moreover, discusses a mathematical tool-kit of heuristic methods
for hypothesis-making which include induction, analogy, figures,
generalization, particularization, hybridation, metaphor, metonymy, and
definition. This substantiates his thesis that the logic of mathematics is a
logic of discovery, not merely of justification, and that it is largely though
not exclusively non-deductive.
In this essay, I examine Celluccis heuristic conception of mathematics
by way of a case study, namely, Christiaan Huygenss On Reckoning in
Games of Chance (1657). I argue that his heuristic conception provides an
insightful way to understand Huygenss method and approaches to solve
mathematical problems in games of chance. As a general thesis, I argue
that Huygenss practice consists in problem-solving that can be described
by the analytic method and its heuristic strategies to search for hypotheses.
As a more specific thesis, I argue that Huygens employs the heuristic
methods of particularization, generalization, and reduction to solve one of
the main problems in his Reckoning. However, I also appeal to a brief
example from the history of mathematics to raise some questions about
Celluccis strong claim that axioms never have a heuristic function or
cannot be regarded as hypotheses.

2
All translations from this work are mine.
 Christiaan Huygenss On Reckoning in Games of Chance 167

1. The Case of Huygenss On Reckoning in Games

of Chance
After his1655 visit to Paris, Christian Huygens set to work on various
problems on the mathematics of chance that had been analyzed in
correspondence by Blaise Pascal and Pierre Fermat the previous year.
These problems included the problem of points and problems with dice
and, since he had no access to the correspondence, he solved them
independently of Pascal and Fermat. Such problems concern the fair
distribution among players of the total sum staked in games of chance
when the game is suspended before it ends. For the case where two players
are involved, for example, we might state the problem of points as follows:
Suppose that two players play a match such that, in order to win, one must
score n points before his opponent does. If they stop the match when
player 1 has won x < n points and player 2 has won y < n points, how
should the total sum that they staked in the match be divided?
Huygenss mathematical work related to the solution of such kinds of
problems is highly significant since, as F.N. David proclaims, [t]he
scientist who first put forward in a systematic way the new propositions
evoked by the problems set to Pascal and Fermat, who gave the rules, and
who first made definitive the idea of mathematical expectation was
Christianus Huygens (David 1962, p. 110). In his 1657 De Ratiociniis in
Aleae Ludo, Huygens put forth the first systematic treatment of the
mathematics of chance, and this work became the standard text for
studying the elements of the doctrine of chances. It was subject to various
English translations, one of them by John Arbuthnot, and Jacob Bernoulli
included it, with his own annotations, as part I of the Ars Conjectandi. The
main body of Huygenss De Ratiociniis in Aleae Ludo consists of the
following fourteen propositions:

I: To have equal chances of getting a and b is worth (a+b)/2.

II: To have equal chances of getting a, b or c is worth (a+b +c)/3.
III: To have p chances of obtaining a and q of obtaining b, chances
being equal, is worth (pa + qb) / (p + q).
IV: Suppose I play against an opponent as to who will win the first
three games and that I have already won two and he one. I want to
know what proportion of the stakes is due to me if we decide not to
play the remaining games.
V: Suppose that I lack one point and my opponent three. What
proportion of the stakes, etc.
VI: Suppose that I lack two points and my opponent three, etc.
168 Chapter Ten

VII: Suppose that I lack two points and my opponent four, etc.
VIII: Suppose now that three people play together and that the first and
second lack one point each and the third two points.
IX: In order to calculate the proportion of stakes due to each of a given
number of players who are each given numbers of points short, it is
necessary, to begin with, to consider what is owing to each in turn
in the case where each might have won the succeeding game.
X: To find how many times one may wager to throw a six with one die.
XI: To find how many times one should wager to throw 2 sixes with 2
dice.
XII: To find the number of dice with which one may wager to throw 2
sixes at the first throw.
XIII: On the hypothesis that I play a throw of 2 dice against an
opponent with the rule that if the sum is 7 points I will have won
but that if the sum is 10 he will have won, and that we split the
stakes in equal parts if there is any other sum, find the expectation
of each of us.
XIV: If another player and I throw turn and turn about with 2 dice on
condition that I will have won when I have thrown 7 points and he
will have won when he has thrown 6, if I let him throw first find
the ratio of my chance to his.3

I propose that Celluccis heuristic notion of mathematics, with its

account of the analytical method, explains well the type of mathematical
practice that Huygenss treatise reveals. There are no axioms serving as
the foundation of Huygenss De Ratiociniis in Aleae Ludo. There is rather
a series of propositions that actually stand for problems of chance and
expectation. In order to solve them, Huygens analyzes them, reducing
them to other problems and posing hypothetical solutions. The solution to
each problem, in turn, suggests new problems for investigation. The
analytical process, then, gradually leads to a deepening of knowledge on
the mathematics of chance. For example, even without discussing the
details here, we might easily imagine that the solution to the problem
stated in proposition VII could proceed by analyzing this problem into
those problems already solved in the immediately preceding propositions.
And as we shall see in detail shortly, proposition IX is a general problem

3
I have listed the propositions as translated in David 1962, p. 116-117. This is a
loose, modernized rendering of Huygenss propositions, but it will do for our
purposes here.
 Christiaan Huygenss On Reckoning in Games of Chance 169

that can be analyzed into simpler problems that are either of easy solution
or already solved in previous propositions, especially II and VIII.
Moreover, in his general treatment of the problem of points in proposition
IX Huygens assumes that all players have equal chances of winning each
game. This suggests a new, more general, problem: What if the players do
not have equal chances of winning each game? Abraham de Moivre took
up this problem and offered an even more general solution to the problem
of points in his 1718 Doctrine of Chances. We find in Huygenss treatise,
then, not an axiomatized theory but a series of interrelated problems
regarding the calculus of chance whose solutions eventually lead Huygens
to offer general rules for the solution of similar problems, such as the
general method for solving particular problems of points stated in
proposition IX. And the same analytical process is taken up by other
inquirers, so that the analytical method does tend towards increasingly
more general problems, potentially ad infinitum.
Proponents of the foundationalist view of mathematics as an affair of
deduction from self-evident axioms might of course deny that Huygenss
treatise is properly a mathematical work. Lorraine Daston, a prominent
historian of probability, in fact observes that even though the famous
correspondence between Blaise Pascal and Pierre Fermat first cast the
calculus of probabilities in mathematical form in 1654, many
mathematicians would argue that the theory achieved full status as a
branch of mathematics only in 1933 with the publication of A. N.
Kolmogorovs Grundbegriffe der Wahrscheinlichkeitsrechnung. Taking
David Hilberts Foundations of Geometry as his model, Kolmogorov
advanced an axiomatic formulation of probability based on Lebesgue
integrals and measure set theory. Like Hilbert, Kolmogorov insisted that
any axiomatic system admitted an unlimited number of concrete
interpretations besides those from which it was derived, and that once the
axioms for probability theory had been established, all further exposition
must be based exclusively on these axioms, independent of the usual
concrete meaning of these elements and their relations (Daston 1988,
p.3). Under such a foundationalist view, therefore, the work of all the early
probabilists, including Huygens, may be regarded as a non-mathematical,
even if scientific, attempt at providing quantified models of chance
phenomena, but not as mathematical theorizing proper. They may concede
Dastons own view that the link between model and subject matter is
considerably more intimate than that between theory and applications so
that, even in the eyes of the early probabilists, the field of mathematical
probability was a mathematical model of a certain set of phenomena,
rather thanan abstract theory independent of its applications (Ibid., p.
170 Chapter Ten

xii). Even conceding this, however, the foundationalists would not confer
upon early mathematical probability the seemingly privileged rank of a
theory.
To the foundationalists endorsement of the axiomatic method as the
only method of mathematical reasoning, and to Dastons seeming
agreement that only axiomatized theories are legitimate branches of
mathematics, Cellucci offers an explicit reply:

[T]he idea that the axiomatic method is available for the whole of
mathematics because all mathematical theories, when sufficiently
developed, are capable of axiomatization, contrasts with mathematical
experience, which shows that axiomatization does not naturally apply to all
parts of mathematics. Some of them are not suitable for axiomatization,
and exist as collections of solved or unsolved problems of a certain kind.
This is true, for example, of number theory and of much of the theory of
partial differential equations. (Cellucci 2005, p. 25)

The same might be said of early mathematical probability. Though it was

not yet axiomatized, it did stand as a collection of solved and unsolved
problems, and the solution of each problemby way of the characteristic
methods of mathematical reasoningled to the recognition of new ones.
Moreover, I think that from Celluccis heuristic perspective the
distinction between model and theory may be of philosophical interest for
understanding some ways of characterizing the structure of mathematical
and scientific knowledge, but it is not relevant for determining whether the
early probabilists were acting and reasoning as mathematicians.
Foundationalist philosophers of mathematics may impose their conceptions
of mathematics on early mathematical probability in order to argue as
much as they want about whether it is a model or a theory. However, from
an open-ended, heuristic perspective, what marks the probabilists
reasoning as genuinely mathematical is that they were posing problems
and seeking hypotheses to solve them. The hypotheses themselves are the
conditions for the solution of the given problems. Whether these problems
as solved amounted structurally to models or to theories is beside our point
of interest.
Nevertheless, let me state that I think that no deeper understanding of
mathematics is gained by arbitrarily circumscribing the notion of
mathematical theory to axiomatized systems of propositions. If anything, it
promotes the erroneous idea that mathematics is the dead stuff printed in a
 Christiaan Huygenss On Reckoning in Games of Chance 171

certain kind of textbook.4 My inclination is to say that a mathematical

theory is a purely ideal system while a mathematical model is a system
that represents an actual problematic phenomenon. Borrowing Peircean
terms, a theory is a pure icon while a model is a symbolic icon. Qua
pure mathematicians, the early probabilists were creating a theory; qua
applied mathematical scientists, they were modeling aleatory phenomena.
Be that as it may, what is crucial to us is that the ideal systems of early
mathematical probability were open-ended and subject to reconception and
revision, as problem-solving demanded and as mathematical theorizing
and the modeling of actual chance phenomena dictated. Whether
theorizing or modeling, their activity was thoroughly mathematical, and it
proceeded by problem-solving and hypothesis-making. Huygenss work
testifies to this, as we shall now see.

2. Generalization and Particularization as Analytical

Heuristics
Proposition IX provides a general rule for the solution of the problem of
points. Let me first expound Huygenss demonstration and then discuss
what it reveals about analytical heuristics. Again, the proposition is the
following:

In order to calculate the proportion of stakes due to each of a given number

of players who are each given numbers of points short, it is necessary, to
begin with, to consider what is owing to each in turn in the case where
each might have won the succeeding game.

4
Preferably one without any figures, actual diagrams, pictures, conjectures or wild
guesses. See, for instance, James Robert Browns discussion of the Bourbaki group
in French mathematics, which equates the highest standards of rigor with a
thorough refusal to use any pictures or figures or other heuristic aides in their
demonstrations (Brown 1999, p. 172-173). In this regard, Cellucci reveals the
ironies of claiming to banish diagrams from mathematical reasoning when he
writes: While the closed world view considers the use of diagrams as redundant
because it assumes that results established using diagrams can always be derived
from given axioms by logical deduction only (at least in principle; in practice
diagrams occur on almost every page of Hilberts Grundlagen der Geometrie),
seeing the solution by means of a diagram is a self-contained procedure that is not
part of a global axiomatic order (Cellucci 2000, p. 163).
172 Chapter Ten

To demonstrate it, Huygens reasons as follows.5 (I will insert my

annotations in parentheses.) He supposes that there are three players, A, B,
and C, and that A lacks one game, B two games, and C two games in order
to win the match. (That is, he begins by considering a particular problem
of points.) He begins by trying to find the proportion of stakes due to B,
calling the sum of stakes q, if either A, or B himself, or C wins the first
succeeding game. There are, therefore, three cases to consider.
(a) If player A were to win the next game, then the match would end
and consequently the sum due to B is 0 (i.e. B is due 0q).
(b) If B were to win the next game, he would therefore lack 1 game,
while A and C would still lack 1 and 2 games respectively. Therefore, by
proposition VIII, B is due 4q/9. (Alternatively, following Fermats
reasoning in his 1654 correspondence with Pascal, imagine a table of
equipossible outcomes for the ensuing situation, such as this one:

aaa bbb ccc

abc abc abc
AAA BBB ABC

The match must be decided in at most two games, and each column shows
a possible sequence of game winners and the resulting match winner. For
example, the first column shows that player A wins the next two games
and therefore the match, granting that the second game would actually be
unnecessary to win the match. The second column shows that player A
wins the first game and therefore the match, so even if B were to win the
second game it would be superfluous. The last column shows player C
winning the next two games and therefore the match. Following this
reasoning, there would be 4 out of 9 possible outcomes that would favor
player B. Huygens does not construct a table, and his problem-solving
approach is different; however, the exercise based on Fermats approach
allows us to understand the result.)
(c) Lastly, if C were to win the next game, then he would lack 1 game,
while A and B would still lack 1 and 2 games respectively. Consequently,
by proposition VIII, B is due 1q/9. (Again imagine a table of equipossible
outcomes for the ensuing situation. There would be only one out of nine
possible outcomes that would favor player B.)

5
My rendition of Huygenss reasoning is a loose translation of his demonstration
as reprinted in Bernoulli 1713, p. 18-19.
 Christiaan Huygenss On Reckoning in Games of Chance 173

Moreover, if we colligate in one summation, that is, if we add, that

which in each of the three cases is due to B, namely 0, 4q/9, and 1q/9, the
result is 5q/9. Dividing this sum by 3, which is the number of players, the
result is exactly 5q/27. By proposition II this is the part sought, that is,
the proportion of the total stakes that is due to B. (Had we diagrammed a
table for the particular version of the problem of points that Huygens
considers, we would have found that there are twenty-seven equipossible
outcomes, out of which only five outcomes favor player B.) As if to
elucidate completely his reasoning, Huygens restates his conclusion that
since B would obtain either 0, 4q/9, or 1q/9, then by proposition II the
proportion of stakes due to B is 0 + 4q/9 + 1q/9 : 3 or 5q/27.
(At this point, Huygens derives a general rule for solving the problem
of points from the foregoing solution to one particular version of the
problem of points.) Therefore, Huygens argues, one must consider in any
problem whatsoever, clearly in the preceding one or in any other version
of the problem, what is due to each player in the case where each might
win the next game. (In the previous particular problem, we would find by
the same method that A is due 17q/27 and C is due 5q/27.) For just as one
cannot solve the preceding problem until we subduce it under the
calculations already done for proposition VIII, so also we cannot solve the
problem in which the three players lack 1, 2, and 3 games respectively
until we calculate how the stakes ought to be distributed when: (i) they
lack 1, 2, and 2 games respectively, which is the preceding problem just
solved, and (ii) they lack 1, 1, and 3 games respectively, which is the
problem already solved in proposition VIII. (Note that when (iii) they lack
0, 2, 3 games respectively, the solution is trivial since A gets all of the
stakes. This is why Huygenss does not list it.) Huygens provides a table
that comprehends the calculations for each subsequent particular
problem of points, up to the problem in which A, B, and C lack 2, 3, and 5
games respectively, noting that the particular solutions can be extended.
(By providing the table, Huygens emphasizes that his general rule will
work no matter how complex the particular problem of points under
study.)
Allow me to draw out now what Huygenss reasoning reveals about
mathematical problem-solving via analytical heuristics. After stating the
proposition or problem to be solved, Huygens provides a particularization
of the problem of points. Cellucci defines heuristic particularization as
the inference by way of which one passes from one hypothesis to another
one that it contains as a particular case (Cellucci 2002, p. 267). We might
state the general problem of points as follows: Given that players A, B,,
X, Y, Z lack a, b,, x, y, z points respectively to win the match, find the
174 Chapter Ten

proportion of the total stakes q that is due to each one of them. Huygenss
finds that trying to find a general rule of solution directly from this general
statement of the problem is too difficult. Thus he particularizes the general
problem and proceeds to solve the particular version.
Next Huygens experiments by analyzing the particular problem into
three alternative diagrams of problems that have already been solved.
Reduction in this sense simply means to resolve the present problem
into one or more alternative problems whose solutions, when composed or
linked in some suitable way, are sufficient for solving the original one. In
this case, Huygens reduces the problem in which players A, B, and C lack
1, 2, and 2 games respectively into three alternative problems: how to
divide the stakes when (a) they lack 0, 2, and 2 games; (b) they lack 1, 1,
and 2 games; and (c) 1, 2, and 1 games. Case (a) has a trivial solution, and
cases (b) and (c) have already been solved in proposition VIII.
Additionally, proposition II provides the rule by which the original
problem can be solved in terms of the solutions to cases (a), (b), and (c).
Finally, Huygens grasps that the method of solution is generalit
can be applied to any particular problem, and it will lead to the correct
solution. Equivalent modifications of the original problem in any play
situation will yield the correct response regarding the fair distribution of
stakes. Cellucci defines heuristic generalization as the inference by way
of which one passes from one hypothesis to another one that contains it as
a particular case (Ibid., p. 267). I submit that Huygens grasps the
generality of the rule quickly due to his vigorous power of generalization.
Any mathematician with a lesser power of generalization, however, could
arrive at the same generalization by conducting other experiments. The
mathematician could experiment with problems in which there are, say,
four players that lack 1, 1, 1, and 2 points. She could resolve this play
situation by the same method into the various possible alternative
problems. Still she would find that Huygenss general rule works.
Accordingly, Huygens emphasizes the generality of his method by
providing a table with the solution to more complex games. No matter
how complex the problem, his general method works, and his readers can
confirm it by conducting alternative experiments themselves.

3. Final Remarks and Considerations

Against the foregoing account of Huygenss reasoning as analytical
problem-solving, however, it may be objected that in the Reckoning
Huygens develops his investigations syntheticallyin the sense that he
proceeds from what is known to what is to be demonstrated instead of
 Christiaan Huygenss On Reckoning in Games of Chance 175

reasoning analytically in the sense of Pappusthat is, reasoning from the

assumption that an unknown is given and deducing a necessary conclusion
from which that unknown can be determined (see Boyer and Merzbach
1991, p. 305). Thus, Huygens establishes propositions I through III
without positing any unknowns, and he can solve the problems stated in
propositions IV through VIII always proceeding synthetically from what is
known. However, this objection would serve to clarify that by analysis
Cellucci means something differentnot analysis in the sense of Aristotle
or Pappus but in the sense he attributes to Hippocrates of Chios and Plato
(Cellucci 2005, p. 204; see Cellucci 1998). Again, he defines the analytic
method as the procedure according to which one analyzes a problem in
order to solve it and, on the basis of such analysis, one formulates a
hypothesis. The hypothesis constitutes a sufficient condition for the
solution of the problem, but it is itself a problem that must be resolved
(Cellucci 2002, p. 174). As I pointed out, the heuristic procedures involved
in this sense of analysis are most often non-deductive. Moreover, even
though the order of exposition of the Reckoning is synthetic, if we focus
on the reasoning involved in proposition IX, we notice that the general
method it both proposes and deploys for the solution of the problem of
points is analytic in Celluccis sense. It proposes that more complex games
must be resolved into simpler games whose solution is known or that can
themselves be resolved into even simpler games whose solution can be
determined.
Overall, I have proposed that in the course of this demonstration
Huygens reasons analytically in Celluccis sense, deploying the heuristic
experimental techniques of particularization, reduction, and generalization
to solve a problem and, consequently, to demonstrate necessarily what
originally stood as a hypothetical proposition. Thus, the heuristic conception
provides an insightful way to understand Huygenss approach and
hypothesis-making methods to solve mathematical problems in games of
chance. I also find in this example a helpful illustration of Celluccis view
that mathematics is a heuristic practice, that the method of mathematical
inquiry is analytical, and that particularization, reduction, and generalization
are among the key heuristic techniques of research that mathematicians
deploy in the course of problem-solving. Beyond illustrating Celluccis
view, moreover, this case provides some corroboration for it. As I have
suggested, the strictures of the foundationalist view lead to the claim that
the work not only of Huygens, but of Pascal and Fermat before him and
Bernoulli, De Moivre, and Laplace after him, did not properly belong to
mathematics since the outcome of their investigations was not an
axiomatized theory. However, such a claim must be rejected in the face of
176 Chapter Ten

the actual nature of mathematical practice and the evidence that the history
of mathematics affords us of such a practice. What mathematicians have
done historically and what they do todayand not any particular set of
conceptual constraints on what the logical form of a mathematical theory
ought to bedefines what mathematics is. In Celluccis account, what
mathematicians do is analytical problem-solving, and the case of
Huygenss On Reckoning in Games of Chance provides evidence for such
a heuristic view.
Nonetheless, I would like to question what appears to be Celluccis
strong claim that axioms never have a heuristic function and cannot be
regarded as hypotheses. What about the possibility that the axiomatic
method may provide a strategy for finding proofs and thus solving
problems? Cellucci rejects it outright by appealing to mathematical
experience:

The idea that the axiomatic method provides a strategy both for finding and
remembering proofs also contrasts with mathematical experience, which
shows that proofs based on the axiomatic method often appear to be found
only by a stroke of luck, and seem artificial and difficult to understand.
Showing only the final outcome of the investigation, established in a way
that is completely different from how it was first obtained, such proofs
hide the actual mathematical process, thus contributing to make
mathematics a difficult subject. (Cellucci 2005, p. 25)

I agree completely in that presenting mathematics as the result of deductive

demonstration from axioms or first principles completely obscures the
nature of mathematical reasoning. In fact, it is such a mode of presentation
that often kills students imaginations and interest in mathematics
(Campos 2010).
However, I think that Cellucci dismisses too strongly the possibility
that axioms themselves may be treated as framing hypothesesthat is, as
hypotheses that frame a general state of things for mathematical
investigationand that exploring the consequences of such framing
hypotheses may lead to mathematical discovery also. This does not mean
that all framing hypotheses are axioms. An example is the idea of the
fundamental probability set which served as the framing hypothesis of
early investigations into mathematical probability without being an axiom
(Campos 2009). Early probabilists framed the concept of a collection of
equipossible outcomes for chance situations in order to solve problems
pertaining to games of chance. The possibility does mean, however, that
axioms may be regarded as framing hypotheses and that by modifying or
transforming these axioms mathematical discovery may follow. An example
 Christiaan Huygenss On Reckoning in Games of Chance 177

is the case of non-Euclidean geometries. The fifth postulate of Euclids

Elements states that if a straight line falling on two straight lines make the
interior angles on the same side less than two right angles, the two straight
lines, if produced indefinitely, meet on that side on which are the angles
less than the two right angles (Euclid 1956, vol. 1, p. 155). According to
the foundationalist view, if the postulate is false, Euclidean geometry is
false, and a new, true geometrical theory must be constructed on the
basis of true axioms. However, from a heuristic standpoint this position
would appear completely inadequate to the inquiring practice of the
mathematicians. There is nothing foundational about postulates. Rather, by
rejecting the fifth postulate of Euclidean geometry mathematicians were
able to discover new geometries. My suggestion is that the alternative
postulates can be regarded as alternative framing hypotheses; their
function is heuristic in the sense of framing a general state of things for
geometrical investigation. According to this view, in modifying the fifth
postulate, a mathematician is simply changing a framing hypothesis and
re-conceiving a mathematical state of affairs, so to speak. Thus, there is no
need to claim that, qua mathematics, the Euclidean geometry is false while
the non-Euclidean geometries are true. They are theories that we can
investigate through analysis. In general mathematicians do not set out to
deduce closed theories on the basis of axioms; they set out to explore what
follows from provisional hypotheses, and axioms may be regarded as such
hypotheses also. It is puzzling to me, therefore, that Cellucci seems to
reject so strongly the possibility that axioms may play any heuristic role in
mathematical inquiry. This possibility enriches, rather than challenges, the
open, heuristic conception of mathematics that he so aptly advances in his
work.

References
Bernoulli J. (1713). Ars Conjectandi. Basil: Thurnisiorum.
Boyer C. and Merzbach U. (1991). A History of Mathematics. Second
Edition. New York: John Wiley and Sons.
Brown J.R. (1999). Philosophy of Mathematics: An Introduction to the
World of Proofs and Pictures. New York: Routledge.
Campos, D.G. (2009). The Framing of the Fundamental Probability Set: A
Historical Case Study on the Context of Mathematical Discovery.
Perspectives on Science, 17 (4): 385-416.
. (2010). Peirces Philosophy of Mathematical Education: Fostering
Reasoning Abilities for Mathematical Inquiry. Studies in Philosophy
and Education, 29(5): 421-439.
178 Chapter Ten

Cellucci C. (1998). Le ragioni della logica. Bari, Italy: Laterza.

. (2000). The Growth of Mathematical Knowledge: An Open World
View. In: Grosholz E. & Breger H., eds. The Growth of Mathematical
Knowledge. Dordrecht, Netherlands: Kluwer Academic Publishers:
153-176.
. (2002). Filosofia e matematica. Bari, Italy: Laterza.
. (2005). Introduction to Filosofia e matematica. In: Hersh R., ed. 18
Unconventional Essays on the Nature of Mathematics. New York:
Springer: 17-36.
Daston L. (1988). Classical Probability in the Enlightenment. Princeton,
New Jersey: Princeton University Press.
David F.N. (1962). Games, Gods and Gambling: A History of Probability
and Statistical Ideas. London: Charles Griffin & Co.
De Moivre A. (1718). The Doctrine of Chances, or a Method of
Calculating the Probability of Events in Play. London: W. Pearson.
Euclid (1956). The Thirteen Books of Euclid's Elements. Heath T., ed.
New York: Dover.
Huygens C. (1657). Ratiociniis in aleae ludo. In: Van Shooten F.,
Exercitionum Mathematicorum. Amsterdam: J. Elsevirii.
Kolmogorov A.N. (1933). Grundbegriffe der Wahrschein-
lichkeitsrechnung. Berlin: J. Springer.
 CHAPTER ELEVEN

NATURAL MATHEMATICS
AND NATURAL LOGIC

LORENZO MAGNANI

SUMMARY In his article Why Proof? What is a Proof? (2008) Carlo

Cellucci focuses on so-called natural mathematics and natural logic,
centering on mathematics and logic embodied in organisms as a result of
natural selection, and stressing the different role of artificial mathematics
and logic, that is, mathematics and logic as disciplines. I will provide
further insight into this interplay. The first issue is related to the
importance of increasing logical knowledge of abduction: Cellucci himself
clearly shows how the study of abduction helps us to extend and
modernize the classical and received idea of logic, understood merely as a
corpus of axiomatic systems. The second refers to some ideas deriving
from so-called distributed cognition and concerns the role of logical
models as forms of cognitive externalizations of preexistent informal
human reasoning performances. In this perspective, natural mathematics
and logic are seen as constitutively intertwined with their artificial
counterparts. Logical externalization in objective formal systems,
communicable and sharable, is able to grant stable perspectives endowed
with symbolic, abstract, and rigorous cognitive features.

KEYWORDS Natural logic, natural mathematics, abduction, distributed

cognition, axiomatic systems.

1. Introduction
In his article Why Proof? What is a Proof? (2008), Carlo Cellucci
stresses the importance of so-called natural mathematics and logic,
centering on mathematics and logical embodied in organisms as a result of
natural selection, and focusing on the different role of artificial
180 Chapter Eleven

mathematics and logic, that is, mathematics and logic as disciplines. I

will provide further insight into this intellectual interplay. The first issue is
related to the importance of increasing logical knowledge of abduction:
Cellucci himself clearly shows how studies on abduction help us to extend
and modernize the classical and received idea of logic, understood simply
as a corpus of axiomatic systems. The second refers to some ideas deriving
from the area of so-called distributed cognition and concerns the role of
logical models as forms of cognitive externalizations of preexistent
informal human reasoning performances. In this perspective, natural
mathematics and logic are constitutively intertwined with their artificial
counterparts. Logical externalization in formal objective systems,
communicable and sharable, can grant stable perspectives endowed with
symbolic, abstract, and rigorous cognitive features. For example, this
character of stability and objectivity, provided by logical axiomatic
achievements, is not present in models of abduction that are merely
cognitive and epistemological and, moreover, they remain central to
computational implementation.
Cellucci says: One may then distinguish a natural mathematics, that
is, the mathematics embodied in organisms as a result of natural selection,
from artificial mathematics, that is, mathematics as a discipline. (Devlin,
2005, p. 249) calls artificial mathematics abstract mathematics, but
artificial mathematics seems more suitable here since it expresses that it
is a mathematics that is not a natural product, being not a direct result of
biological evolution but rather a human creation (Cellucci, 2008, p. 16).
Natural mathematics is based on natural logic, which is that natural
capacity to solve problems that all organisms possess and is a result of
biological evolution. On the other hand, artificial mathematics is based on
artificial logic, which is a set of techniques invented by organisms to solve
problems and is a result of cultural evolution.
As I noted about ten years ago in my book Abduction, Reason, and
Science (Magnani, 2001) Peirce clearly indicated the importance of logic
(first order syllogisms) for grasping the inferential status of abduction, at
the same time creating a wonderful new broad semiotic view accompanied
by the well-known philosophical commitment to the new vision of
pragmatism. Given the restricted scope - classical, in terms of first order
syllogisms - of the logical tools available to him, the logical framework
depicted abduction merely as the well-known fallacy of affirming the
consequence. Recently, Alisedas book (2006) provided a summary of
the results that derive from this fundamental Peircean logical tradition: it
presents numerous recent logical models of abduction which are clearly
 Natural Mathematics and Natural Logic 181

illustrated in their rigorous demonstrative frameworks, that is in their

artificial logics.

2. Model-based reasoning in demonstrative frameworks

It is well known that the kind of reasoned inference that is involved in
selective and creative abduction1 goes beyond the simple relationship that
exists between premises and conclusions in valid classical deductions,
where the truth of the premises guarantees the truth of the conclusions, but
also beyond the relationship that exists in probabilistic reasoning, which
renders the conclusion merely more or less probable.
By contrast, we can see selective and creative abduction as being
formed by the application of heuristic procedures that involve all kinds
of good and bad inferential actions, and not only the mechanical
application of rules. It is only by means of these heuristic procedures that
the acquisition of new truths is guaranteed. Peirces mature view on
creative abduction as a kind of inference2 also seems to stress the
strategic component of reasoning.
Many researchers in the field of philosophy, logic, and cognitive
science have maintained that deductive reasoning (which resorts to
artificial logic, in Celluccis (2008) terms) also consists of the
employment of logical rules in a heuristic manner, even maintaining the
truth-preserving character: the application of the rules is organized in a
way that is able to recommend one particular course of action instead of
another. Moreover, very often the heuristic procedures of deductive
reasoning are in turn performed by means of an in-formal model-based
abduction.3 So humans apply rudimentary natural abductive/strategic

1
Epistemologically selective abduction occurs when we reach a hypothesis among
to use a word of the logical tradition also exploited by Aliseda already
available abducible hypotheses (as for instance in the case of a medical
diagnosis). Creative abduction occurs when, through our reasoning processes, we
are able to create new abducibles, which can be tested and added to [or which
can replace] the available ones (Magnani, 2001).
2
From Peirces philosophical point of view, all thinking is in signs, and signs can
be icons, indices or symbols. Moreover, all inference is a form of sign activity,
where the word sign includes feeling, image, conception, and other
representation (Peirce, 1931-1958, 5.283), and, in Kantian terms, all synthetic
forms of cognition. That is, a considerable part of thinking activity is model-based.
3
In general, model-based abduction takes advantage of internal (or of suitably re-
internalized external) models that are not merely symbolic/propositional but which
182 Chapter Eleven

ways of reasoning in formal deduction too. The most common example of

a strategic process that leads to the formation of new hypotheses (creative
abduction) is the common experience people have of solving problems in
geometry in a model-based way, trying to devise proofs using diagrams
and illustrations: of course the attribute of creativity we give to the
abduction in this case does not mean that it has never been made before by
anyone or that it is original in the history of a branch of knowledge (in this
particular case the ancient Greek geometers were the real creative
abducers!).
Hence, we must say that a kind of in-formal model-based abductions
also operate in deductive reasoning performed by humans who use
artificial logical systems. Following Hintikka and Remess analysis
(1974), proofs of general implication in first order logic need to use
instantiation rules by which new individuals are introduced, so they are
ampliative. In ordinary geometrical proofs, auxiliary constructions are
present in terms of conveniently chosen figures and diagrams. In Beths
method of semantic tableaux the strategic ability enacted by humans to
construct impossible configurations is undeniable (Hintikka, 1998;
Niiniluoto, 1999). Aliseda (2006) also provides interesting uses of semantic
tableaux as constructive representations of theories, where for example,
abductive expansions and revisions, derived from the belief revision
framework, operate over them. In the case of tableaux, their symbolic
character is certainly fundamental, but it is particularly clear that they are
also model-based configurations of proofs externalized through
suitable notations.4
Following Hintikka, we can say that the ground floor of deductive
reasoning, first-order logic, is none other than operating with certain
models or approximations of models, as is simply demonstrated by some
fundamental techniques such as Beths semantic tableaux. It is important
to note that Hintikka is perfectly aware of the double character of these
models, internal (more natural, in terms of mental models) and/or
external (artificially built for the occasion):
These models can be thought of as being mental, or they can be taken to

for example exploit diagrams, visualization, configurations, schemes, thought

experiments, and so on (Magnani, 2001).
4
It is worth noting that the semantic tableaux method provides further insight into
the problem of theory evaluation, intrinsic to abductive reasoning. In chapters six
and eight, Aliseda (2006) shows how semantic tableaux can deal with causal
aspects of abductive reasoning that cannot be considered with the help of the logic
programming tradition alone.
 Natural Mathematics and Natural Logic 183

consist of sets of formulas on paper or in this day and age perhaps rather
on the screen and in the memory of a computer. In fact, from this
perspective all rules of logical inference obviously involve mental
models. Johnson-Lairds discovery hence does not ultimately pertain to
the psychology of logic. It pertains, however confusedly, to the nature of
logic itself. The most basic deductive logic is nothing but experimental
model construction (Hintikka, 1997, pp. 69-70).

In this way Hintikka rejoins the distributed cognition approach to logic

I also stressed in Magnani (2005, 2009), where the interplay between
internal and external (as kinds of semiotic anchors symbolic, in this
case, Magnani 2006b) aspects of logical reasoning are illustrated. For
example, the role of strategies of experimental (counter) model-construction
in logical deduction is stressed, as is the importance of the introduction of
the right new individuals by means of existential instantiation to be
introduced into the model. The most important strategic question in
deductive reasoning is to determine in what order the instantiations are
to be treated. In classical geometrical reasoning the role of existential
instantiation is obvious and occurs through the iconic so-called auxiliary
constructions, which involve conceptually manipulating a configuration
of geometrical objects and extending it by introducing new individuals.
The possible creative character is reflected, for example, in the fact that
there is not always a mechanical (recursive) method for modeling these
human deductive performances. Of course, as Aliseda shows in chapter
four Abduction as computation (2006), a suitable computational
counterpart can take advantage of algorithms which render mechanical the
suitably chosen reasoning processes, and so suitable for implementation in
a computational program.

2.1. Formal Logical Deduction as an Optical Illusion

The logical tradition of Frege and Russell rejected all reasoning that
had been made in terms of geometrical icons as being responsible for
introducing an appeal to intuition. On the contrary, I am highly inclined to
agree with Hintikka, who maintains that the traditional idea of logical
reasoning as a discursive process is wrong, it is an optical illusion,
because all deduction is a form of experimental model construction that
follows that interplay between internal and external representations
already described. It is important instead to note that for instance already
at the level of elementary geometry:
184 Chapter Eleven

[] geometrical figures are best thought of as a fragmentary notation for

geometrical proofs alternative to, but not necessarily intrinsically inferior
to, the purely logical notation of formalized first order logic. [] They
are intrinsic features of certain deductive methods. They are part of the
semantics of logical reasoning, not only of its psychology or its heuristics.
If it is suggested that heuristic ways of thinking are needed to make
mathematical reasoning intuitive, I will borrow a line from Wittgensteins
Tractatus 6.233 and say that in this case the language (notation) itself
provides the intuitions (Hintikka, 1997, p. 73).

Moreover, in the case of human performances, in many forms of

deductive reasoning there are no trivial and mechanical methods of
making inferences; we must use models and heuristic procedures that
refer to a whole set of strategic principles. All the more reason for
Bringsjord (1998) to stress his attention to the role played by a kind of
model-based deduction that is part and parcel of our establishing
Gdels first incompleteness theorem, showing the model-based nature of
this great abductive achievement of formal thought.5

3. External and Internal Representations

3.1 Logic Programs as Agents: External Observations
and Internal Knowledge Assimilation

It is in the area of distributed cognition that the importance of the

interplay between internal and external representations has recently
acquired importance (cf. for example Clark, 2003, and Hutchins, 1995).
This perspective is particularly coherent with the so-called agentbased
framework (Magnani, 2006a and 2009, chapter seven). It is interesting to
note that direct attention to the agentbased nature of cognition and to the
interplay between internal and external aspects can also be found in the
area of logic programming, which is one of the two main ways the other
is the semantic tableaux method as already mentioned of logically and
computationally dealing with abduction.

5
Many interesting relationships between model-based reasoning in creative
settings and the related possible deductive dynamic logical models are analysed
in Meheus (1999) and Meheus and Batens (2006). Dynamic logic is also related to
the formal treatment of inconsistencies. Cf. also Cellucci (1998), who illustrates
Gdels implicit acknowledgment of the aforementioned optical illusion.
 Natural Mathematics and Natural Logic 185

I think that in logic programming a new idea of logic contrasted with

the classical idea arises, which certainly opens to abduction the door that
grants access to its full treatment through artificial logical systems.
Indeed, logic programs can be seen in an agent-centered, computationally-
oriented and purely syntactic perspective. Already in 1994 Kowalski
(1994) in Logic without model theory introduced a knowledge assimilation
framework for rational abductive agents, to deal with incomplete
information and limited computational capacity.
Knowledge assimilation is the assimilation of new information into a
knowledge base, as an alternative understanding of the way in which a
knowledge base formulated in logic relates to externally generated input
sentences that describe experience. The new pragmatic approach is based
on a proof-theoretical assimilation of observational sentences into a
knowledge base of sentences formulated in a language such as CL.6
Kowalski proposes a pragmatic alternative view that contrasts with the
model-theoretical approach to logic. In model theory, notions such as
interpretation and semantic structures dominate and are informed by the
philosophical assumption that experience is caused by an independent
existing reality composed of individuals, functions and relations, separate
from the syntax of language.
By contrast, logic programs can be seen as agents endowed with
deductive databases considered Kowalski says as theory presentations
from which logical consequences are derived, both in order to internally
solve problems with the help of theoretical sentences and in order to
assimilate new information from the external world of observations
(observational sentences). The part of the knowledge base which includes
observational sentences and the theoretical sentences that are used to
derive conclusions that can be compared with observational sentences, is
called the world model, considered a completely syntactical concept:
World models are tested by comparing the conclusions that can be
derived from them with other sentences that record inputs, which are
observational sentences extracted assimilated from experience. The
agent might generate outputs that are generated by some plan formation
process in the context of the agents resident goals which affect its
environment and which of course can affect its own and other agents
future inputs. Kowalski concludes The agent will record the output,

6
CL, computational logic, refers to the computational approach to logic that has
proved fruitful for creating nontrivial applications in computing, artificial
intelligence, and law.
186 Chapter Eleven

predict its expected effect on the environment using the world model and
compare its expectations against its later observations.
I think the epistemological consequence of this approach is
fundamental: in model theory truth is a static correspondence between
sentences and a given state of the world. In Kowalskis computational and
pragmatic theory, what is important is not the correspondence between
language and experience, but the appropriate assimilation of an inevitable
and continuously flowing input stream of external observational
sentences into an ever-changing internal knowledge base (of course the
fact that the computational resources available are bounded motivates the
agent to make the best use of them, for instance by avoiding the redundant
and irrelevant derivation of consequences). The correspondence (we can
say the mirroring) between an input sentence and a sentence that can be
derived from the knowledge base is considered by Kowalski as only a
limiting case. Of course the agent might also generate its own hypothetical
inputs, as in the case of abduction, induction, and theory formation.
The conceptual framework above, derived from a computationally
oriented logical approach that strongly contrasts with the traditional
approach in terms of model theory, is extremely interesting. It stresses
attention to the flowing interplay between internal and external
representations/statements, so epistemologically establishing the importance
of the agentbased character of cognition and thus of logical - in the
broad sense - cognition. In the recent cognitive science approach in terms
of distributed cognition this perspective is also helpful for depicting the
cognition of actual beings in so far as we are interested in studying its
essential distributed dynamics.

3.2. Distributed Cognition in Human Organic Agents:

External and Internal Representations
Mind is limited, both from a computational and an informational point
of view: the act of delegating some aspects of cognition becomes
necessary. It is in this sense that we can say that cognition is essentially
multimodal.7 In addition, we can say that, adopting this perspective, we

7
Thagard (2005, 2006) observes that abductive inference can be visual as well as
verbal, and consequently acknowledges the sentential, modelbased, and
manipulative nature of abduction I have illustrated in my books on this subject
(Magnani, 2001, 2009). Moreover, both data and hypotheses can be visually
represented: For example, when I see a scratch along the side of my car, I can
 Natural Mathematics and Natural Logic 187

can give an account of the complexity of the whole human cognitive

system as the result of a complex interplay and coevolution of states of
mind, body, and external environments suitably endowed with cognitive
(in the cases illustrated in this commentary logical) significance. That is,
taking advantage of Celluccis words, I can say that there is a coevolution
between natural mathematics and logic and their artificial counterparts.
The agent-based view I have illustrated in the previous subsection aims
at analysing the features of real (natural) human thinking agents and
ideal (artificial) logical agents by recognizing the fact that a being-like-
us agent functions at two levels and in two ways. I define the two
levels as explicit and implicit thinking. Agent-based perspective in logic
has the power to recognize the importance of both levels.
We maintain that representations are external and internal. We can say
that
- external representations are formed by external materials that re-
express (through reification) concepts and problems that are already
present in the mind or concepts and problems that do not have a natural
home in the brain;
- internalized representations are internal re-projections, a kind of
recapitulations, (learning) of external representations in terms of neural
patterns of activation in the brain. They can sometimes be internally
manipulated like external objects and can originate new internal
reconstructed representations through the neural activity of transformation
and integration.

3.3. Internal, External, and Hybrid Inducers and Abducers:

External Semiotic Anchors
In what follows I will illustrate some features of this extraordinary
interplay between human brains and the ideal cognitive systems they
make, and so of ideal logical (and computational) agents. We acknowledge
that material artifacts such as inductive and abductive logical and
computational agents are tools for thought, as is language: tools for

generate the mental image of grocery cart sliding into the car and producing the
scratch. In this case both the target (the scratch) and the hypothesis (the collision)
are visually represented. [...] It is an interesting question whether hypotheses can
be represented using all sensory modalities. For vision the answer is obvious, as
images and diagrams can clearly be used to represent events and structures that
have causal effects (2006). Indeed hypotheses can also be represented using other
sensory modalities.
188 Chapter Eleven

exploring, expanding, and manipulating our own minds. A novel

perspective on external ideal logical agents can be envisaged.
Human beings (and animals, as Peirce already maintained) spontaneously
perform more or less rudimentary abductive and inductive reasoning.
Starting from lowlevel in-formal (natural) inferential performances such
as hasty generalization or simple abductive diagnoses, widespread in
children and adult humans, that certainly may represent a strategic success
(for instance survival, as also maintained by Cellucci (2008)) and a cognitive
failure (they are not at all truth-preserving, and thus epistemologically
unsatisfactory) human beings arrived at the externalization of theoretical
inductive and abductive agents as ideal agents, logical/axiomatic and
computational. It is in this way that merely natural successful strategies
are replaced with artificial successful strategies that also tell the more
precise truth about things. These external representations can be usefully
re-represented in our brains (if this is useful, simple, and possible), and
they can give rise to new improved organic (mentally internal) ways of
inferring or be suitably exploited in a hybrid manipulative interplay, as I
have already said above.
In summary, we can partially copy (recapitulating them through
internalization) ways of reasoning (or fragments of them) from some
aspects of what we have externalized out there, in the external
environment, for instance in ideal logical systems/agents. This
recapitulation/internalization relies on the fact, recently stressed by some
biologists, that general/genetic inheritance (natural selection among
organisms influences which individuals will survive to pass on their genes
to the next generation) is not the only inheritance system to play a
fundamental role in Darwinian biological evolution: cultural niche
construction (in this case artificial logical agents) plays a diachronic role
in a time span covering several generations, and this introduces a second
general inheritance system (also called ecological or cultural inheritance).
According to the theory of cognitive niches, the two systems coevolve
(Odling-Smee, et al., 2003).
From this perspective, human beings are hardwired for survival and for
truth alike so that the best inductive and abductive strategies can be built
and made explicit, through self-correction and re-consideration (a process
that is at work at least, for example, since the time of Mills ideal
inductive methods). Furthermore, human beings are agents that can
cognitively behave as hybrid agents that exploit both internal representations
and externalized logical representations and tools in reasoning, but also a
mixture of the two.
 Natural Mathematics and Natural Logic 189

Let us consider the example of the externalization of some inferential

skills in logical demonstrative systems, such as those that are at the basis
of logic programming and semantic tableaux, both illustrated and
improved on by Aliseda (2006). They present interesting cognitive
features (cf. also Longo, 2005) which I believe deserve further analysis:

1. Symbolic: they activate and semiotically anchor meanings in

material communicative and intersubjective mediators in the framework of
the phylogenetic, ontogenetic, and cultural reality of the human beings and
their language. I have already said that it can be hypothesized that these
logical agents originated in embodied cognition, gestures, and
manipulations of the environment we share with some mammals but also
with non-mammals (cf. the case of the complicated monkeys knots,
endowed with implicit mathematical features, and pigeons categorization,
as implicit concept formation, in Grialou, Longo, and Okada, 2005);
2. Abstract: they are based on a maximal independence regarding
sensory modality; they strongly stabilize experience and common
categorization. Maximality is especially important: it refers to their
practical and historical invariance and stability;
3. Rigorous: the rigour of proofs is attained through difficult practical
experience. For instance, in the case of mathematics and logic, as the
maximal place for convincing and sharable reasoning. Rigour lies in the
stability of proofs and in the fact they can be replicated. In this
perspective, mathematics is the best example of maximal stability and
conceptual invariance. Logic is in turn a set of proof invariants, a set of
structures that are preserved from one proof to another or which are
preserved by proof transformations. As the externalization and result of a
distilled praxis, the praxis of proof, it consists of maximally stable
regularities;
4. I also say that a maximization of memorylessness8 variably
characterizes demonstrative reasoning. This is particularly tangible in the
case of the vast idealization of classical logic and related approaches. The
inferences described by classical logic do not yield sensitive information
so to speak about their real past life in human agents use, contrary to
conceptual narrative descriptions of human informal non-demonstrative
processes, which variously involve historical, contextual, and

8
I derive this expression from Leyton (2001) who introduces a very interesting
new geometry where forms are no longer memoryless as in classical approaches
such as the Euclidean and the Kleinian in terms of groups of transformations.
190 Chapter Eleven

heuristic memories. Indeed many informal thinking behaviours in

human agents for example abductive inferences, especially in their
generative part are context-dependent. As already noted, their stories
vary with the multiple propositional relations the human agent finds in
his/her environment and which he/she is able to take into account, and
with various cognitive reasons to change his/her mind or to think in a
different way, and with multiple motivations to deploy various tactics of
argument.
In this perspective Gabbay and Woods say:

Good reasoning is always good in relation to a goal or an agenda which

may be tacit. [. . . ] Reasoning validly is never itself a goal of good
reasoning; otherwise one could always achieve it simply by repeating a
premiss as conclusion, or by entering a new premiss that contradicts one
already present. [. . . ] It is that the reasoning actually performed by
individual agents is sufficiently reliable not to kill them. It is reasoning that
precludes neither security not prosperity. This is a fact of fundamental
importance. It helps establish the fallibilist position that it is not
unreasonable to pursue modes of reasoning that are known to be imperfect
(Gabbay and Woods, 2005, pp. 19-20)

Human agents, as practical agents, are hasty inducers and abducers and
bad predictors, unlike ideal (logical and computational) agents. In
conclusion, we can say that informal abductive inferences in human agents
have a memory, a story: consequently, an abductive ideal logical agent
which formalizes those human skills has to variably weaken many of the
aspects of classical logic and overcome the relative demonstrative
limitations. The informal/natural aspects embody what Cellucci (2000)
calls the open world view, while ideal/artificial agents represent the
closed world view. Indeed, Cellucci stresses that only thanks to the open
world view can we acknowledge the richness of a wider perspective on
logic and mathematics, where communication, cooperation, and
negotiations (p. 173) are in play and other central cognitive components
can be taken into account, such as abduction, analogy/meta-
phor/metonymy, induction/specialization/generalization, the role of tacit
knowledge, etc.
We can conclude by stressing the fact that the human informal non-
demonstrative inferential process of abduction (and of induction) is
increasingly artificialized, thanks to externalization and objectification, in
at least three ways:

1. Through Turings Universal Practical Computing Machines we can

have running programs often based on logic, that are able to mimic and
 Natural Mathematics and Natural Logic 191

enhance the actions of a human computer very closely (Turing, 1950),

and so - amazingly - also those actions of human agents that correspond
to complicated inferential performances like abduction (cf. the whole area
of artificial intelligence);
2. Human non-demonstrative processes are increasingly externalized
and made available in the form of explicit narratives and learnable
templates for behaviour (cf. also the study of fallacies as important tools in
that human kit that provides evolutionary advantages, in this sense any
fallacy of the affirming the consequent which depicts abduction in
classical logic can be better than nothing Woods, 2004).9
3. New demonstrative systems ideal/artificial logical agents are
created, able to model and make rigorous in a demonstrative way many
non-demonstrative thinking processes like abduction, analogy, creativity,
spatial and visual reasoning, etc.

A skeptical conclusion about the superiority of demonstrative over

non-demonstrative reasoning is provided by Celluccis philosophical
argumentation (2005), which seems to emphasize the role of ignorance
preservation in logic and with which I agree: To know whether an
argument is demonstrative one must know whether its premises are true.
But knowing whether they are true is generally impossible, as Gdel
teaches. So they have the same status as the premises of non-
demonstrative reasoning. Moreover: demonstrative reasoning cannot be
more cogent than the premises from which it starts; the justification of
deductive inferences in any absolute sense is impossible, they can be
justified as much, or as little, as non-deductive ampliative inferences.
Checking soundness is also a problem.

References
Aliseda, A. (2006). Abductive Reasoning. Logical Investigations into
Discovery and Explanation. Berlin: Springer.
Bringsjord, S. (1998). Is (Gdelian) model-based deductive reasoning
computational?, Philosophica, 61, 51-76
Cellucci, C. (1998). The scope of Logic: deduction, abduction, analogy.
Theoria, 64/2-3, 217-242.

9
Cf. also Gabbay and Woods (2005, pp. 33-36).
192 Chapter Eleven

. (2000). The growth of mathematical knowledge: an open world view.

In: E. Grosholz and H. Breger (eds.), The Growth of Mathematical
Knowledge. Dordrecht: Kluwer Academic Publishers.
. (2005). Mathematical discourse vs. mathematical intuition. In:
Cellucci, C., Gillies, D. (eds.) Mathematical Reasoning and Heuristics,
pp. 138166. Kings College Publications, London.
. (2008). Why proof? What is a proof?. In: G. Corsi and R. Lupacchini
(eds.), Deduction, Computation, Experiment. Exploring the
Effectiveness of Proof . Berlin, Springer, pp. 1-27.
Clark, A. (2003). Natural-Born Cyborgs. Minds, Technologies, and the
Future of Human Intelligence. Oxford: Oxford University Press.
Devlin, K. (2005), The math instinct. Why youre a mathematical genius
(along with lobsters, birds, cats, and dogs). New York: Thunders
Mouth Press.
Gabbay, D. M. and J. Woods, J. (2005). The Reach of Abduction.
Amstedam: North-Holland.
Grialou, P., Longo, G., and Okada, M. (Eds.). Images and Reasoning,
Tokyo: Keio University.
Hintikka, J. (1997). On creativity in reasoning. In: . E. Andersson and
N.-E. Shalin (Eds.), The Complexity of Creativity (pp.67-78).
Dordrecht: Kluwer Academic Publishers.
. (1998). What is abduction? The fundamental problem of
contemporary epistemology, Transactions of the Charles S. Peirce
Society, 34, 503-533.
Hintikka, J. and Remes, U. (1974). The Method of Analysis. Its
Geometrical Origin and Its General Significance. Dordrecht: Reidel.
Hutchins, E. (1995). Cognition in the Wild. Cambridge, MA: The MIT
Press.
Kowalski, R. (1994). Logic without Model Theory. In: What is a Logical
System?, D. M. Gabbay (Ed.) (pp. 3571), Oxford: Oxford University
Press.
Leyton, M. (2001). A Generative Theory of Shape. Berlin: Springer.
Longo, G. (2005). The cognitive foundations of mathematics: human
gestures in proofs and mathematical incompleteness of formalisms. In:
P. Grialou, G. Longo, and M. Okada (EDs.) (pp. 105-134). Images and
Reasoning. Tokyo: Keio University.
Magnani, L. (2001). Abduction, Reason, and Science. Processes of
Discovery and Explanation. New York: Kluwer Academic/Plenum
Publishers.
. (2005). Abduction and cognition in human and logical agents. In: S.
Artemov, H. Barringer, A. Garcez, L. Lamb, and J. Woods (Eds.) (pp.
 Natural Mathematics and Natural Logic 193

225-258), We Will Show Them: Essays in Honour of Dov Gabbay, vol.

II. London: College Publications.
. (2006a). Hasty generalizers and hybrid abducers. In: P. A. Flach, A. C.
Kakas, L. Magnani, O. Ray (Eds.) (pp.1-8). Workshop Abduction and
Induction in AI and Scientific Modeling. Trento: University of Trento.
. (2006b). Multimodal abduction. External semiotic anchors and hybrid
representations, Logic Journal of IGPS, 14(1), 107-136.
. (2009). Abductive Cognition. The Epistemological and Eco-Cognitive
Dimensions of Hypothetical Reasoning. Berlin/Heidelberg: Springer.
Magnani, L., N.J. Nersessian, and Thagard, P. (Eds.) (1999). Model-Based
Reasoning in Scientific Discovery. New York: Kluwer Academic/
Plenum Publishers.
Meheus, J. (1999). Model-based reasoning in creative processes. In: L.
Magnani, N.J. Nersessian, and P. Thagard (Eds.) (pp. 199-217).
Meheus, J. and Batens, D. (2006). A formal logic for abductive
reasoning, Logic Journal of IGPL, 14(1), 221-236.
Niiniluoto, I. (1999). Abduction and geometrical analysis. Notes on
Charles S. Peirce and Edgar Allan Poe. In: L. Magnani, N.J.
Nersessian, and P. Thagard, (Eds.) (pp. 239-254).
Odling-Smee, F.J., Laland, K.N., Feldman, M.W. (2003). Niche
Construction. The Neglected Process in Evolution. Princeton:
Princeton University Press.
Peirce, C.S. (1931-1958). Collected Papers of Charles Sanders Peirce.
Cambrideg: Harvard University Press; vol. 1-6, Hartshorne, C., Weiss,
P. (eds.); vols. 7-8, Burks, A.W. (ed.)
Thagard, P. (2005). How does the brain form hypotheses? towards a
neurologically realistic computational model of explanation. In: P.
Thagard, P. Langley, L. Magnani, and C. Shunn (2005), Generating
explanatory hypotheses: mind, computer, brain, and world,
Proceedings of the 27th International Cognitive Science Conference.
Stresa, Italy, June 2005, CD-Rom.
. (2006), Abductive inference: from philosophical analysis to neural
mechanisms. In: A. Feeney and E. Heit (Eds.), Inductive Reasoning:
Cognitive, Mathematical, and Neuroscientific Approaches. Cambridge:
Cambridge University Press.
Turing, A. M. (1950). Computing machinery and intelligence, Mind, 49,
433460.
Woods, J. (2004). The Death of Argument. Dordrecht: Kluwer Academic
Publishers.
 CHAPTER TWELVE

FOR A BOTTOM-UP APPROACH

TO THE PHILOSOPHY OF SCIENCE

MARIA CARLA GALAVOTTI

SUMMARY The aim of this article is to extend Celluccis bottom-up

approach to mathematics to the philosophy of science at large, in the
conviction that a bottom-up perspective will lead to a better understanding
of the nature of scientific knowledge and its methodology. The first
section summarizes the main traits of Celluccis position. The second
section discusses the innovative ideas of Patrick Suppes, who embraces a
bottom-up approach to philosophy of science having a strong affinity with
Celluccis viewpoint. The third section outlines the constructivist approach
to statistics recently developed by Christian Hennig in a bottom-up
perspective. The fourth and final section focusses on the notion of context,
which represents an essential ingredient of philosophy of science in a
bottom-up outlook.

KEYWORDS philosophy of science, epistemology, statistics, constructivism,

pluralism, context

1. Carlo Celluccis bottom-up approach to the foundations

of mathematics
In a series of original writings Carlo Cellucci argues in favour of a bottom-
up approach to the foundations of mathematics, revolving around the
conviction that mathematics is developed from below, namely starting
from specific problems that arise within the natural and social sciences.
The bottom-up approach is developed by the analytic method, which
moves upward from problems to hypotheses derived non-deductively
therefrom. Cellucci contrasts his own approach with the long-standing top-
down approach to mathematics which has been dominant thanks to a
number of authors including Dirichlet, Riemann, Dedekind, Klein, Hilbert,
196 Chapter Twelve

Noether and Bourbaki. Unlike the bottom-up approach, the top-down

strategy moves from general principles to specific problems, and makes
use of the axiomatic method according to which one assumes as starting
points axioms which are true in some sense, and deduces logical
consequences from them (Cellucci 2013a, p. 100). Cellucci argues that
the axiomatic method is beset with a number of difficulties, including first
and foremost Gdels incompleteness theorems. The top-down approach is
not limited to mathematics, also being adopted within empirical sciences.
According to Cellucci, the top-down approach has proven unsuccessful in
a number of fields, including economics and the life sciences, and that
even in those fields where it has been successfully applied, its success
remains unexplained.
By contrast, the bottom-up approach starts from the specific problems
arising within specific disciplines and looks for possible solutions making
use of non-deductive inferences. Once some hypothesis is suggested as the
source of a possible solution to a given problem, its plausibility is tested
against evidence by comparing its consequences with the available data.
The bottom-up approach makes use of the analytic method, which Cellucci
traces back to Hippocrates of Kos, who grounded on it medical practice.
The analytic method represents a problem-solving activity, whose
peculiarity is that

to solve a given problem, one looks for some hypothesis that is a sufficient
condition for its solution. The hypothesis is obtained from the problem,
and possibly other data, by means of some non-deductive rule. The
hypothesis must be plausible, that is, compatible with the existing data. But
the hypothesis is in turn a problem that must be solved, and will be solved
in the same way (Cellucci 2013b, p. 34).

The notion of plausibility, which plays a crucial role within the

bottom-up approach, is not a priori, being rather based on experience.
According to Cellucci, a certain hypothesis can be deemed plausible if
comparing the arguments for and the arguments against the hypothesis on
the basis of the existing data, the arguments for must be stronger than
those against (Ibid.).
The strong link Cellucci establishes between mathematics and
experience is reflected by his distinction between natural and artificial
mathematics. Natural mathematics corresponds to the ability possessed by
human beings (and other living creatures) to capture the primary
information in the system of positive integers and in the system of
Euclidean plane geometry (Ibid., p. 33). This ability is developed to
guarantee survival and the satisfaction of basic needs, whereas artificial
 For a Bottom-Up Approach to the Philosophy Of Science 197

mathematics is an artifact acquired through education. While natural

mathematics is taken as resulting from biological evolution, artificial
mathematics is the product of cultural evolution. There emerges a
naturalistic view of mathematics, which according to Cellucci is supported
by recent literature in cognitive science and the foundations of mathematics.
It is worth mentioning that Celluccis viewpoint discards the
distinction between a context of discovery and a context of justification,
which is part of the received view of philosophy of science inherited
from logical empiricism. As put by Cellucci:

The analytic method cancels the separation between discovery and

justification because, according to it, the solution of a problem is both a
process of discovery and a process of justification. It is a process of
discovery, because it involves finding hypotheses by means of non-
deductive rules. It is a process of justification because it involves
comparing the arguments for and the arguments against the hypotheses
thus found, on the basis of the existing data (Ibid., pp. 35-6).

The justification of the analytic method, and more specifically of the non-
deductive rules it makes use of, relies on their usefulness for the solution
to problems. As observed by Cellucci, this kind of justification is akin to
what Herbert Feigl called vindication. In an influential paper published
in 1950, called De principiis non disputandum...? On the Meaning and
the Limits of Justification Feigl made a distinction between two kinds of
justification procedures: validation and vindication. The validation
procedure consists in justifying a given argument by appealing to more
general standards, until some fundamental justifying principles are
reached. This corresponds to what has traditionally been done in deductive
logic, where the deductive chain is traced back to the axioms in order to
justify a theorem. By contrast, vindication is in terms of means with
respect to ends. In other words, it appeals to pragmatic considerations, like
the evaluation of whether the means employed are suitable to the
achievement of some desired end. According to Feigl, vindication applies
to the basic principles of both deductive and inductive logic, the difference
being that the end of deductive logic is to make truth-preserving
deductions, while the end of inductive logic is to widen our knowledge by
formulating successful predictions.
Cellucci is led by his naturalistic perspective to reformulate Feigls
notion of vindication in tune with his own view of the role of deductive
and non-deductive rules, which revolves around the notion of plausibility.
In the first place, for Cellucci the purpose of deductive rules is to make
explicit the content or part of the content that is implicit in the premises
198 Chapter Twelve

(Cellucci 2011, p. 133). Given that this is what makes deductive rules
useful for knowledge, in order to accomplish this purpose deductive
inferences must have plausible premisses and conclusions. Then, the
usefulness of deductive rules essentially depends on a comparison with
experience (Ibid., p. 134). Furthermore, Cellucci identifies the task of
non-deductive rules with the discovery of hypotheses, and maintains that
the usefulness of a non-deductive argument depends on the plausibility of
its premisses and conclusion, which again calls for a comparison with
experience. Cellucci reaches the conclusion that

both deductive and non-deductive rules can be vindicated with respect to

an end that agrees with their role in knowledge. [...] The usefulness of both
deductive and non-deductive rules essentially depends on a comparison
with experience, so with something external to inference (Ibid., p. 137).

This establishes a fundamental symmetry between deductive and non-

deductive rules that supports the conclusion that there is nothing
distinctively perplexing about induction as opposed to deduction (Ibid., p.
141). Such an attitude obviously clashes with the conviction accredited by
the top-down approach that deduction is the canon of reasoning and that
there is a substantial asymmetry between deduction and induction. In its
place Cellucci reaffirms a continuity between induction and deduction based
on their utility for knowledge, thereby embracing a pragmatist attitude.
Some further aspects of Celluccis bottom-up approach will be
highlighted in the next section.

2. Patrick Suppes bottom-up approach to philosophy

of science
In a series of writings ranging from Models of Data (1962) to the
monumental book Representations and Invariance of Scientific Structures
(2002), Patrick Suppes develops an approach to philosophy of science and
epistemology that can be deemed bottom-up, although this is not the
terminology he uses. Suppes perspective marks a sharp turn with respect
to the received view, which he contrasts with a pragmatist standpoint that
regards theory and observation as intertwined rather than separate,
establishes a continuity between the context of discovery and the context
of justification, and takes scientific theories as principles of inference,
which prove useful in making predictions and choosing between
alternative courses of action.
 For a Bottom-Up Approach to the Philosophy Of Science 199

The received view of theories, portrayed by Hempels famous image of

scientific theories as complex spatial networks, is definitely top-down.
According to this view theories represent nets whose knots stand for
theoretical terms, connected by definitions and/or hypotheses included in
the theory. The net floats on the plane of observation, to which it is
anchored by means of rules of interpretation1. This view assigns to
philosophy of science the task of clarifying the relationship between
theoretical and observational terms, while taking the data coming from
observation and experimentation as given. Data are not intended as objects
of study for philosophy of science because they belong to the context of
discovery, not to that of justification.
Suppes alternative approach abandons the clear-cut distinction between
theoretical and observational terms made by logical empiricists, to hold
that the relation between empirical theories and data calls for a hierarchy
of models (Suppes 1962, p. 253) characterized by different degrees of
abstraction, where there is a continuous interplay between theoretical and
observational model components. Models in this hierarchy range from
models of data, describing empirical evidence, to the abstract
mathematical models characterizing a theory. A multiplicity of experimental
models stand at the intermediate levels of this hierarchy, moving from
bottom to top because given a model of the data exhibiting a certain
statistical structure of some phenomenon under investigation a fitting
theoretical model is sought. Suppes maintains that in order to understand
the nature of scientific knowledge it is essential to analyse theories not
abstractly, but in connection with experimentation, and claims that to
recognize [...] the existence of a hierarchy of theories arising from the
methodology of experimentation for testing the fundamental theory is an
essential ingredient of any sophisticated scientific discipline (Suppes
1967, pp. 63-64). The lack of consideration of the role played by statistical
methods in experimentation and theory making is regarded by Suppes as a
lacuna of the top-down approach traditionally taken by philosophers of
science who write about the representation of scientific theories as logical
calculi and go on to say that a theory is given empirical meaning by
providing interpretations of coordinating definitions for some of the
primitive or defined terms of the calculus (Ibid.).
Moreover, Suppes calls attention to the complexity of data delivered
by observation and experimentation. In his words: the data represent an
abstraction from the complex practical activity of producing them. Steps

1
See Hempel (1958).
200 Chapter Twelve

of abstraction can be identified, but at no one point is there a clear and

distinct reason to exclaim, Here are the data! (Suppes 1988, p. 30).
Depending on the desired level of abstraction different pieces of
information will then count as data, and what qualifies as relevant will
inevitably depend on a cluster of context-dependent elements. The strong
emphasis put on the context is a key feature of Suppes approach, which is
deeply pluralist and pragmatist.
Suppes pluralism is grounded in the conviction that the complexity of
phenomena and the variety of practical situations in which phenomena are
investigated are such that most important notions of science, as well as
philosophy, cannot be forced into some definition given once and for all.
Instead of looking for a unique way of representing what he calls
scientific structures, one should admit that a multiplicity of representations
can be produced, resulting in a multi-faceted view of scientific knowledge.
Plurality represents for Suppes one of the tenets of the new metaphysics
which serves in his book Probabilistic Metaphysics to fight the chimeras
of the traditional view of rationality retained by logical empiricists. The
ideal of the unity of science is deemed one of such chimeras, as is the idea
of completeness of knowledge, to which Suppes objects that the
collection of past, present, and future scientific theories is not converging
to some bounded fixed result that will in the limit give us complete
knowledge of the universe (Suppes 1984, p. 10). Other pillars of the
traditional view of rationality that Suppes regards as chimeras are the
ideals of certainty and determinism.
It is worth noting that Cellucci, in his book Perch ancora la filosofia,
also deems chimeras a number of tenets of the traditional view of
knowledge as true justified belief. Among such chimeras he mentions
truth, objectivity and certainty. These deceptive ideals go hand in hand
with the chimera of intuition, namely the conviction that intuition is the
source of immediate, absolute knowledge; the chimera of deduction, that
is to say the claim that deduction is the essence of reasoning; the chimera
of rigour, to mean that by borrowing from mathematics the axiomatic
method philosophy can attain the rigour of mathematical reasoning; and
the chimera of mind, taken as the tenet that the mind exists
independently of the body and that knowledge is an entirely mental
process. To these chimeras Cellucci opposes a heuristic conception of
knowledge aiming at plausibility rather than truth, and agreement rather
than objectivity. Cellucci also maintains that knowledge is intrinsically
uncertain, and progress is obtained by formulating hypotheses that prove
fruitful for the solution to particular problems.
 For a Bottom-Up Approach to the Philosophy Of Science 201

In a similar vein, Suppes takes science to be a perpetual problem-

solving activity, and regards scientific theories as constructs which like
our own lives and endeavours [...] are local and are designed to meet a
given set of problems (Suppes 1981, pp. 14-15). Furthermore, Suppes
regards uncertainty as an ingrained feature of science and human knowledge
in general, and claims that it is the responsibility of a thoroughly-worked-
out empiricism to include an appropriate concept of uncertainty at the
most fundamental level of theoretical and methodological analysis
(Suppes 1984, p. 99). The uncertain nature of knowledge requires a
fundamental role to be ascribed to probability, which provides the tool to
build a new metaphysics according to which the basic laws of natural
phenomena, causality, the theory of meaning, and rationality have a
probabilistic character. In the conviction that probability enters all stages
of a comprehensive analysis of knowledge, Suppes labels his own
approach probabilistic empiricism, as opposed to logical empiricism.
As suggested by the preceding overview, Suppes probabilistic
empiricism has much in common with Celluccis bottom-up approach.
Both of them are inspired by a pragmatist attitude, regard science as
problem-solving, and retain a naturalistic view of rationality. The
contributions made by Suppes and Cellucci can be seen as complementary
to the extent that they develop different aspects of the bottom-up approach.

3. Statistics in a bottom-up perspective

In recent years, the statistician Christian Hennig has developed a bottom-
up approach to statistics that can be regarded as an expansion of Celluccis
approach to mathematics and Suppess view of philosophy of science.
Hennig focusses on mathematical modelling, more precisely he is not
concerned with mathematical objects in themselves, but with their
relation to the real entities that are modelled (Hennig 2010, p. 31). The
central question addressed by Hennig is how well does mathematics
describe nature and why? (Ibid.). In order to answer this question he takes
a constructivist perspective based on the idea that mathematical
modelling can be seen as a tool for arriving at an agreement about certain
aspects of reality, and therefore for constructing a stable and consensual
reality (Hennig 2009, p. 40).
Hennig distinguishes three levels of reality, namely personal reality,
social reality, and observer-independent reality. Personal reality is the
reality experienced by an individual (Ibid.), which comprises all our
perceptions and thoughts, including our conception of the world.
According to Henning personal reality is a construction of the individual
202 Chapter Twelve

interpreted as a self-organising system (Ibid.). The constructive procedure

leading an individual to form some conception of the world from his
perceptions does not have to be conscious and explicit, for example, a
construct can be regarded as made up behaviour implying tacit assumptions
(Ibid.). Social reality is generated by acts of communication among
individuals. This kind of reality is something between communicating
individuals, separated from the personal reality within each individual
(Ibid.). Personal and social are distinct albeit interconnected realities: on
the one hand social reality is generated by attempts of individuals to
communicate their personal perceptions; on the other hand, social reality is
not only perceived by individuals, but has a strong impact on the
construction of personal reality, because people use language to think.
Perceptions are connected, structured and even changed by language
(Hennig 2010, p. 34). The patterns of perceptions and/or actions that are
formed through self-organisation and communication are called constructs,
and can be personal or social. Finally, by observer-independent reality
Hennig means the reality outside us. This is described by constructs that
are stable. In Hennigs words:

there are personal constructs that are consistent with different sensual
perceptions at different times for different points of view. Furthermore, we
observe that the corresponding social constructs (i.e. the ones to which the
same word refers in communication) are stable as well and that other
people behave consistently with the communicative acts and our
perception concerning these constructs (Ibid., p. 35).

The persistence of such stable constructs may suggest the existence of a

third level of reality, independent of the observer. However, observer-
independent reality is only accessible through the perceptions processed
by our brain. In other words, this level of reality is not accessible
independently of the observer, and therefore nothing can be said about its
ontological status. Hennig embraces an agnostic attitude towards the
existence of the reality outside the observer, but adds that constructivism
is compatible with a relatively weak form of realism (Hennig 2009, p.
41).
As to the place of mathematical modelling within this picture, Hennig
states that mathematics in its recent formalised form can be regarded as a
closed social system generating its own reality: mathematical (or formal)
reality. The claim of mathematics is to provide a communicative domain
in which absolute agreement is possible, and constructs are absolutely
stable (Hennig 2010, p. 37).
 For a Bottom-Up Approach to the Philosophy Of Science 203

In a broad sense mathematical modelling is a social construct, but one

of a peculiar kind because of its capacity to deliver stable representations
of phenomena on which all agree. Like Cellucci, Hennig identifies the
origins of mathematics with the practice of counting, made necessary by
the satisfaction of basic needs. In the course of its history, mathematics
evolved towards increasing abstraction. This process led to a point where
the correspondence between mathematical modelling and practical situations
became blurred. Hennings conclusion is that although mathematics
originated from human activity, it has acquired a reality of its own quite
independent from personal and social realities.
Regarding the question of the truth of mathematical constructs, Hennig
maintains that the idea of absolute truth in mathematics can be explained
by a historical process of construction that made binding agreement the
essential aim of mathematical communication (Hennig 2009, p. 42). The
claim here is that there is no need for a strong, realistic conception of truth
in order to understand or justify mathematical modelling. Mathematics
belongs to science, and the main (defining) objective of science,
interpreted as a social system [...] is to establish an ideally growing body
of stable constructs about which general agreement is possible (Henning
2010, p. 35). The assessment of a mathematical model is based on the fact
that it fits its purpose. In other words, the criterion for the justification of
mathematical models is given by their success. Models serve a number of
practical purposes, being used for prediction, manipulation and in some
cases explanation, they reduce complexity, support decision-making,
improve mutual understanding, support agreement on the modelled
phenomena, and explore different scenarios. The very fact that models
prove fruitful for the fulfilment of these tasks provides good grounds for
their vindication, to adopt Feigls - and Celluccis - terminology. As a
matter of fact, Hennigs views on the truth of mathematical constructs
largely agrees with Celluccis claim that there is no more to mathematical
existence than the fact that mathematical objects are hypotheses tentatively
introduced to solve problems. Mathematical objects have no existence
outside our mind, since they are simply mental tools which we use to solve
problems (Cellucci 2013b, p. 36).
Hennig pays special attention to statistical modelling. In accordance
with the approach just outlined, he focusses on the main purpose of
statistics, identified with the representation and appraisal of the strength of
evidence. Hennig emphasizes the complex nature of statistical evidence,
and calls attention to the assumptions underlying probabilistic methods for
evaluating evidence, warning against the widespread tendency to quantify
evidence in a unified way regardless of the subject matter (Hennig 2009,
204 Chapter Twelve

p. 44). Such a practice - he claims - goes hand in hand with the conviction
that the mere application of statistical methods to data uncritically taken as
given can produce objective results. By contrast, the bottom-up
analysis recommended by Hennig starts from the context in which data are
collected to move on to the formation of models representing them, and to
the application of methods devised for the quantitative appraisal of
evidence. It is essential that at each step of this process all the assumptions
that are made are spelled out and justified in view of the aim of enquiry.
Hennig examines two different methods for the statistical quantification
of evidence, namely tests of significance and Bayess method, and
compares them in the light of the frequency interpretation of probability,
usually associated with statistical testing, and the subjective interpretation,
often associated with the Bayesian approach. After a detailed analysis that
cannot be recollected here, Hennig comes to the conclusion that the
frequentist assumptions about the world outside seem to stand on a more
or less equal footing with the Bayesian ones about rational reasoning
(Ibid., p. 50). The frequentist assumption Hennig refers to is the claim that
indefinitely long series of repeatable experiments sufficiently identical and
independent can be produced on which relative frequencies are calculated.
The probabilities obtained on such basis are taken by frequentists as
approximations of the true, unknown probabilities characterizing
phenomena.2 By contrast, Hennig claims that the Bayesian approach rests
on the crucial assumption that the individual can always be forced to bet
either in favour of or against an outcome, according to her specified
betting rates (Ibid., p. 49). While it seems questionable that subjective
probability and the betting scheme are inextricably entrenched as
described by Hennig,3 there is no doubt that the Bayesian model of
rationality faces a number of objections, extensively discussed in the
literature. Now, for Hennig the constructivist viewpoint does not contend
that one statistical method is better than another, nor does it affirm the
superiority of one particular interpretation of probability over the others. It
rather leaves the choice of a particular method to the context in which a
particular problem is addressed, in the awareness that different
approaches have different merits and fulfil different aims (Ibid., p. 51). In
other words, statistical methods are vindicated in view of the purpose they

2
See Galavotti (2005) for an account of the frequency interpretation of probability.
3
See Dawid and Galavotti (2009), where the operational definition of probability
on the basis of penalty methods - largely adopted by Bruno de Finetti - is
investigated.
 For a Bottom-Up Approach to the Philosophy Of Science 205

are meant to accomplish. This conclusion, together with Hennigs insistence

on the need to take into account the context in which one operates and to
justify all assumptions that are made, are very much in tune with Suppes
pluralism and pragmatism. Hennigs contribution can be seen as
complementing Suppes perspective, by offering an account of the
formation of models as resulting from the concurrence of personal and
social elements.
It is noteworthy that a plea for contextualism also emerges from the
literature on the statistical methodology for the assessment of causal
relations. Theories of causal modelling include the approach in terms of
Bayesian networks developed by Judea Pearl; the alternative approach,
also making extensive use of graphs, of Clark Glymour, Peter Spirtes,
Richard Scheines, Kevin Kelly and collaborators; the so-called potential
response (PR) method of Donald Rubin, Paul Holland and others; and the
decision-theoretic approach of Philip Dawid.
Judea Pearl defines causal relationships on the basis of the notion of
directed acyclic graph (DAG) he also calls Bayesian network to
emphasize three aspects: (1) the subjective nature of the input information;
(2) the reliance on Bayes conditioning as the basis of updating
information; (3) the distinction between causal and evidential models of
reasoning, a distinction that underscores Thomas Bayes paper of 1763
(Pearl 2000, p. 14). Put briefly, causal Bayesian networks represent
ordered structures of variables exhibiting certain stability conditions which
can lead to manipulations. Such a mechanism-based conception of
interventions (Ibid., p. 24) is the cornerstone of causality viewed as a
useful tool for prediction and intervention. A clear-cut distinction between
seeing and doing underlies Pearls treatment of causality, where the
quantities determined through observation are systematically distinguished
from those obtained through experiment. This distinction plays a crucial
role in predicting the results of controlled experiments from observed
probabilities, which from this perspective is the main task of causality.
Pearl also contemplates the explanatory use of causal models to provide
an explanation or understanding of how data are generated, or to
convey information on how things work (Ibid., pp. 25-26). A crucial
role is assigned to the stability of causal structures, which should be
durable over time and invariant across a variety of situations. Models
characterized by such robust features allow for predictions and
manipulations that are meant to hold for a wide range of circumstances. So
conceived, the explanatory account of causation is merely a variant of the
manipulative account, albeit one where interventions are dormant (Ibid.,
p. 26). Remarkably, Pearls work on explanation, done in collaboration
206 Chapter Twelve

with Joseph Halpern, reaches the conclusion that when taken in its
explanatory sense causality is context-dependent. This simply follows
from the fact that the whole edifice of causation is made to rest on
modelling, which in turn requires various assumptions so strictly linked
with the context as to justify the claim that the choice of a model depends
to some extent on what the model is being used for (Halpern and Pearl
2005, p. 878).
A slightly different perspective is taken by Philip Dawid, who
advocates a decision-theoretic approach to causation, which is entirely in
terms of conditional probabilities and expectations based on information,
known or knowable, and makes use of models and quantities that are
empirically testable and discoverable (Dawid 2000, p. 408).4 Dawid
identifies the task of causal analysis with making use of past data to take
decisions about future interventions. The distinction between seeing and
doing, taken to be fundamental for the accomplishment of that task, is
articulated by Dawid into the specification of three different situations: an
idle regime, which represents a purely observational, undisturbed,
setting; a regime in which a certain intervention is performed; and a
regime in which an alternative intervention is performed. The probabilistic
consequences of interventions are then compared through the distributions
obtained by observing the behaviour of variables under different regimes,
and causality is associated with the invariance of given conditional
distributions across regimes. Dawid puts special emphasis on the need to
state and justify the assumptions that are made in each particular situation,
on the account that this surgical separation of the formal language from
ad hoc causal assumptions enforces clear and unambiguous articulation of
those assumptions, allows us to develop the logical implications of our
assumptions, and clarifies exactly what needs to be justified in any
particular context (Dawid 2010, p. 83).5 Plainly, reference to the context
in which one operates is also considered essential by the literature on
statistical causal modelling. In Dawids words: appropriate specification
of context, relevant to the specific purposes at hand, is vital to render
causal questions and answers meaningful (Dawid 2000, p. 422).
An important conclusion to be drawn from the work of the authors
mentioned in this section is that it is crucial that all assumptions

4
Dawid puts forward this approach as an alternative to the potential response
model for causal inference developed by Donald Rubin, Paul Holland and other
statisticians, which makes use of counterfactuals. See Holland (2001) and the
bibliography therein.
5
See also Dawid (2007).
 For a Bottom-Up Approach to the Philosophy Of Science 207

underpinning the application of statistical methodology should be spelled

out and justified case by case. For instance, within causal modelling it is
common practice to make exogeneity and/or invariance assumptions, and
in a vast array of situations it is customary to make extensive use of
independence. While putting great emphasis on the need to spell out and
justify the assumptions made by researchers the bottom-up approach
acknowledges that such a justification can only be produced with reference
to a given context. Consequently, the notion of context acquires a primary
role within the bottom-up perspective.

4. Some remarks on context

This last section will offer some hints towards a characterization of the
notion of context. Before embarking on that discussion, let me add that in
a number of writings I have myself embraced a bottom-up approach to the
philosophy of science.6 Such an attitude has been prompted by a careful
analysis of the debate on explanation and causality. This debate, which has
grown to a proportion that cannot be recollected here, has gradually
reached the conclusion that different contexts call for different notions of
both explanation and causality. A growing number of authors working on
these - as well as other - key notions of the philosophy of science seem to
be willing to abandon the search for univocal theories apt to cover a wide
range of disciplines and applications, in favour of a pluralistic attitude
aware of the importance of context-dependent elements.7
Since context constitutes the bedrock on which the bottom-up approach
to the philosophy of science rests, it seems appropriate to conclude my plea
in favour of the bottom-up perspective with some remarks on the notion of
context itself. The notion of context is too complex to be embodied in

6
See Galavotti (2001b), (2006) and (2012). After adopting the locution bottom-
up in my (2001b), I switched to the expression from within in (2006), to resume
the expression bottom-up in the second part of Campaner and Galavotti (2012).
It should be noted that while I use of from within as a variant of bottom-up, to
mean a form of epistemology done from within the sciences, Cellucci uses from
within and from without in the opposite way, namely he associates from
within with the top-down approach, and from without with the bottom-up
approach. Once the way in which we interpret these expressions is clarified, our
attitude towards epistemology is consonant.
7
For a survey of different pluralistic approaches to causality see Galavotti (2008).
A pluralistic approach is heralded in Galavotti (2001a), and Campaner and
Galavotti (2007).
208 Chapter Twelve

some once-and-for-all definition, but with no claim to completeness an

attempt can be made to identify a number of elements that are relevant in
that connection.
In the first place, a context includes the disciplinary framework in
which some problem originates, and more specifically its conceptual
reference setting, compounded by the body of theoretical and methodological
knowledge shared by the scientific community addressing the problem in
question.
A further component of context is the nature of the available evidence,
and also its amount. Evidence can be obtained by observation or by
experiment, and both of these operations admit of a whole array of
possibilities. Experimentation can be done with or without randomization,
on a smaller or larger scale, and it may involve the use of more or less
sophisticated instruments, which require more or less complex
measurement procedures. When evidence consists of statistical data, as is
often the case, these may convey information of a different kind. For
instance, they might express the frequency of a certain property within a
given population, or the opinion of single individuals obtained by
interviewing. The nature of statistical evidence has a bearing upon the
methodology to be adopted within a given context and the interpretation of
the obtained results. Also relevant in the case of statistical data is the
information on the composition and size of the population (reference class)
from which they are obtained. In fact the choice of the reference class is a
major problem underlying statistical methodology and probability
evaluations, especially in the realm of the social sciences. These are just a
few of the problems connected with the notion of evidence, which is the
object of increasing attention on the part of statisticians, computer
scientists, and researchers working in a wide array of fields from psychology
to medicine, economics, archaeology, law, and also philosophers of
science.8
A context is also characterized by the aims of a given investigation.
Among other things, there are contexts in which what is sought is
explanation, and others which aim at prediction. Explanation and
prediction are distinct conceptual operations that do not always match.
Prediction is often associated with manipulation, namely models that are
good for making predictions are meant to provide grounds for planning
manipulations. Disciplines like economics, epidemiology and medicine

8
On the topic of evidence see the recent collection of articles in Dawid, Twining
and Vasilaki, eds. (2012).
 For a Bottom-Up Approach to the Philosophy Of Science 209

offer plenty of examples of situations in which models that do not carry

explanatory information are used for prediction, as well as for taking
decisions on interventions that may concern economic policy, medical
treatments, preventive measures, and the like.9
The preceding remarks were intended merely to hint at the complexity
characterizing the context in which research is carried out. There seems to
be little doubt that by acknowledging the relevance of a number of
contextual factors the bottom-up approach promises to promote a better
understanding of the nature of scientific knowledge.

References
Campaner, R. and Galavotti, M.C. (2007). Plurality in Causality. In
Machamer, P. and Wolters, G., eds., Thinking about Causes. From
Greek Philosophy to Modern Physics. Pittsburgh: University of
Pittsburgh Press: 178-199.
Campaner, R. and Galavotti, M.C. (2010), Some Remarks on Causality
and Invariance. In Carsetti, A., ed., Causality, Meaningful Complexity
and Embodied Cognition. Dordrecht: Springer: 211-230.
Campaner, R. and Galavotti, M.C. (2012). Evidence and the Assessment
of Causal Relations in the Health Sciences. European Studies in the
Philosophy of Science, 26: 27-45.
Cellucci, C. (2008). Perch ancora la filosofia. Bari: Laterza.
. (2011). Classifying and Justifying Inference Rules. In: Cellucci C.,
Grosholz E. and Ippoliti E., eds., Logic and Knowledge. Newcastle
Upon Tyne: Cambridge Scholars Publishing: 93-106.
. (2013a). Top-down and Bottom-up Philosophy of Mathematics.
Foundations of Science XVIII: 93-106. DOI 10.1007/s10699-012-
9287-6.
. (2013b). Philosophy of Mathematics: Making a Fresh Start. Studies in
History and Philosophy of Science 44: 32-42.
Dawid, A.P. (2000). Causal Inference Without Counterfactuals. Journal of
the American Statistical Association, 95: 407-424.
. (2007). Counterfactuals, Hypotheticals and Potential Responses: A
Philosophical Examination of Statistical Causality. In Russo, F. and
Williamson, J., eds., Causality and Probability in the Sciences.
London: College Publications: 503-532.

9
For more on this see Campaner and Galavotti (2010) and (2012), and Galavotti
(2010).
210 Chapter Twelve

. (2010). Beware of the DAG! In Guyon, I., Janzing, D., and Scholkopf,
B., eds., Proceedings of the NIPS 2008 Workshop on Causality.
Journal of Machine Learning Research Workshop and Conference
Proceedings, 6: 59-86.
Dawid, A.P. and Galavotti, M.C. (2009). De Finettis Subjectivism,
Objective Probability, and the Empirical Validation of Probability
Assessments. In Galavotti, M.C., ed., Bruno de Finetti, Radical
Probabilist. London: College Publications: 97-114.
Dawid, P.A., Twining, W. and Vasilaki, D., eds. (2012). Evidence,
Inference and Enquiry. Oxford: Oxford University Press.
Feigl, H. (1950). De principiis non disputandum ...? On the Meaning and
the Limits of Justification. In: Black M., ed., Philosophical Analysis.
Ithaca, N.Y.: Cornell University Press: 119-56. Reprinted in Feigl, H.
(1980). Inquiries and Provocations: Selected Writings 1929-1974,
Cohen, R.S., ed., Dordrecht: Reidel: 237-68.
Galavotti (2001a). Causality, Mechanisms and Manipulation. In Galavotti,
M.C., Suppes, P. and Costantini, D., eds., Stochastic Causality.
Stanford: CSLI Publications: 1-13.
. (2001b). On Discovery and Justification. La Nuova Critica, 37-38: 5-
20.
. (2005). Philosophical Introduction to Probability. Stanford: CSLI
Publications.
. (2006). For an Epistemology From Within. An Introduction to
Suppes Work. Epistemologia, 29: 215-224.
. (2008). Causal Pluralism and Context. In Galavotti, M.C., Scazzieri, R.
and Suppes, P., eds., Reasoning, Rationality and Probability. Stanford:
CSLI Publications: 233-252.
. (2010). Probabilistic Causality, Observation and Experimentation. In
Gonzalez, W.J., ed., New Methodological Perspectives on Observation
and Experimentation in Science. A Corua: Netbiblo: 139-155.
Halpern, J. and Pearl, J. (2005). Causes and Explanations: A Structural-
Model Approach, Part I: Causes, Part II: Explanations. British Journal
for the Philosophy of Science, 56: 843-887; 889-911.
Hempel, C.G. (1958). The Theoreticians Dilemma. In: Feigl, H., Scriven,
M. and Maxwell, G., eds., Minnesota Studies in the Philosophy of
Science, volume II. Minneapolis: University of Minnesota Press.
Reprinted in Hempel, C.G. (1965): 173-226.
. (1965). Aspects of Scientific Explanation and Other Essays in the
Philosophy of Science. New York: The Free Press.
Hennig, C. (2009), A Constructivist View of the Statistical Quantification
of Evidence. Constructivist Foundations, 5: 39-54.
 For a Bottom-Up Approach to the Philosophy Of Science 211

. (2010). Mathematical Models and Reality: A Constructivist

Perspective. Foundations of Science, 15: 29-48.
Holland, P. (2001). The Causal Interpretation of Regression Coefficients.
In Galavotti, M.C., Suppes, P. and Costantini, D., eds., Stochastic
Causality. Stanford, CSLI Publications: 173-187.
Pearl, J. (2000). Causality. Models, Reasoning, and Inference. Cambridge:
Cambridge University Press.
Suppes, P. (1962). Models of Data. In Nagel, E., Suppes, P. and Tarski,
A., eds., Logic, Methodology and Philosophy of Science. Stanford:
Stanford University Press: 252-61.
. (1967). What is a Scientific Theory? In: Morgenbesser, S., ed.,
Philosophy of Science Today. New York: Basic Books: 55-67.
. (1981). The Plurality of Science. In Asquith, P.D. and Hacking, I.,
eds., PSA 1978, vol. II. East Lansing: Philosophy of Science
Association: 3-16. Reprinted in Suppes, P. (1993): 41-54.
. (1984). Probabilistic Metaphysics. Oxford: Blackwell.
. (1988). Empirical Structures. In Scheibe, E., ed., The Role of
Experience in Science, Berlin-New York: Walter de Gruyter: 23-33.
. (1993). Models and Methods in the Philosophy of Science: Selected
Essays. Dordrecht-Boston: Kluwer.
. (2002). Representation and Invariance of Scientific Structures.
Stanford: CSLI Publications.
 CHAPTER THIRTEEN

MATHEMATIZING RISK:
A HEURISTIC POINT OF VIEW

EMILIANO IPPOLITI

SUMMARY In this paper I argue for a heuristic point of view about

mathematization, in particular about the issue of applicability and
effectiveness of mathematics, using Celluccis characterization of
mathematical objects. To this end, I examine the notion of risk and
investigate four main approaches, namely the probabilistic, the
psychological, the fractal and the evolutionary approaches. I show that the
lack of success of the various approaches in the treatment of risk is due to
the ways in which they conceptualize and mathematize it. I set out to show
that the heuristic point of view can offer a better characterization of risk,
which requires a different approach, bottom-up, local and oriented to
problem-solving.

KEYWORDS heuristics, discovery, risk, mathematization, bottom-up

Introduction
There is a received thesis that states that secret of modern science relies on
the successful application of mathematics to the phenomena under
investigation and this, in turn, builds on the successful quantification of
the phenomena it attempts to describe (Schnemann 1994, 150). This
application, and its success, is not without cost since it has its price in
limiting the scope of subject matter which can be so studied. Hence, the
choice of fundamental variables is critical for the success of the intended
mathematization of nature (Ibid). The mathematization has to limit the
study of a phenomenon and I will show that the concept of risk is
exemplary in this respect.
214 Chapter Thirteen

Carlo Cellucci (Cellucci 2005, 2008, 2013b) provides a penetrating

view on this issue. He describes the scientific revolution as the result of a
philosophical turn, since modern science was founded in the seventeenth
century by means of a philosophical turn: giving up the search to know the
essence of natural substances, content with knowing some of their
accidents, such as space, motion, shape, size (Cellucci 2008, 83, my
translation). Such a philosophical turn was put forward by Galileo, which
made it possible to have a mathematical treatment of natural phenomena
that was not possible in Aristotelian science, due to his claim to know the
essence of natural substances, as the essence is not a quantity (Ibid.).
Therefore, it was only by confining itself to the study of accidents that
modern science generated a quantitative treatment of a few fundamental
physical variables. Accordingly, the choice of the fundamental variables
over which to quantify is critical for the scientific enterprise. Galileo made
this choice in physics in order to measure what is measurable and to
render measurable what is not yet so. His problem then became that of
isolating those aspects of natural phenomena which are basic and capable
of measurement (Kline 1953, 186f). To solve this problem, he
investigated a small set of conceptssuch as space, time, weight, velocity,
acceleration, inertia, force, and momentum: in the selection of these
particular properties and concepts Galileo again showed genius, for the
ones he chose are not immediately discernible as the most important nor
are they readily measurable (Ibid.).
In this sense, modern mathematics is a tool designed to treat natural
phenomena, and not the social ones. The idea of mathematizing social
phenomena is relatively recent and, so far, at large extent ineffective. The
fact that a large part of known mathematics was designed to treat physical
problems explains why it fails in social domains: the mathematization of
social domains has been constructed by imitating the mathematization of
physics, expressed mostly in the form of axiomatic systems. But, first the
axiomatization of parts of physics represents the end of a long, bottom-up
process of investigation. On the contrary, in the social domain this step is
put forward at the beginning of the investigation, and hence in a top-down
fashionhoping to model the phenomena with the same rigor and
precision of mathematics and physics. Secondly, even in physics and
mathematics, this final stage of post hoc axiomatization has been achieved
only in a few, narrowly delimited fields (Schnemann 1994, 151). In
effect, there are large parts of physics and mathematics (e.g. number
theory) that are not axiomatized but which are making important
progresses: often the mathematization and axiomatization did not work at
all. The bottom line: a foundational, top-down and formal approach to
 Mathematizing Risk: A Heuristic Point of View 215

problems in the social domains is even more controversial than in physics

and mathematics.
More important, the axiomatic method is affected by strict limits, as
argued by Cellucci (Cellucci 2013b): it is not able to provide a method
for the growth and ampliation of knowledge and it arbitrarily stops the
process of questioning the hypotheses of our knowledge (i.e. the axioms).
Above all, it is not able to detect the specific, novel features of the field
under investigation since it goes top-down, that is from the information
encapsulated in the axioms down to the predictions deduced from them.
Thus, it is only when the phenomena under investigation display features
that fit the information of the axioms (they are known somehow) that the
axiomatic systems can be somehow useful. In this sense, the known
mathematics is super-imposed to social-economic phenomena, and hence
many features of social domains are lost in this mathematical modeling.
The point of this paper is to argue that these limits are overcome by a
heuristic approach. This approach, theorized by Cellucci (Cellucci 2008,
2013b), maintains that knowledge is essentially a problem-solving activity,
and in order to solve a problem it is necessary to start from a specific
problem, searching for a hypothesis that is a sufficient condition for its
solution. The hypotheses can be found by means of some non-deductive
inference (e.g. induction, analogy, metaphor), and once they have been
found, it is necessary to check that they are plausible by deducing
consequences from them and testing that they are compatible with the
existing knowledge. If a hypothesis passes this test, it becomes a new
problem that must be solved in the same way. And so on, ad infinitum.
This process is bottom-up, local and provisional, since the hypotheses
obtained in this way are partial and subject to revision. More specifically,
in this paper I argue for a heuristic approach to the issue of the
mathematization by using the concept of risk as a case study. The notion
of risk is a major factor in the art and science of decision-making. It is
explicitly developed to improve our forecasts and to understand how to
put the future at the service of the present (Bernstein 1996, 1): it is a
distinctive feature of modern Western society, since the effort of
understanding risk, measuring it, and weighing its consequences has
transformed risk-taking into one of the prime catalysts that drives modern
Western society (Ibid.).
Nowadays risk theory (RT) is a tool for treating the issue of the ratio
costsbenefits in the decision-making process: risk management aims at
minimizing the costs of decisions made to protect a system against
possible damages. This protection has a cost: it requires that certain
resources of the system have to be locked, so that they cannot be employed
216 Chapter Thirteen

in other activities. This protection narrows the functions of the system,

since it requires a specific allocation of resources. RT tries to deal with
this conflict between long-term benefits and short-term costs: since
protection subtracts resources that would otherwise be directed towards
optimal outputs and benefits that became visible only when a harsh crisis
does happen, RT pursuits an optimization in terms of equilibrium
between risk-protection and risk-taking.
A seminal conceptualization of risk is the one provided by Frank
Knight, who separates risk and uncertainty in terms of measurability
(Knight 1921). Roughly, while risk is measurable, in the sense that the
distribution of the outcome in a group of instances is known (either
through calculation a priori or from statistics of past experience) (Knight
1921, 233), uncertainty is not: it represents situations where it is not
possible to form a group of instances. In this paper, I argue that it is not
possible to separate these two notions easily, since a reasonable risk
management cannot really selects and isolates a riskin virtue of its
dynamical nature. In particular it is not possible to separate risk and
uncertainty because the interpretative and qualitative side of risk does not
allow this separation without limiting the notion of risk and its
mathematical modeling. On the contrary, I will show that often it is just
the idea that a risk can be selected, isolated and handled that generates
damages or destructive courses of actions.
Furthermore, in our global, interconnected, competitive and complex
society it is getting more and more difficult to identify risks and,
accordingly, to take good decisions. But risks have to be taken and in
effect are continuously taken: in this sense, risk is the other side of
opportunity. Every decision offers an opportunity, but it does not come
free (i.e. risk-free). For instance, in finance is common to observe that
within every market there is an opportunity:

whether you are long or short does not matter, but how you play the game
does. People like John Paulson and Kyle Bass identified profitable
opportunities when they were but a brief idea in the heads of their
counterparts [...]. Through discovering these various opportunities, these
men and these men alone executed what I consider the greatest trades of all
time (Veneziani 2011, xvii).

The ability to manage risks and to take the opportunity that it offers can
provide a decisive competitive advantage in the course of actions that
follow a decision. Thus, there is no surprise that the conceptualization and
the mathematization of the notion of risk have widespread in the last few
decades. I will examine four main approaches to this issuethat is the
 Mathematizing Risk: A Heuristic Point of View 217

probabilistic one, the psychologistic, the fractal and the evolutionaryand I

will discuss their main hypotheses and limits. In the end, I will argue for a
heuristic view about risk.

1. The Probabilistic point of view

The orthodox approach to risk is the probabilistic one. It explicitly relies
on the hypothesis that risk is measurable and manageable and states that
risk is simply a random variable that maps unexpected future events into
numerical values representing profits and losses. A risk can be studied or
individually or as part of a stochastic process where it depends on previous
risks. The potential values of a risk have a probability distribution, which
even if it can never be observed exactly, can be obtained by data of past
losses about similar risks, if available.
In this way the probabilistic approach mathematizes riski.e.
generates a calculus and a measurement of it. Such a calculus assigns a
numerical value to it by quantifying over two variables: known probability
distribution and known function of damages (losses). In effect, it is
possible to quantify and rank risks simply by multiplying the numerical
value of these two variables. Thus the representation of risk in
probabilistic terms simply requires two mathematical variables:

1. A probability distribution about phenomenon;

2. A function that associates damages/benefits to each state.

It follows that the probabilized version of the risk of an event is equal to

the probability of failure times the consequence of failure. Hence, risk in
this sense is simply the probability that certain damages can occur.
A typical product of the probabilistic approach to risk is the well-
known VaR (Value@Risk), which is a standard tool for evaluating the
exposure of a given entity to a certain risk. VaR denotes the maximal loss
that is not exceeded with the probability p (where p, in general, can be
equal to 95% or to 99%) at a specific time t. It is not the maximal possible
loss: the VaR is a percentile of the profit and loss distribution such that,
with a small given probability, we can face that loss (or more) over the
fixed time horizon. It is designed in particular for financial instruments but
is not confined to them. Roughly, VaR works in the following way. The
first step is to set a confidence level (the value of p): you start off by
deciding how safe you need to be. Say you set a 95 % confidence level.
That means you want [] a 95 % probability that the losses will stay
below the danger point, and only a 5 percent chance they will break
218 Chapter Thirteen

through it (Mandelbrot 2006, 272). The second step is the calculation of

the VaR, which requires some additional hypotheses:

suppose you want to check the risk of your euro-dollar positions. With a
few strokes on your PC keyboard, you calculate the volatility of the euro-
dollar market, assuming the price changes follow the bell curve. Let us say
volatility is 10 percent. Then, with a few more strokes, you get your
answer: there is only a 5 percent chance that your portfolio will fall by
more than 12 percent (Ibid., 273).

The VaR does not require the tail of the distribution and its strategy can be
expressed by the following principle: to have a good knowledge of the
parameters when they are in some domain and to content oneself with an
upperbound of the probability of the exterior of the domain (Bouleau
2009, 3). But the point is that practically, at present, in the banks and in
most of the textbooks, VaR is computed with the help of explicit
probability distributions calibrated on reality by usual statistical
parametrical tests, hence hypotheses on tails of distributions are assumed
(Ibid.). In practice, the hypotheses on tails of distribution are crucial for
the calculation of the probabilistic version of risk and of VaR. There are
two main approaches to distributions: the mild and the wild one.
The mild approach relies on the hypothesis that every process sooner
or later will be aggregated in a Gaussian distribution. This implies that it is
completely defined by two parametersmean (V) and standard deviation
(). Moreover this hypothesis implies that, metaphorically speaking, the
world is considered as being made mostly of tiny pieces: the majority
of the elements (95%) is under the threshold of 2V. Thus, in Gaussian
distributions there is not much room for great exceptions: all the
constituents are of similar magnitude, not far from the mean, and all
together shape the process.
The wild approach conjectures that the process shows a Cauchy
probability density1, which implies the so called fat tails. In particular,
the process follows a power law and, contrary to exponential Gaussian
decay, has infinite variance. Hence, the aggregation of Cauchy variables is
not the result of the sum of elements of similar magnitude, but is the sum

dP 1
1 
The Cauchy distribution can be expressed by the equation dx Q (1  x 2 ) ,
dP 2
which for xo is dx  x .
 Mathematizing Risk: A Heuristic Point of View 219

of elements whose magnitude varies rapidly. Tellingly, some of the

elements can have a size comparable to the magnitude of the whole
aggregation.
It is worth noting that the distributions (Gauss and Cauchy) are stable,
since they are instances of the general class of Levy-stable distributions.
This means that the fundamental assumption here is that a process (like
prices of the stock markets) can be modeled by L-Stable distributions.
Stability is important because it exhibits heavy tails and asymmetry, and
these properties seem able to fit the data of the markets.
Unfortunately, the probabilistic approach and its tools have several
limits, both local and general.
As concerns the local ones, first of all the VaR is not sub-additive: if
VaR(X1) and VaR(X2) are the absolute values of the maximal losses on
the portfolios X1 and X2 at some given date and with some given
probability threshold, for the union of the two portfolios, where
compensations may occur, we would expect to have VaR(X1+X2)
Var(X1)+Var(X2) but this inequality is not fulfilled in general (Bouleau
2009, 4).
Moreover this approach has to face the problems of probability
distribution: once you are riding out on the far ends of a scaling
probability curve, the journey gets very rough. There is no limit to how
bad it could get (Mandelbrot 2004, 273). In effect, the distributions are
badly known in the region where the probability is low, in particular in the
neighborhood of infinity. Although well known today, the awareness of
this important question has been progressive during the history of sciences
and explains some present difficulties (Bouleau 2009, 1). In the above
euro-dollar example the flaw is manifest: the potential loss is actually far,
far greater than 12 percent. [] Assume the market cracks and you land in
the unlucky 5 percent portion of the probability curve: How much do you
lose?. Answer: you can lose everything. Events in the tail of the
distribution can have a size close to one of the entire distribution and,
hence, they are destructive. Even the VaR is able to detect this error and it
is possible to improve the probabilistic approach under this respect. Two
notable examples are boundary@risk (BaR) and Extreme Value Theory
(EVT).
They follow different strategies. In the former case, the strategy is to
set boundaries on data and parameters in order to control the output
without making any assumptions on the tails of distributions. In the latter
case, the strategy is to assume scaling and fat tails of the distributions.
In effect, the BaR aims at avoiding any assumption on the tails of the
distributions [] by reasoning inside a domain with explicit boundary
220 Chapter Thirteen

(Ibid., 7). The boundaries are fixed by frontiers, which are defined by
specifying a domain to all the quantities: to the data, to the parameters,
and, thanks to the model to the computed outputs (Ibid.). But in this sense
the BaR is simply a special case of the VaR, since it is not able to treat the
phenomenon when the boundary is over-crossed: as a matter of fact it tells
us nothing on what happens when the boundary is crossed (Ibid,, 8).
Thus, it does not really improve the main strategy of the VaR and the
probabilistic approach.
On the other hand, the strategy of the EVT (e.g. Burry 1975, Coles
2001) is able to overcome this flaw of the VaR (and the BaR), by
generating the best possible estimate of the tail area of the distribution. In
order to do that, the EVT explicitly conjectures that the phenomena vary
wildly and exhibits fat tails and scaling, so that big and overwhelming
losses are incorporated in the calculus (e.g. with block-maxima and peak-
over-threshold method) and the risk is better modeled and mitigated. The
crucial point of the theory is the determination of a cut-off between the
central part of the distribution and the upper tail, i.e. a numerical value that
separates ordinary realizations of the random variable considered from
extreme realizations of the same variable.
Unfortunately a basic assumption of EVT is that data are independent
and identically-distributed, which does not hold for most financial time
series: it ignores another big source of risk, namely the so-called long-term
dependence, i.e. the clustering of extreme values due to dependence or
the tendency of bad news to come in flocks. A bank that weathers one
crisis may not survive a second or a third (Mandelbrot 2006, 273).
Therefore, in this case, EVT generates incorrect valuation, which in turn
produces the two most unwanted results: unexpected losses or excessively
conservative positions.2
Despite these improvements, the probabilistic approach is affected by
strong limits, which will be better clarified at the end of the paper. At this
stage, we can note that the representation of risk by means of two variables
(probability of occurrence and size of damage) can be misleading, since (i)
it is not possible to know the tail of a distributionbecause rare events are
not well-covered (and can be covered) by the dataand (ii) the estimation
of the damages (their magnitude) is its only concern. Moreover even
assuming that we can have a knowledge of (i) and (ii), the model is
simplistic because it is hiding the reasons why we are interested to these

2
This weakness is handled by means of a toolkit able to take dependence of the
data into consideration, e.g. the so called extremal index .
 Mathematizing Risk: A Heuristic Point of View 221

events by doing as if they could be translated automatically and

objectively into costs (Bouleau 2009, 5).
The fundamental limit of this approach, in the end, relies in the process
of probabilization and mathematization itself, as Bouleau underlines
(Bouleau 2009 and 2011). In fact, probabilizing a situation is
fundamentally an ousting, an erasing of the meaning, while the risk
analysis is necessarily understanding of interpretations (Ibid.). This
erasing could be defended by arguing that it is necessary in order produce
a calculus, i.e. a computation of risk that associates numbers to certain
events. Unfortunately, computations, even with methodological
precautions, have the innate defect to hide ignorance (Ibid.). And there is
a lot of ignorance hidden in risk and risky situations. For instance:

we dont know precisely to quantify neither the return risk of a loan, nor
the market or liquidity risk, nor the risks due to human mistakes or due to a
regulation change, very accurate computations are mixed with rough
estimates hopping they will have no sensitive consequences on the result.
During the manufacturing of packages of several credits, a standardization
of the description occurs [] which moves away from reality. At the limit,
a mortgage loan in Iowa or Kansas is thought in the same manner as a
credit at New York on Madison avenue if they are both well assessed
(Ibid., 5-6).

2. The Psychologistic point of view

The psychologistic view on risk is based on the idea that risky situations
require choices under conditions that violate the so-called concavity
hypothesisa cornerstone of the Expected Utility Theory (EUT).
The concavity hypothesis simply states that, in a decision involving
outcomes whose probabilities are known or knowable, the preferences of a
subject are such that the utility function is concave for gains and convex
for losses. In particular, the utility function is (i) defined on deviations
from the reference point; (ii) generally concave for gains and commonly
convex for losses; (iii) steeper for losses than for gains (Kahneman
Tversky 1979, 279).
More specifically, the concavity for gains represents diminishing
marginal utility: the increase in the amount of money, ceteris paribus,
implies a decline in the marginal utility of each additional unitary amount
of wealth. Each increment to wealth implies progressively smaller
increments to the utility. Thus, the slope of the utility function gets flatter
as wealth increases (in this sense it is concave). So, concave utility
functions represent risk-aversion (KahnemanTversky 1979, 264). A risk-
222 Chapter Thirteen

averse choice prefers a definite outcome to a risky one, even if the risky
outcome has the same mathematical expectation3.

u u

m m
a. concavity of the utility function b. convexity of the utility function

Fig.1

Similarly, the hypothesis of convexity for losses represents

diminishing marginal sensitivity to increasing losses. Thus, in this respect
convexity expresses risk-seeking, since the slope of the utility function
gets steeper as the amount of money of losses increases.
These tenets characterize EUT, which is a mathematical cornerstone in
decisions-making under risk (see modern portfolio theory). Unfortunately,
it is possible to demonstrate several phenomena which violate these
tenets of expected utility theory (Ibid., 264). For instance, various forms
of insurance refute that the utility function for money is concave
everywhere: people choice insurance programs that offer limited
coverage with low or zero deductible over comparable policies that offer
higher maximal coverage with higher deductibles-contrary to risk
aversion (Ibid.).
In this way, the psychologistic point of view on risk captures an
essential flaw of the EUT:

risk has been described as derived from risk aversion as a result of the
structure of choices under uncertainty with a concavity of the muddled

3
For example, when asked to choose one of these cases:
1) with probability 0.5 you get 100 and with probability 0.5 you get nothing;
2) you can get 50 for sure.
Usually people choose the not risky case (i.e. 2). It means that if they have utility
function u, then u(50)0.5 u(100)+0.5u(0). Since concave functions satisfy
these inequalities, the property is called concavity.
 Mathematizing Risk: A Heuristic Point of View 223

concept of utility of payoff, [] but this utility business never led

anywhere except the circularity [] risk is what risk-averters hate.
Indeed limiting risk to aversion to concavity of choices is a quite unhappy
result the utility curve cannot be possibly monotone concave, but rather,
like everything in nature necessarily bounded on both sides, the left and the
right, convex-concave and, as Kahneman and Tversky (1979) have
debunked, both path dependent and mixed in its nonlinearity (Taleb
Douady 2012, 2).

The psychologistic view draws on the probabilistic approach and

accordingly shares some of its limits. In particular, it does not help in the
estimation of extreme events, as the use of utility functions

to escape from the concept of mathematical expectation, although yielding

true services to represent the behavior of the agents with their subjective
probabilities (and avoid St Petersburg paradox), doesn't solve at all the
problem of the tails of probability distributions because the utility function
also is badly known on extreme events (Bouleau 2009, 2).

3. The Fractal point of view

The fractal approach is based on the idea that risk cannot be understood,
assessed and managed without a radical change in the way the social and
natural phenomena are viewed. In effect, the fractal view developed by
Mandelbrot looks at the phenomena in a new way: discontinuity,
concentration and scaling are considered essential features of phenomena
and it is impossible to make sense of the future course of events, forecasts,
and accordingly treat risk, without taking them into close consideration.
Discontinuity is the property by which processes change suddenly,
randomly jumping between very different and distant values. It is worth
noting that the strongest short argument to account for discontinuity (like
IBM moving down, and later up, by 10%, then 13.2%) involves the fact
that exogenous stimuli can be very large. But a large exogenous stimulus
need not come out as one single piece of news (Mandelbrot 1997, 55).
Concentration is the property by which the big events are such to occur
in a small interval of time. For instance, a very small number of days are
responsible for the majority of stock market movements: seven trading
days can generate about half the returns of a decade.
Scaling is the property of statistical invariance under dilation or
reduction in space and time (self-affinity). This characterizes fractal
objects, whose small parts resemble the whole (e.g. the veins in leaves
look like branches; branches look like small trees, etc.). This property
224 Chapter Thirteen

emerges also in economic data, where parts often relate to the whole
according to a power law. For example, in economics a power-law
distribution implies that the likelihood of a daily or weekly drop exceeding
20% can be predicted from the frequency of drops exceeding 10%, and
that the same ratio applies to a 10% vs. a 5% drop (MandelbrotTaleb
2010, 99).
The main point of the fractal view is that it is impossible to assess a
risk without considering these three properties. A huge amount of data and
findings show that a consistent fraction of our world exhibits a winner-
take-all extreme concentration. In markets, there is a predominance of
random jumps, and discontinuity and concentration are major ingredients
in a realistic evaluation of risks (Mandelbrot 1997, 56), while the
common tools of finance are designed for random walks, that is for a
process moving in small, smooth steps.
These properties reshape the assessment of risks and its calculation:
take a portfolio and compare the risks using the Random Walk on the
Street and the M 1963 model. The former estimates the risks as small, say,
one thousandth, one millionth or less, while the latter may estimate the
same risk to be a hundredth, a tenth or more (Mandelbrot 1997, 56). This
implies that if we define reward and risk as a ratio, then the standard
arithmetic must be wrong. The denominator, risk, is bigger than generally
acknowledged; and so the outcome is bound to disappoint. Better
assessment of that risk, and better understanding of how risk drives
markets, is a goal of much of my work (Mandelbrot 2006, 5).
Another key factor in risk identification and assessment is dependence:
for instance markets keep the memory of past moves, particularly of
volatile days, and act according to such memory (MandelbrotTaleb
2005, 99). In other words, volatility generates volatility in rapid sequences,
producing clusters: this is not an impossibly difficult or obscure
framework for understanding markets. In fact, it accords better with
intuition and observed reality than the bell-curve finance that still
dominates the discourse of both academics and many market players
(Ibid.).
So the fractal view offers a new approach to risk and its calculation.
Moreover, in order to treat it properly and to reflect these features
(discontinuity, concentration, scaling and dependence) in the formal treatment
of it, new concepts and new pieces of mathematics are producedlike
fractal mathematics. As a consequence, the fractal view suggests a new
strategy. First, in order to manage risk you have to diversify as broadly as
you can. In financial markets, for instance, since long-run returns are
dominated by a small number of investments, there is a high probability
 Mathematizing Risk: A Heuristic Point of View 225

that the risk of missing them must be mitigated by investing as broadly as

possible. The exposure to risks is larger than the one conjectured by the
received approaches, and management has to consider this.
Let us consider a typical risk management problem, such as the
calculation of a reservoira typical strategy for managing risk that covers
us against the risk of damages, losses and disasters. It applies to many
domains (banking, hydrology, meteorology, etc.), for to calculate the
needed amount of reservoirs is essential for a good management of risk.
Let us discuss an example in hydrology, in particular the dam of Nile
examined by Mandelbrot (Mandelbrot 2006, 178-80). According to the
probabilistic approach, which employs statistic independence and bell
curve, if you want to replace, say, a twenty-five-year-old dam with a new
and higher one that protects you against one hundred years of flood, the
new dam should be twice as high as the old (since the timescale of the new
dam is four times that of the old). Unfortunately, this calculation is
completely wrongand can lead to disasters. The dam has to be higher
than this and its dimension has to be calculated with a completely different
piece of mathematics - the Hursts formula, which takes into account not
only the size of the floods, but also their precise sequences. The exposure
to risk is higher than the one foreseen by a probabilistic point of view and,
accordingly, your protection has to be much greater.
One of the central strategies of fractal viewpoint, the idea of a broad
diversification, shapes another influential approach, the evolutionary one.

4. The Evolutionary point of view

The evolutionary viewpoint on risk is based on the analogy between risk
management and biology, in particular the natural process of reliability
and robustness. Nature and the living beings can be seen as examples of
handling risk in order to increase the chances of survival and, hence, the
relation between organisms and environment can shed lights on the notion
of risk and risk management.
The evolutionary approach has been recently popularized by Taleb
(Taleb 2010) who argues that Mother Nature teach us how to manage
risk, not because of the optimality of evolution, but entirely for
epistemological reasons, how we should deal with a complex system with
opaque causal links and complicated interactions (Taleb 2010, 362). In
effect, Nature is an example of a complex system, plenty of webs of
interdependences, nonlinearities, which create a robust ecology. The
solutions that the evolutionary arguments offer us to manage risks are
essentially based on two strategies: control of size and redundancy.
226 Chapter Thirteen

On the former, it is based on the connection between the risk and the
size of an entity (organism). Evolution offers an interesting perspective on
it: size generates risks, and that is why Mother Nature does not like
anything too big (Ibid., 349). An increment in size of an entity can
increase its exposure to risk and eventually lead to a disaster (e.g.
extinction). This simple evolutionary fact offers a hint for the management
of social structures. For instance, it applies to a well-known process in
economics, that is the merger-mania and the economy of scale, according
to which a company gains money and efficiency as it becomes larger and
larger. It is a popular concept, but it is prevalent in the collective
consciousness without evidence for it; in fact, the evidence would suggest
the opposite. Yet, for obvious reasons, people keep doing these mergers
they are not good for companies, they are good for Wall Street bonuses; a
company getting larger is good for the CEO (Ibid., 350). The economy of
scale shows us that man-made structures should not be too large: in fact,
the continuous enlargement of a company seems to create islands of
efficiency, but it also makes it more vulnerable to contingencies.
On the latter, the evolutionary view notes that natural phenomena are
plenty of redundancies, like defensive, functional and spandrel
redundancy. The defensive redundancy increases our chance to survive
under adversities by means of replacements. For example, the redundancy
of organs (eyes, lungs) provides us extra capacity than the one needed in
ordinary circumstances. Thus, redundancy equals insurance, and the
apparent inefficiencies are associated with the costs of maintaining these
spare parts and the energy needed to keep them around in spite of their
idleness (Ibid., 346). In this sense redundancy is the opposite of nave
optimization, which characterizes the orthodox economics that is
essentially based on the notion mathematized by Paul Samuelson. As a
matter of fact, in the light of this theory it is inefficient, for example, to
maintain two lungs, due to the costs required by the transportation of them,
but such optimization would, eventually, kill you, after the first accident,
the first outlier (Ibid.).
Biology also inspires functional and spandrel redundancy. In the
first case, the same function can be obtained by two different structures,
and not by two identical elements. In the second case, an organ can
perform a function that is not its current central one.
The concept of redundancy is not a theoretical novelty in RT, since it is
a well-known strategy to build reliable systems in engineering, i.e.
reliability engineering (see Pieruschla 1963, Roberts 1964, Sandler 1963,
BarlowProschan 1965, Bazovsky 1961). Reliability can be broadly
defined as the capability of a system to perform its functions over a period
 Mathematizing Risk: A Heuristic Point of View 227

of time under conditions of stress. Redundancy is a way to obtain this goal

as it offers a way to reduce weaknesses (failures) or mitigate their effect.
In order to be established, reliability has to be derived from the data, for
instance the total number of failures occurring during a given period. Since
the 1960s the notion of redundancy has been successfully employed in
aircraft maintenance and operations. This strategy aims at detecting
weaknesses in a system in order to eliminate or minimize the likelihood of
their occurrence. In particular, redundancy here is as a form of parallelism,
a duplication of critical components or functions of a system, and it has
been a crucial strategy to face risks in reliability engineering. In this
respect, redundancy is generally broken down in two ways, just like in
biology: the active one and the passive one. These two ways serve the
same aim: to prevent a process from exceeding certain limits without
human intervention by using extra-capacity. More specifically, the passive
form uses excess capacity to reduce the impact of component failures. A
notable example is the extra strength of cabling and structures used in
bridges. Such an extra strengthor margin of safetyallows some structural
components to fail without bridge collapse. This is analogous to the
passive form of redundancy in our body: vision loss in one eye does not
cause blindness (even though depth perception is compromised). The
calculation of reliability shows how to handle and mitigate risks and how
it can generate a profit. Nevertheless, the mathematics and the assumptions
of this approach are not very different from the ones of the received
approaches.
Let us consider an example from engineering design of plantsthe
concept of r/n systems. Such a concept is a common way to handle risk by
increasing the reliability of systems using redundancy. Its strategy is to
multiply the critical components in such a way that systems employ r units
from a total population of n to be available in a processe.g. a system with
four units and requiring three of them to be operable. So, the calculation of
the overall reliability of an r/n system can be expressed by a simple
cumulative binomial distribution:

m
u!
Re = p a (1  p )u  a ,
a 0 r !(u  a !)

where Re is the reliability of a system given the actual number of failures

(a), which is less or equal to the maximum allowable (m) of failures; u is
the total number of units in the system and p is the probability of survival,
or the subcomponent reliability, for a given time t. Assuming that the
228 Chapter Thirteen

subcomponents reliability is 0.90, the solution of the equation shows that

the likelihood of the system to function over the period of time t in the
stated conditions is p=0.9477 (about 95%). Therefore, the redundancy
really improves the reliability of the system since the reliability of the
single components is, by hypothesis, 90%. Nevertheless the reliability
analysis requires that the data about failures behaving as a bell curve
hence employing the assumptions of the received approaches about tails.
On the other side, the active redundancy aims at eliminating the
processes decline (and their performances) by monitoring the performance
of the individual components. A common way of monitoring is a switch
between the components, for instance in electrical power distribution.4
The bottom line: to manage risk by controlling the two main variables
of the mathematical risk, i.e. the probability of failure and the
consequences of failure.
Another key concept connected to notion of size in risk-management is
density. In this respect, the evolutionary view moves from the
observation that Mother Nature does not like too much connectivity and
globalizationbiological, cultural, or economic (Ibid., 351). An interesting
example is the distribution of species: simply, larger environments are
more scalable than smaller onesallowing the biggest to get even bigger,
at the expense of the smallest (Ibid.). Nevertheless, on the other side,
Taleb notes that Nature provides evidence that small islands have many
more species per square meter than larger ones, and, of course, than
continents. As we travel more on this planet, epidemics will be more
acutewe will have a germ population dominated by a few numbers, and
the successful killer will spread vastly more effectively (Ibid., 352). So,
lack of variety, concentration and exposure to risks are interconnected
phenomena: a system with little redundancy and many interconnections
can completely fail also if only one part does not perform as intended,
leading to a disaster (e.g. Fisher 2012).
Evolution offers a solution to this problem that enables the
construction of more stable and safer systems: to limit the size of its units,
but not their interactions. This principle can be applied to man-made
structures in order to obtain a better management of risk.

4
Typically, each of the several power lines has both a monitor detecting overload
and circuit breakers. The combination of power lines generates redundancy, i.e. an
extra capacity that eliminates or mitigates risks. Circuit breakers disconnect a
power line when the monitors detect an overload and the power is redistributed
across the remaining lines.
 Mathematizing Risk: A Heuristic Point of View 229

Moreover, the evolutionary approach shows the reflexive and dynamic

features of risk. The notion of adaptation can be used to model and handle
risk: the environment-organism relation is not static, in the sense that
environment is not fixed, but is continually modified by the activity of the
organisms. As the activity of the population unfolds, the environment
changes, creating new pressures on the population and so on. Such a
dynamic process holds in particular for human beings due to the cultural
activity. Since humans actively modify the environment by means of
technology, it continually changes in response to their activity. The
cultural changes are deeper, more extensive and faster than natural
changes. So risks are continually changing, affecting the system and being
affected by it: the detection of new risks in a system requires a response
and this, in turn, changes the system itself, exposing it to new potential
risks as a consequence. In the same way, an organism changes its process
of adaptation on the basis of the changes it provokes in the environment.
A fruitful development of this approach is the one based on a co-
evolutionary view, supported by D. Sornette (see e.g. SornetteKovalenko
2012), who extends the notion of risk in order to include an endogenous5
view and offers a new measure to predict risksTime@Risk. In this view,
risk is co-evolutionary related to the notion of stress6, so that its
definition has considered a new property: a vulnerability and related
counter-measures and mitigation techniques, that specify how disruptive is
the potential stressor to the system (SornetteKovalenko 2012, 4). So,
while the two properties that define risk in the probabilistic approach are
external forces influencing the system (stressors), here we have a third one
that is internal, and collectively

they control the overall losses that the stressor can bring to the system. As
a consequence, risk is understood as the combination of these three
characteristics of the potential stressor. Thus, risk is equal or proportional
to the possible internal response of the system, and therefore is a proxy for
the stress developing within the system (Ibid.)

In this sense, risk is the complement of resilience:

On the one hand, risk provides a measure of the nature and amplitude of
stressors, present and future. As a consequence, from risk measurements,

5
The endogenous forces are responsible of a new kind of extreme event that
Sornette calls dragon-king.
6
A stress is defined here as an internal response/reaction of a system to a
perturbation called stressor.
230 Chapter Thirteen

one can infer the possible level of stress that may develop within the
system. On the other hand, resilience characterizes the internal stress
response within the system, quantified by the capacity of a system to cope
with stressors and remain essentially the same. In other words, resilience is
the amount of stress that a system can bear without a considerable
transformation (Ibid., 6).

The co-evolutionary view enables us to make predictions about risks (see

Sornette 2003). Just like in biology, there are symptoms of stresssuch as
attention, mobilization of resources, and concentrations on key areas,
which can be used to forecast risks and react in a dynamic way by building
up resilient strategies. More specifically:

you could prepare [] on what is not working and what could be

improved or changed. You could start a process towards building stronger
resilience, catalyzed by the knowledge of the nature and severity of the
stressors forecasted to come. [] Advanced diagnostics could
revolutionize risk management by pushing us into action to build defenses.
A working advanced diagnostic system would not be static, but would
provide continuous updates on possible scenarios and their probabilistic
weights, so that a culture of preparedness and adaptation be promoted []
Here, we go one step further by suggesting that forecasting the occurrence
of crises promotes the evolution of the system towards a higher level of
resilience that could not be achieved even by evolution (which is backward
looking). Advanced diagnostics of crises constitutes the next level of
evolution for cognizant creatures who use advanced scientific tools to
forecast their future (Ibid., 23).

5. The heuristic point of view

The heuristic view draws on Sornettes view and on the idea that risk is a
dynamic process, the result of a continuous process of interpretation of
phenomena and problem-solving. Thus, it contrasts the other points of
view, which rely on a static approach that arbitrarily stops the process of
interpretation and problem-solving. In effect, all these viewpoints do not
take into consideration the interpretative and qualitative nature of risk and,
accordingly, the risk is represented and modeled in an improper way. They
treat it in a static way: the process of interpretation is considered closed,
and the risks have a precise meaning as they are identified, known and can
be quantified.
Of course it is possible to argue that a static point of view, and the
closure of process of interpretation, is necessary in order to quantify and
translate into numbers the interpretation and meaning of risk. But the point
 Mathematizing Risk: A Heuristic Point of View 231

here is that especially the notion of risk is a concept that cannot be

captured in this way. As a matter of fact, the process of interpretation of
risk instead of being closed, it is in permanent innovation. As soon as a
new reading appears, it generates new risks which are only perceived by
those ones who understand it (Bouleau 2009, 6). Let us consider as
example the financial crisis of 2008. In this case, the risks were in the
several possible interpretations of some correlationsin particular the
correlation between two variables such as the real estate prices and
earning. In effect, if in 2006, nobody sees the increase of the real estate
prices and the decrease of the households savings in the US as a
phenomenon allowing several readings, the corresponding risk is not
detected (Ibid., 6). Risk is the result of correlations, that is the possible
relations between variables affecting the behavior of a system. These
correlations obviously do not admit a single, unique interpretation and, in
fact, they can be explained in several ways: in principle there are infinite
variables that can affect a phenomenon and to find these variables is
crucial to identify risks. Hence, the interpretation and the meaning of the
phenomenon are crucial to determine if and what risk can be identified. So
what a risk is depends on the chosen interpretation and the meaning given
to the process itself. In the meaning of an event lays a risk. And meaning
is generated by the choice of the variables to investigate and connect. For
instance:

suppose a particular type of cancer be enumerated within the population.

This subset of the suffering people is the concrete event. The ratio to the
whole population will be taken for an estimate of the risk. If it would
happen that observation and inquiry reveal that a significant part of the
concerned persons have taken, say cannabis, twenty years before, then all
the cannabis consumers become potential patient. If another common
property of the sick persons is pointed out-e.g. they have used mobile
phones as soon as these were available, then almost the whole population
is now threatened. The risk changes when the interpretation modifies the
meaning of the event (Ibid.).

The fact that risk is the result of interpretation and meaning implies
that it is not possible even only to affirm that the tails of distributions are
badly known: they are never granted in the investigation of the risks. But
without the possibility to make such a hypothesis on tails, it is not possible
to mathematize a phenomenonespecially risk. To be applied, mathematics
requires that the process of interpretation is closed or could be closed.
Unfortunately in social domains continuously the agents think to improve
their knowledge and forecasting, and for this re-interpret all the
232 Chapter Thirteen

information they capture (Ibid., 9). This process generates continually

new readings of the phenomena, new correlations between their variables,
which in turn generate new risks and so on. The tails of distribution, and
then risks, are by nature temporary and fluctuating depending on the
interpretative knowledge that the agents acquire by their understanding of
economical phenomena (Ibid., 6).
Therefore, in order to work, the mathematization of risk needs a trick
the hypothesis on the tails of the distributions, which de facto ignores all
the problems and the difficulties of the interpretative side of risk. This
follows from the fact that mathematization requires the arbitrary stop of
the multiplicative process of interpretation.
On the contrary, as our knowledge is extended, new interpretations of a
given phenomenon can be generated and, accordingly, new risks can be
detected: any advance in the knowledge makes us discover features of the
world on which we were not aware before and, by this, generate new risks.
Whatever way is taken for mathematizing the risks, they are congealed,
canned in a box which hides the interpretative (Ibid., 6).
Therefore, risk is a notion that cannot be fixed a priori, once and for
all, from the very beginning of a process. It is intrinsically dynamical and
multiplicative. New knowledge generates new interpretations and
viewpoints about a phenomenon (i.e. new entity and new connections
between entities), which in turn generates new risks that can be forecasted,
taken in consideration and handled. In particular, the identification and
management of risks is reflexive: once a risk of a system is detected, the
decisions taken to handle it change the system itself and this, in turn,
generates new risks for the system. New consequences and branches in the
decision-tree are created. A notable example, again, is the 2008 financial
crisis: the derivatives designed to handle and mitigate the risks at a local
level, turned out to increase and in some cases generate systemic risks.
The dynamic and interpretative nature of risk is manifest in the relation
between innovation and risk. Innovation (e.g. a new technology) not only
is essential to manage risk, but it also generates new and unknown risks. In
effect, risk is the product of the constant search for novelty and innovation
in order to solve problems and amplify our knowledge that enables control
on the environment. To take a risk may increase our chances of survival,
and once an innovation is produced, it opens new lines of consequences,
generating new risks. This explains why risk is a dynamic concept and is
inexhaustible: we can only have a partial and incomplete representation
of it. The bottom line: an innovation can change the dynamics of the
process itself, and accordingly new risks can appear as new interpretations
of it are generated and problems solved.
 Mathematizing Risk: A Heuristic Point of View 233

This process is particularly evident in the social domains. Let consider

one more time the 2008 financial crisis. Insurances, options, futures,
derivatives etc. are financial innovations generated in the last two decades
to better treat and manage financial risksnay a firms exposure to
financial riskand stabilize the economic system. Financial risks can be
represented as the variability in cash flows and market values generated by
changes in the commodity prices, interest rates and exchange rates. In
effect, these financial innovations seem to serve their purpose as they had
brought about a fundamental improvement in the efficiency of the global
capital market, allowing risk to be allocated to those best able to bear it
(Ferguson 2008, 6), in particular by improving the transfer of risks of
mortgage credits. More specifically a prominent portion of mortgages risks
has been transferred via securitization to worldwide investors. This
broader transfer has really improved the management of some financial
risks, as it stabilizes the system since the banks are not alone in bearing the
ensuing losses. These innovations revealed to be crucial for the markets
and the real economics, since they were a means of competitive
advantageincreasing the shareholder value. But at the same time they
have generated new and also unknown risks, like credit risk, counterparty
risk, model risk, rating agency risk, settlement risk (Gibson 2007),
showing at what extent financial innovations can held unknown risks
(VoineaAnton 2009, 141).
Moreover, and more deeply, these financial innovations changed the
interpretations and the dynamics of the market. In particular, the massive
transfer of risks changed the dynamics of the market, since credit risks
were not evaluated only by a small amount of local, centralized experts,
but was examined by thousands of worldwide participants.
Thus, the relation between risk and innovation is such that the latter,
providing a new reading of a phenomenon, changes the dynamics of that
phenomenon: the use of the same tool by different operators can converge
their decisions on a single option (or a small set of options), falsifying the
calculation of risk. For instance, in the case of the financial crisis of 2008,
the use of the same model (i.e. VaR) have converged the conclusions of
the investors, at the same time, to a similar decision, increasing systematic
risk. This example shows the limits of the mathematization of risk and its
quantitative models, since it is reasonable to state that in the 2008 financial
crisis the standard quantitative models for risk management evaluation/
assessment and the users of these models (analysts) underestimated the
systematic nature of risks (Ibid., 143).
This process is very similar to the one called reflexivity by George
Soros (Soros 1987): once an innovative tool for evaluating risk and the
234 Chapter Thirteen

consequent decision is taken, it affects and modifies the process itself,

reinforcing itself and increasing the exposure to certain risks. This is a
qualitative process, which cannot be expressed simply by means of
mathematical modeling. In effect, it was just by means of a qualitative
analysis that, for instance, Kayle Bass anticipated the financial crisis of
2008betting against it and eventually earning a huge amount of moneys.
On the basis of the unreasonable divergence between two variables, the
stationary earning and savings of American citizens and the increasing
prices of houses, Bass investigated the process of stipulation of mortgages.
In particular:

Bass reminded us that mortgages are more than just numbers for
accountants and actuaries to crunch. They also contain a qualitative aspect:
What you had to figure out was theres the quantitative aspect of things
and theres the qualitative aspect of things. And the quantitative aspect,
everybody had. Everybody had [all the] terabytes of mortgage data,
modeling software that they could buy. But the qualitative aspect was kind
of: who were the originators that had literally no standards at all?...So
what we did was, we went out and we found the bad guys (Veneziani
2011, 9).

A qualitative, verbal approach and the interpretation of the origin of

mortgages revealed new aspects of the phenomenon and then new risks
that could not be revealed by a mere quantitative approach. In effect, the
discovery made by Bass has been achieved by a bottom-up approach,
starting from arguments in a non-mathematical form and field, and trying
to transform them into mathematical arguments. So Bass found plenty of
bad guys. He investigated the lifestyles and backgrounds of these
mortgage originators and their companies. Who were they? What was their
incentive to offer someone making $30,000 a year a $500,000 loan?
(Ibid.). After this investigation Bass was able to exploit the opportunities
created by risks hidden in housing market. Thus, this is a clear example of
how a potential risk can generate a potential profit, both in economic and
epistemic terms.
The change generated by these financial innovations was so deep that
systemic risks emerged and spread all over the market much easier and
faster. In the summer of 2007, as soon as doubts about the reliability of the
ratings and the prices of the derivatives became public, a huge and
simultaneous exit of investors from the market occurred, creating a steep
fall of the prices and of the market liquidity. Moreover, due to the nature
of the securitization, this crisis has been extended to other fractions of the
marketssuch as commercial buildings, credits to finance acquisitions,
 Mathematizing Risk: A Heuristic Point of View 235

etc.spreading huge losses at the speed of light. Thus not only the risks
were out of control, but were also able to rapidly pass from one section of
the market to the other. Other crucial discoveries of this kind (and
economic profits) were obtained in a similar way by George Soros, who
heavily relied on qualitative analysis and bottom-up approach in order to
provide new reading of the phenomena and, consequently, to discover
risks and opportunities (e.g. Soros 1987).
A heuristic view on risk helps us in the difficult task of interpreting
phenomena by individuating new and unknown risks. In order to do this,
we need procedures for identifying and choosingthat is discoveringthe
relevant variables for the analysis of a phenomenon. In effect, these
variables are crucial for forecasting possible risks. Of course this step is
problematic, since it is hazardous and never guaranteed. Nevertheless the
heuristic view offers a way for discovering risk (and potential gains): the
ampliative inferences (analogies, metaphors, etc.) that enable us to
progress from a problem to the hypotheses that can solve it. In effect,
analogy and in general ampliative inferences are a means for identifying
new risks. This is common knowledge in basic risk-management (e.g.
KowszunStruijve 2005). The strategy to identify risks in non-trivial
situations relies on past experience of kind of risks that have been
identified within the field under investigation and relies on analogies with
other similar fields. In order to do that, it is necessary to start bottom-up,
from arguments that are non-mathematical, and to transform them into
mathematical arguments. This method is the only way to represent features
of a field in its formal treatment. In some cases, it enables important
discoveries and predictions to be made and exploited (e.g. Bass and
Soros), while in the most innovative cases, it allows new mathematics to
be generatedas with Mandelbrots fractal theory.

6. Conclusions
The concept of risk is an interesting case in the study of the effectiveness
of the heuristic view applied to mathematization. Such a view explains the
reasons for the multiple, infinite characterizations of the notion of risk
and the weakness of the other approaches to it. While the latter rely on a
top-down, static view that aims at employing known mathematics to make
sense of phenomena, the heuristic approach aims at translating the
arguments of a field from the bottom-up into a mathematical-formal
treatment of it. Such a different perspective is crucial since it determines
the choice of the variables over which to quantify. In principle, the choice
of the standard approach, which limits the definition of risk to two
236 Chapter Thirteen

parameters (frequency and magnitude), stems from known mathematics

(like probability and statistics) and not from the demand for a genuine
description and understanding of the phenomenon. Most of known
mathematics is designed to model physical phenomena and not social
phenomena, and this fact in part explains why it is not able to express and
properly model social phenomena. So, except for few and very limited
domains (like engineering), risk in the traditional sense cannot be isolated
and modeled. In this sense, it is not separable from the notion of
uncertainty, since its dynamic nature is such that you cannot consider it as
a definitely closed phenomenonmeasurable and quantifiable.
On the contrary, in order to select the variables for the identification of
risks, it is necessary to put forward a qualitative analysis (since this is the
only way to incorporate the features of the field under investigation in
mathematical arguments), which requires the interaction with both
experience and other hypotheses. As risk interacts with experience and
conjectures, new knowledge is produced and, in turn, new properties and
determinations of risk emerge, so that its representation is always
provisional, partial, and incomplete. What defines a particular risk depends
on this process. An event that seems to be risky might not be so in the
light of a different interpretation and reading and vice versa.
Since the process of reading and interpreting phenomena cannot be
closed, the mathematization of risk is problematic and is always under
revision: risk is a complex object from a mathematical point of view,
whose characterization is continually open to new determinations. In this
sense, this concept shows that mathematical objects are hypotheses
tentatively introduced to solve problems. A mathematical object is the
hypothesis that a certain condition is satisfiable (Cellucci 2013a, 103).
Nothing more. Nothing less. So, the hypotheses through which
mathematical objects are introduced characterize their identity. The
identity of a mathematical object can be characterized differently by
different hypotheses, and this implies that hypotheses do not
characterize the identity of mathematical objects completely and
conclusively, but only partially and provisionally. For the identity of
mathematical objects is always open to receiving new determinations
through interactions between hypotheses and experience (Ibid., 104).
In effect, as risk interacts with other experiences and factors, its
mathematical representation changes:

new properties of mathematical objects may emerge since, by putting the
hypotheses, which introduce them in relation with other things,
mathematical objects may get new determinations. Such new properties
may also suggest to modify or completely replace the hypothesis through
 Mathematizing Risk: A Heuristic Point of View 237

which the identity of mathematical objects has been characterized. This is a

potentially infinite process, so mathematical objects are inexhaustible. This
is the lesson of Gdels first incompleteness theorem (Ibid.).

This lesson turns out to be useful also for risk management, since the
exposure to risk is not only greater than the one supposed by the received
approaches, but also the dynamical nature of risk is such that its
mathematization is continually open. In order to capture some of its
features we need to re-interpret and read it over and over again and only
after we can try to express it in mathematical fashion. Sometimes this
passage can be accomplished with the help of known mathematics, while
sometimes it requires the generation of new mathematics.

References
Barlow, R.E.Proschan, F. (1965). Mathematical Theory of Reliability,
New York: John Wiley & Sons
Bazovsky, I. (1961). Reliability Theory and Practice. Englewood Cliffs:
PrenticeHall
Bernstein, L. (1996). Against the Gods. The remarkable history of risk.
New York: Wiley & Sons
Bouleau, N. (2009). Mathematization of risks and economic studies in
global change modelling. Halshs-00435959
. (2011). Mathematics and real-world knowledge. Real-world economics
review, 57, 90-104
Burry, K.V. (1975). Statistical Methods in Applied Science. John Wiley &
Sons
Cellucci, C. (2005). Filosofia e matematica. Roma: Laterza
. (2008). Perch ancora la filosofia. Roma: Laterza
. (2013a). Top-Down and Bottom-Up Philosophy of Mathematics,
Foundations of Science, 18(1), 93-106
. (2013b). Rethinking Logic. Logic in Relation to Mathematics,
Evolution, and Method. Springer: Berlin (to appear)
Coles, S. (2001). An Introduction to Statistical Modelling of Extreme
Values. Springer, London
Ferguson, N. (2008). The ascent of money. New York: Penguin Press
Fisher, T. (2012). Designing to avoid disasters: the nature of fracture-
critical design. New York: Routledge
Gibson, M. (2007). Credit Derivatives and Risk Management. Board of
Governors of the Federal Reserve System Finance and Economics
Discussion Series, paper 2007-47
238 Chapter Thirteen

Kline, M. (1953). Mathematics in western culture. London: Oxford

University Press
Kahneman, D.Tversky, A. (1979). Prospect Theory: An Analysis of
Decision under Risk. Econometrica, Vol. 47(2), 263-292
. (1981). The Framing of Decisions and the Psychology of Choice,
Science, New Series, Vol. 211(4481), 453-458
. (1986). Rational Choice and the Framing of Decisions, The Journal of
Business, 59(4), Part 2: The Behavioral Foundations of Economic
Theory. S251-S278
. (1992). Advances in prospect theory: Cumulative representation of
uncertainty. Journal of Risk and Uncertainty, 5. 297-323
Knight, F. (1921). Risk, Uncertainty and Profit. Reprint. Chicago:
University of Chicago Press
Kowszun, J.-Struijve, O. (2005): Guidance on risk assessment, Education
for Change Ltd
MacKenzie, D. (2005) Mathematizing risk: models, arbitrage and crises.
In: Hutter, B. and Power, M. (eds.). Organizational Encounters with
Risk. London: Cambridge University Press, 167-189
Mandelbrot, B. (1997). Fractals and scaling in finance: discontinuity,
concentration, risk. New York: Springer
. (2006). The Misbehavior of Markets: A Fractal View of Financial
Turbulence. New York: Basic Books
Mandelbrot, B.Taleb, N.N. (2005): How the Finance Gurus Get Risk All
Wrong, Fortune 7/11/2005, 152(1), 99
Odling-Smee, F.J.Laland, K.N.Feldman, M.W. (2003). Niche Construction.
The Neglected Process in Evolution. Princeton: Princeton University
Press
Pieruschla, E. (1963). Principles of Reliability, Prentice-Hall, Englewood
Cliffs
Roberts, N.H. (1964). Mathematical Methods in Reliability Engineering,
New York: McGraw-Hill
Sandler, G.H. (1963). System Reliability Engineering, Englewood Cliffs:
Prentice-Hall
Schnemann, H. (1994). Measurement: The reasonable ineffectiveness of
mathematics in the social sciences. In: Borg, I., Mohler. Trends and
perspective in empirical social sciences. New YorkBerlin: Walter de
Gruyter, 149-160
Sornette, D.Kovalenko, T. (2012), Dynamical Diagnosis and Solutions
for Resilient Natural and Social Systems
arXiv:1211.1949v1 [physics.soc-ph]
 Mathematizing Risk: A Heuristic Point of View 239

Sornette, D.Johansen, A. (2002). Endogenous versus Exogenous Crashes

in Financial Markets, arXiv:cond-mat/0210509
Sornette, D.Filimonov, V. (2012). Quantifying reflexivity in financial
markets: towards a prediction of flash crashes, arXiv:1201.3572v2
Soros, G. (1987). The alchemy of finance. New York: Wiley & Sons
Taleb, N. N. (2010). The Black Swan. Second Edition. New York: Random
House Trade Paperbacks
Taleb, N.Douady, R. (2012). Mathematical Definition, Mapping, and
Detection of (Anti)Fragility. draft
Veneziani, V. (2011): The Greatest Trades of All Time: Top Traders
Making Big Profits from the Crash of 1929 to Today. Hoboken (New
Jersey): Wiley & Sons Inc
Voinea, G.Anton, S.G. (2009). Lessons from the current financial crisis.
A risk management approach. Review of Economic and Business
Studies. 3. 139-147
 CHAPTER FOURTEEN

REFLECTIONS ON THE OBJECTIVITY

OF MATHEMATICS

ROBERT S.D. THOMAS

SUMMARY The faith of a mathematician in mathematics can seek

understanding, an understanding that, unlike experience, might be shared
with non-mathematicians. I understand with similes and assimilations, for
example, assimilating mathematics to art and to sciences as the study of
relations as such. A few approaches to epistemology of mathematics lack
dependence on ontology. Applying the notion of assimilation to a
conceptual approach suggests that objectivity is achieved in mathematics
by public agreement (including agreements to differ) on styles of inference
and definitions in terms of relations.

KEYWORDS assimilation, grounding, mathematics, objectivity, object

1. Introduction
The motto of the greatest theologian of a thousand years ago, Anselm of
Canterbury, was faith seeking understanding (fides quaerens intellectum).
Quite aside from its original theological application, this motto can serve
the philosopher of mathematics that begins from mathematics like the
author, little sense though it would make for a philosopher that did not
come to philosophy from mathematics. For such a one, the faith is a
scientific reliance on mathematics as a body of knowledge and the
understanding a philosophical contextualization.
According to the Stanford Encyclopedia of Philosophy (Thomas
Williams) Anselms motto lends itself to at least two misunderstandings,
both of which apply to its mathematical guise. First, it could express a
wish to replace faith with understanding. When the faith to which one is
referring is the acceptance of the corpus of mathematical knowledge, some
of which one has seen proved convincingly and some small part of which
242 Chapter Fourteen

one has contributed, philosophical understanding is a non-starter as a

replacement. Speaking personally, what I want is an understanding that is
not mere conviction, not justification and certainty. I want mathematics set
in the context of everything else that one knows, some of it on less good
evidence, but much of it on the same, namely, conventional wisdom. As
David Lewis famously opined, on a basis of their track records philosophy
is not going to tell mathematics how to improve doing mathematics. A
fortiori, philosophy will not replace it.
The other misunderstanding of the motto by some philosophers (again
according to Williams) is that, because it begins with the faith it is of
significance only to the faithful, in this case mathematicians. That too
would be wrong. It is important that non-mathematicians have some
understanding of the mathematical aspect of such mathematics as they do
and of what they do not themselves do, just as it is of some value to
everyone to appreciate what they do not themselves do a lot or at all. The
great importance of understanding mathematics corresponds to the
importance of mathematics.
My aim, a mathematicians, of getting an understanding of mathematics
does not compel or even tempt me to produce or even to seek answers to
all of the questions that interest philosophers. Not only are questions of
ethics and aesthetics of only passing interest to me, but also I have never
been concerned with some of the philosophical problems to do with
mathematics that so much concern some philosophers. A topic that I find
of particular interest is the struggle philosophers have with the relation of
equinumerositythe very prototype of the relations I consider
mathematics to be about. Ill return later to my preference for avoiding
ontological issues, being content to bracket them most of the time. There
are likewise questions of interest to some mathematicians that do not
concern me. Seeing that mathematics is built downwards into so-called
foundations as well as upward to higher-level theories, the mathematical
quest for certainty in foundations, however affected by philosophical
predilections, is not one I follow.
Knowing whatever amount of mathematics, it is important to have
some understanding of the activity and results. What is this activity whose
results are so dependable that many of them have been accepted for now
over two thousand years? This question occasions in my case a detour of
investigation on account of there being no commonly accepted answer to
such questions in general. What is something? is not an easy question to
answer even when it is something less mysterious than the stock item of
philosophy of mathematics, the number two.
 Reflections on the Objectivity of Mathematics 243

2. Assimilation
My notion of assimilation is roughly that of Piaget in which new things we
meet are assimilated to notions we already have, which notions in their
turn are accommodated to accept the new arrivals. I see this process as in
contrast to access to pre-ordained universals. I have made two attempts at
publishing my proposal of how this is done (or how I do it), (1991) and
(2010) in the volume (Cellucci et al. 2011). Our ability to characterize
and classify empirical objects, the editors summarize, depends on
processes of assimilation. Experience furnishes initial examples of a kind
of thing (like cats) and we go on to make decisions about what other kinds
of objects to place in the same grouping; because such assimilation is often
based on a shifting, fuzzy, and socially constructed notion of sufficient
likeness, assimilation classes are not determinate enough to be sets. Since
the world is relatively orderly, our individual and cultural assimilation
classes overlap enough that we can communicate... (Cellucci et al. 2011,
p. xxii).
I have quoted a summary by others in order to point out that, in using
standard terms like kind twice, they are going around the point. We make
decisions about the inclusion in our classes of individuals (idiosyncratic
particulars, as Russell called them); we are forming our own idea about
each kind we recognize (radical constructivism). Naturally, growing up in
a language community our assimilation classes correspond usually and
roughly to the vocabulary we learn (social constructivism). As Piaget as
epistemologist pointed out, our assimilation classes are accommodated to
what we assimilate to them. We then without further ado assimilate our
classes to the kinds supposedly determined by our culture. Aside from our
effective guessing, our culture transmits its supposed knowledge of kinds
to us by correcting our classification of individuals and statements about
categories like Not all ferries carry automobiles. We do not recollect
their platonic forms from before our incarnationor so I suppose. So what
we each regard as the kinds of our culture are our assimilations of many
persons assimilation classes.1 If we know more than one language, our
approach to this process is made more sophisticated. If we study
languages, we see this happening on a large scale. I took George Lakoff's

1
Word types, corresponding to assimilation classes of word tokens, are among the
most successful examples with some English exceptions like cleave and enjoin.
244 Chapter Fourteen

(1987) as vindicating this point of view, which I had adopted even before
seeing (Lakoff and Johnson 1980).2
How does assimilation work beyond the need to do it? We use it in a
principled way by ignoring most differences among things that we
assimilate, which differences are noticed being determined by our present
circumstances and interests. We sort sensory-input bundles as somethings,
as philosophers as varied as Husserl and Wittgenstein have observed. In a
zoo cats are a wider class than in a pet shop. Inside an assimilation class
we use a principle of assimilation in ignoring differences that we want to
ignore while being able to take account of any differences that matter to us
and expecting things to be the same in ways that make sensealways
conscious of the possibility of exceptional ways in which things, in spite
of being assimilated, differ.
On this basis, my idea of understanding something is to know what
assimilations are appropriate to iteach assimilation having positive
features recommending it and negative features that need to be set aside
and what mere similarities are helpful. My understanding of mathematics
in such terms is of course based upon my experience of doing mathematics
since I began doing arithmetic in grade two and doing research as a
graduate student. That experience is an understanding of a sort, but my
concern here is shareable verbally expressed understanding. I assimilate
mathematics to science (2008) and art (disclaimers in (2008; 2010)) and
have put forward for consideration similarities of research to the strategic
thinking connected with certain games (2009; 2011a) and of written
mathematics to stories (2002b; 2007). My only paper in a philosophy
journal was intended to be a sustained argument for my view that
mathematics is about relations and not about mathematical objects, that the
mathematical objects are just things that we wish on the relations we want
to talk about in order to be able to do the talking. It is very hard to talk
about relations without relata, and I regard it as a triumph of the
mathematical method to arrange to talk about relations with artificial
relata. I attempted to show this in (2000; 2002a) by showing the
difficulties that a number of philosophers got into by their assumption that
mathematics is about mathematical objects. The attempt took the form of a
consideration of philosophical identifications of mathematics with fiction
(an assimilation that I regard as a mistake) and analogies drawn between

2
I shall for ever be grateful to Barron Brainerd, who pointed me to that book in
1982 when he read what I had written.
 Reflections on the Objectivity of Mathematics 245

mathematics and fiction (a package of similarities that too much can be

made of).
The assimilation to art appears only in disclaimers because, while it is
important to me, I have nothing to say about it except to emphasize its
importance. Mathematicians motivations are often aesthetic; the mathematics
that I have done recently is an example in that it has studied geometric
patterns of fabric weaving or floor tiling that I find attractive (2011b). The
appeal is in the material not the mathematics. The more usual aesthetic
motivation for mathematical research is the attraction of what one can say
about invisible and abstract material. No one alive has produced a
mathematical result as beautiful as ei + 1 = 0, but mathematics is not just
results. Proofs too can be attractiveand theories. The notorious
subjectivity of beauty does not remove it as motivation from us subjects.
I should mention that my main concern both as an editor and as an
occasional author is that philosophy and teaching of mathematics should
be about mathematics as I have experienced or observed it and not about
something that it would be more convenient to talk about or teach. It is
obviously a temptation analogous to purity of mathematics to discuss
something that is invented for convenience rather than recalcitrant
actualities. In this sense it is important that philosophy of mathematics
should be analogous to applied mathematics rather than to pure
mathematics. This of course restricts what one can say. Another temptation
that I have attempted to resist is that of educators at the primary and
secondary levels to substitute something easier than arithmetic, geometry,
and trigonometry for those topics and call it mathematics. In my (1996),
followed up in my part of (1997), I attempted to express my view of
mathematics in a brief compass mainly to contrast it with proto-
mathematics, that is, the places and practices where one can see
mathematical ideas or mathematical possibilities, ethnomathematics for
example. As I said there, building on what children know or can easily
pick up is a good way to begin but the wrong place to stop. Having spelled
out the difference, I then pointed out that situations like word problems,
into which one can read mathematics, have to have their mathematical
relations extracted before the mathematics necessary to solve the problem
can be done with the piece of school mathematics that gave rise to the
problem. Weaving, the topic of my mathematical research, has both a
substantial ethnomathematical literature as well as mathematical literature,
both distinct from the literature of weavers.
246 Chapter Fourteen

3. Mathematics assimilated to sciences

With the notion of assimilation in place it is possible to state the position
of my (2008) more adequately. The position is not that mathematics is
something else (everything being what it is and not another thing) but that
it can be useful to assimilate mathematics to the other sciences
(themselves merely an assimilation class) because certain of the sciences
analogous features have been modelled on the corresponding features of
mathematics. Galileo in particular launched the revolution in science by
using Husserls useful word againbracketing the natures of falling
bodies, for example, and studying the relations between distances moved
and elapsed times in what we now see as the appropriately quantitative
way.3 This bracketing of natures and focus on relations mimics that in
arithmetic or geometry where it does not matter whether you are counting
fruit or galaxies or what Euclids points are if they can be collinear. The
assimilation is not a metaphysical claim about the nature of mathematics
but a epistemological claim that it can be instructive to think relations are
the subject matter of mathematicswhatever it is in itself if anything. I
dont know, of course, about the itself, but it certaily includes activities,
which the current enthusiasm for practice emphasizes, and results,
physically represented in texts. The assimilation suggests exploring the
ways in which mathematics is like a science, even the assumption that,
except where counterindicated, it may work the same way as a science in
accordance with the principle of assimilation. The obvious counterindication
is that the things whose relations are the subject matter of mathematics are
not usually or necessarily physical things in order to enter into the
relations of interest. In order to enter into chemical relations, for example,
things must be real substances (giving rise to the bumper sticker, chemists
are persons of real substance). Assimilation of mathematics to science
does not solve or even address philosophical problems of long standing; it
merely suggests a way in which discussions in philosophy of science can
be seen as relevant to mathematics. Axiom choice, for example, can be
seen as problem solving with an aesthetic component rather than as
something leading to metaphysically certain foundations, which must shun
plurality because it is more important to get it right than to get it fruitful.4

3
Cf. (Cellucci 2013, 16).
4
A philosopher coming around to this conclusion --- in her own way --- is
Penelope Maddy in her recent book (2011). Cf. (Cellucci 2013).
 Reflections on the Objectivity of Mathematics 247

Let me try to spell out how objects play different roles in mathematics
and the other sciences. In a science we are interested in the behaviour of
things, say in physics physical thingsassimilations of their time-slices,
which is typically how they are experienced. What we sense and measure
are relations among such. We abstract things like, at an extreme, point
massesnot themselves assimilations, but we assimilate them to the
things from which they have been abstracted. The models, as they are
often called, are also things, but not real ones, about which we weave our
theories. As Mary Leng points out in her (2010), the imaginary nature of
these entities does not keep science from studying them nor philosophers
from considering them as if (nod to Vaihinger) they were real in spite of
the undoubted fact that they do not exist. We apply mathematics to them
and have theories. The contrast that I think is useful is that in mathematics
we are not interested in the things at the farthest reaches of our study but
in the relations among them. We invent the things to bear the relations---
and to do nothing else, to have no other relationsand then theorize about
them because we need things to think about. They are not assimilations of
time-slices. What Mark McEvoy writes in his (2012) is, I think, typical.
My own view (McEvoy 2004), heavily indebted both to Katz (1998)
and Resnik (1997), is that our basic mathematical concepts arise from
causal interaction with physical objects that approximate mathematical
objects (e.g., approximately square objects, or n-membered sets of physical
objects). The elementary concepts so obtained are then available for
examination by reason which can establish some elementary truths
involving those concepts (e.g., elementary arithmetical and geometrical
truths). The development of proof and of axiomatization further extends
our ability to reason about these concepts. Some time after we have begun
to establish mathematical truths, we notice that propositions involving
mathematical concepts are not precisely true of anything in the empirical
world. (McEvoy 2012)
His next step is to conclude that if mathematical truths are to be true
at all, they must be true of something else. One is not bound to do this
unless one is bound by philosophers narrow notion of true. The
existence of those things is of no consequence despite being argued about
ceaselessly. In both cases, as Leng argues, we theorize about invented
things but with different matters in mind to understand, in mathematics the
relations among the invented thingswhatever they areand in science
the real things that the inventions are abstracted from and their relations.
The importance of the whatever they are is that application is easy and
unproblematic in the simplest cases; their nature is not bracketed for there
is no nature to bracket. I sometimes put this by saying that, within the
248 Chapter Fourteen

mathematics, mathematical objects names are not nouns but antecedentless

pronouns.

4. Objectivity
Following up the impatience that mathematicians (and some philosophers)
feel at the unproductive and unending focus on ontological questions about
things (and in my case also the ignoring of relations), I suggested in my
(2004) that it would be sensible to try to work out epistemology that was
not based on being right on the ontological coin flip. Since we must be
able to reason as dependably about what does not existeven in a
mathematical senseas about what does, for instance in reductio proofs,
whether some things exist or not is not of any practical importance. It is
not just in mathematics that we need to be able to reason effectively about
what does not exist; it seems to me that the evolutionary advantage to our
reasoning ability is primarily our capacity for reasoning about the future.
No one even argues that the existence question is important; one
encounters sheer dogmatism. Its sole importance is that philosophers base
other arguments on having the right answer to it. I am aware of three
approaches to mathematics that do as I suggest. About category theory,
which underlies the work of several philosophers surveyed in (Landry and
Marquis 2005), I have nothing to say. There are modal approaches
(Chihara 1990; Hellman 1989; Tharp 1989; 1991), about which I have
recently written in general terms (2011a). And there is conceptual realism
attributed to Gdel (for example by Wang (1996, 8.5.20)), defended by
Isaacson (1994), and elucidated and criticized by Detlefsen (2011). It
seems to me that these two philosophical approaches address sufficiently
different aspects of mathematicsmodality and conceptsthat they are
not necessarily in conflict with one another. Both work without regard to
what their advocates Hellman and Isaacson call object-platonism.
It seems entirely uncontroversial that mathematics is objective and, on
that account, intersubjective. When the question is examined as in
(Shapiro 2007), no doubt is left, but the important point is made that
objectivity is a matter of more/less not either/or. And Georg Kreisel claims
that the objectivity of certain notions can be considered without having
to answer whether in addition, some reality or realizability external to
 Reflections on the Objectivity of Mathematics 249

ourselves is involved (1970, p. 20). This is the standard mathematical

stance: not being concerned with the existence question.5
Consider what the availability of the notion of assimilationwith its
pervasive use outside mathematics and near-inapplicability within written
mathematics (saying nothing about thinking)can do for a discussion of a
position like concept-platonism, the alternative to object-platonism
presented by Isaacson. His reason for wanting to avoid object-platonism is
worth mentioning because it is not arguable metaphysical difficulties with
the supposed platonic objects but the reason with which I sympathize
that they are not what mathematics is about. Two further voices in this
direction have recently come to my attention.

Le mathmaticien fait compltement abstraction de la nature des objets et

de la signification de leurs relations: il na qu numrer les relations et
les comparer. (Gauss 1832, p. 176)

In all Mathematical Science we consider and compare relations. In algebra

the relations which we first consider and compare, are relations between
successive states of some changing thing or thought. And numbers are the
names or nouns of algebra, marks or signs, by which one of these
successive states may be remembered and distinguished from another [...].
6
(Hamilton 1831, fol. 49)

Isaacson's interest is to ground objectivity in concepts rather than in

objects. To avoid objects as a ground is an aim also of Azzouni (2010),
Ernest (1988), Leng (2010), Priest (2007), Weir (2011), and Yablo (2005),
and their being the sole ground is opposed by Balaguer (1998) and Maddy
(2011) with their nothing-to-choose attitude between realisms and
irrealism. So this is worth trying. But it is not common to put concepts in
their place. I suspect that a reason for that is that concepts are slippery. As
a sentence of the right sort expresses a proposition, which is non-physical,
a definition expresses a conceptor tries to. Since everyday concepts like
chair are different in different persons, what we call the concept of chair is

5
John Searle (1995, p. 8) makes the distinction between the epistemic sense of the
objective-subjective distinction with which I am concerned here and an
ontological sense mathematicians are typically less interested in. I owe this
reference to Julian Cole.
6
Both of these quotations appear in (Flament 2011, p. 53), which is why Gausss
Latin has been translated into French. Perhaps eventually a critical mass of such
statements by mathematicians will be reached, and philosophers will realize that
concentration on objects really does misrepresent the subject.
250 Chapter Fourteen

an assimilation class of those personal concepts, themselves reifications of

assimilation classes. Since there is no way even to compare them, there
can be no way of certifying that they are identical; so they are assimilated
in spite of being different in unknown ways. There are very few purposes
for which this is inadequate despite there being no agreed-upon concept of
chairno necessary and sufficient condition for being a chair. Reading an
accepted definition, say in a dictionary, one sees reflected in it what can be
interpreted as ones own personal concept. This is presumably the way
explicit mathematical definitions start out. For instance, continuity. There
must have been varied notions of pencil-and-paper continuity, because the
notions were trying to capture the nature of curves that were continuous,
regarded as a non-natural kind like chair. This is something discussed by
Lakoff and Nez (2000). It is hard to ground the objectivity of
mathematics in concepts like chair. But a properly mathematical
definition is not like the dictionary definition of chairpace Lakoff and
Nez. It is not an attempt, in each time period, to capture the vague and
personal notions of mathematicians working at that time as a dictionary
definition attempts to capture the gradually shifting assimilation classes of
its time. Reuben Hersh claims (2011, p. 316) that mathematical concepts
(mental models) are subject to social control; I claim that the chief control
device is definitions.
I attempt a description, not justification, of what has happened in the
example of pencil-and-paper continuity. First, a curve is now defined as
the co-ordinate functions representing it. The matter of continuity is
thereby passed back to the co-ordinate functions. Then a function f is
defined to be continuous in its argument at x if (for any > 0)(there exists
> 0)(0 < |y x| < |f(y) f(x)| < ). This is not a concept formed by
assimilating what is in the heads of mathematicians; it is an entirely
objective matter whether a function is continuous. The definition took
some time to establish. No nature of continuity is argued for. Such a
definition is expressed entirely in terms of the satisfaction of certain
relations, and it is classically an objective matter whetherby the law of
excluded middleeach object in the domain of discourse satisfies each
such pertinent relation. The basis in definitions depends on this kind of
definition not just on there being written definitions.7 No doubt there is

7
The concept of a democracy is not particularly objective. A written constitution,
to use an example from (Rosen 1994, pp. 300 f.), imparts much more objectivity
especially in the hands of expert interpreters. But its basis in natural-language
definitions makes its objectivity inferior to that of contemporary mathematics.
252 Chapter Fourteen

mathematizing involved in shifting the definition of continuous from the

graph of a curve to the co-ordinate functions. The grounding in reality is in
relations,9 and the grounding in knowledge is in definitions. How is this
duality possible?
It has been suggested to me that I am forced to use non-standard
semantics. That might be the case if the standard is set too low, but it
seems to me that the semantics I need ought to be standard for it is old and
customary. Steve Yablo used to call mathematical language figurative,
which I think brings with it suggestions (like Lakoff and Johnsons use of
metaphor) that we do not want or needbesides being vaguer than I
want. The talk of objects is definitely literal; what we have is not the
standard figurative situation where the literal makes no or little sense and
so one resorts to a figurative interpretation in order to make good sense.
The word that seems to be needed is parabolic. For in parables one has
language that makes perfect literal sense, which can be thought to be what
is being said, but also a larger or deeper interpretation is simultaneously
available. Oliver Twist can be thought of as being about an English orphan
or about child welfare. One does not need to deny anything about the child
to see the further intention of Dickens in the story; it is about child welfare
by being about Oliver. Ontology is irrelevant. The further meaning need
not deny the meaning under the standard semantics, and mathematical
meaning about relations need not deny the meaning about the objects. But
the further meaning is what can be applied to the physicalespecially the
idealized physicalby an interpretation that is easier and much less
arbitrary than parabolic interpretation or whatever one ought to call the
elevation of subject matter of Oliver Twist from the fictional character to
Victorian society, as I wrote in the final section of (2007).
In (2002b) I claimed that proofs and algorithms are the characteristic
literary genres (pp. 50 f.) of mathematics. Another is properly mathematical
definitions both explicit and implicit, assimilation classes of physical
representations of what the definitions express. One can ask whether these
definitions are intended to existand what that might mean. That depends
on what you mean by exist. It seems to me that definitions stand a better
chance than concepts, but I do not see either question as important.
Concepts and definitions matter, which is more important. Since I am not a
nominalist I see no reason to deny existence of some sort to definitions or
to concepts provided they are not promoted beyond what seems to be their

9
Cf. (Rosen 2011) for a notion of grounding. Lakoff and Nez (2000) also write
of grounding, constructively criticized in my (2002c).
 Reflections on the Objectivity of Mathematics 253

natural status as reified assimilation classes. On account of the fact that a

mathematical definition is an assimilation class of expressions of what we
regard as the same meaning, the objectivity that we achieve is not
perfect. It is on account of the presence of humans in the process that
assimilation is needed and objectivity suffers in consequence and in
principle. In practice there is little loss in these assimilations; if it turns out
that there is any divergence in interpretation, then we eliminate it.10 The
only way, presumably, to have perfect objectivity would be to have
definitions equivalent under Turing-machine processes so that objectivity
would be attained at the price of meaninglessness, perfect formalism.11
An interesting case can conclude. Solomon Feferman has been saying
for some years that the so-called continuum hypothesis is not a definite
mathematical problem. One of his cases for this can be expressedin my
termsas being that it arises from the assimilation of six distinguishable
continua (Feferman 2009): the line in Euclidean the plane, the continuum
of real numbers discussed by Cantor in 1872 and 1883, the continuum of
Dedekind cuts, the explicitly continuous line of the later editions of
Hilberts Grundlagen der Geometrie, the set of all paths in the full binary
treeall functions from the natural numbers to {0,1}, and the set of all
subsets of the natural numbers. His conceptual structuralism, set out in
ten theses (Feferman 2009) requires as thesis five completely clear
conceptions, the clear and distinct ideas of Descartes again. This
assimilation is not a completely clear conception (Feferman 2011).

I am grateful for assistance with this project from Julian Cole, Sol
Feferman, and Hugh Thomas.

References
Azzouni J. (2010). Talking about Nothing: Numbers, Hallucinations, and
Fiction. Oxford University Press.
Balaguer, M. (1998). Platonism and Anti-Platonism in Mathematics.
Oxford University Press.

10
Discrepancies are recognized and worked out, either by correcting errors,
reconciling differences, or splitting apart into different, independent pathways.
(Hersh 2011, p. 316)
11
This appears to be the conclusion of Stewart Shapiro in his lengthy discussion
(2007) of mathematical objectivity.
254 Chapter Fourteen

Cellucci C. (2013). Philosophy of Mathematics: Making a Fresh Start,

Studies in History and Philosophy of Science, 44: 32-42
Cellucci C., Grosholz E.R. and Ippoliti E. eds. (2011). Logic and
Knowledge. Newcastle upon Tyne: Cambridge Scholars Publishing.
Chihara C. (1990). Constructibility and mathematical existence. Oxford:
Clarendon Press.
de Groot J. (2009). The Historical Novel. Routledge.
Detlefsen M. (2011). Discovery, invention, and realism: Gdel and others
on the reality of concepts. In: (Polkinghorne 2011): 73-94.
Devlin K. (2008). A mathematician reflects on the useful and reliable
illusion of reality in mathematics. Erkenntnis, 68: 359-379.
Ernest P. (1998). Social constructivism as a philosophy of mathematics.
Albany: State University of New York Press.
Feferman S. (2009). Conceptions of the continuum. Intellectica, 51: 169-
189.
. (2011). Is the continuum hypothesis a definite mathematical problem?.
MS for the Exploring the Frontiers of Incompleteness project, Harvard
University, 2011-2012.
http://math.stanford.edu/~feferman/papers/IsCHdefinite.pdf
Flament D. (2011). L'algbre comme science chez W.R. Hamilton: Le
recours au temps pur. In Flament D. and Nabonnand P., eds. Justifier
en mathmatique. Paris: Maison des sciences de l'homme: 49-102.
Gauss C.F. (1832). Theoria residuorum biquadaticorum, commentation
secunda. Societi Regiae scientiarum Gottingensis recentiones, 7.
Werke, vol. 2, pp. 169-178.
Hamilton W.R. (1831). `Metaphysical remarks on algebra', February 1831,
in {\em TCD Notebook} No. 24.5.
Hellman G. (1989). Mathematics without Numbers. Oxford University
Press.
Hersh R. (2011). Mathematical intuition: Poincar, Polya, Dewey. In:
(Cellucci et al. 2011): 297-323.
Isaacson D. (1994). Mathematical intuition and objectivity. In: George A.,
ed. Mathematics and mind. Oxford University Press: 118-140.
Katz J. (1998). Realistic Rationalism. Cambridge, Mass.: Bradford MIT.
Kreisel G. (1970). The formalist-positivist doctrine of mathematical
precision in the light of experience. L'age de la Science, 3: 17-46.
Lakoff G. (1987). Women, fire, and dangerous things. Chicago University
Press.
Lakoff G. and Johnson M. (1980). Metaphors we live by. Chicago
University Press.
 Reflections on the Objectivity of Mathematics 255

Lakoff G. and Nez R.E. (2000). Where mathematics comes from. New
York: Basic Books.
Landry E. and Marquis J.-P. (2005). Categories in context: Historical,
foundational, and philosophical. Philosophia Mathematica (3), 13:1-43.
Leng M. (2010). Mathematics and Reality. Oxford University Press.
Maddy P. (2011). Defending the Axioms: On the Philosophical
Foundations of Set Theory. Oxford University Press.
McEvoy M. (2004). Is reliabilism compatible with mathematical
knowledge? Philosophical Forum, 35: 423-439.
. (2012). Platonism and the `Epistemic Role Puzzle '. Philosophia
Mathematica (3), 20: 298-304.
Polkinghorne J., ed. (2011). Meaning in Mathematics. Oxford University
Press.
Priest G. (2007). Towards Non-Being: The Logic and Metaphysics of
Intentionality. Oxford University Press.
Resnik M. (1997). Mathematics as a Science of Patterns. Oxford
University Press.
Rosen G. (1994). Objectivity and modern idealism: What is the question?
In: Michael M. and O'Leary-Hawthorne J., eds. Philosophy in mind:
The place of philosophy in the study of mind. Dordrecht: Kluwer: 277-
319.
. (2011). The reality of mathematical objects. In: (Polkinghorne 2011):
113-131.
Searle J. (1995). The construction of social reality. New York: The Free
Press.
Shapiro S. (2007). The objectivity of mathematics. Synthese, 156: 337-
381.
Tharp L. (1989). Myth and Mathematics: A Conceptualistic Philosophy of
Mathematics I. Synthese, 81: 167-201.
Tharp L. (1991). Myth & Math, Part II (Preliminary Draft). Synthese, 88:
179-199.
Thomas R.S.D. (1991). Meanings in ordinary language and in
mathematics. Philosophia Mathematica (2), 6: 3-38.
. (1996). Proto-Mathematics and/or Real Mathematics. For the Learning
of Mathematics, 16, No. 2: 11-18.
. (2000). Mathematics and Fiction I: Identification. Logique et Analyse,
43: 301-340.
. (2002a). Mathematics and Fiction II: Analogy. Logique et Analyse, 45:
185-228.
. (2002b). Mathematics and Narrative. The Mathematical Intelligencer,
24, No. 3: 43-46.
256 Chapter Fourteen

. (2002c). Idea Analysis of Algebraic Groups. Philosophical

Psychology, 15: 185-195.
. (2004). Mathematicians and mathematical objects. In: Link G., ed. One
Hundred Years of Russell's Paradox. Papers from the 2001 Munich
Russell Conference. Berlin: de Gruyter: 577-590.
. (2007). The comparison of mathematics with narrative. In: Van
Kerkhove B. and Van Bendegem J.P., eds. Perspectives on
mathematical practices: Bringing together philosophy of mathematics,
sociology of mathematics, and mathematics education. Dordrecht:
Springer: 43-59.
. (2008). Extreme Science: Mathematics as the Science of Relations as
such. In: Gold B. and Simons R., eds. Proof and other dilemmas:
Mathematics and philosophy. Washington, D.C.: Mathematical
Association of America: 245-263.
. (2009). Mathematics is not a game but... . The Mathematical
Intelligencer, 31, No. 1: 4-8. Reprinted in: Pitici M. ed. The best
writing on mathematics 2010. Princeton: Princeton University Press:
79-88. Online on the journal site.
. (2010). Assimilation: Not only indiscernibles are identified. In:
(Cellucci et al. 2011): 363--379.
. (2011a). Acts of geometrical construction in the Spherics of
Theodosios. In: Sidoli N. and Van Brummelen G., eds. From
Alexandria, through Baghdad: Surveys and Studies in the Ancient
Greek and Medieval Islamic Mathematical Sciences in Honor of J.L.
Berggren. Springer: 227-237.
. (2011b). Perfect colourings of isonemal fabrics by thick striping. Bull.
Australian Math. Soc., 85: 325-349. Online on the journal site and
http://arxiv.org/abs/1109.2254 [math.CO].
Thomas R.S.D. and Gerofsky S. (1997). An Exchange about Word
Problems. For the learning of mathematics, 17, No. 2: 21-23.
Wang Hao (1996). A Logical Journey: From Gdel to Philosophy.
Cambridge, Massachusetts: MIT Press.
Weir A. (2011). Truth Through Proof: A Formalist Foundation for
Mathematics. Oxford University Press.
Yablo S. (2005): The myth of seven. In: Kalderon M.E., ed. Fictionalism
in metaphysics. Oxford University Press: 88-115.
 CONTRIBUTORS

Vito Michele Abrusci is full professor in logic, at the University Roma

Tre in Rome. His research fileds are proof theory, linear logic and
developments of linear logic, history of logic and foundations of
mathematics. He is the author of Autofondazione della matematica. Le
Ricverche di Hilbert sui fondamenti della matematica, in David Hilbert.
Ricerche sui fodamenti della matematica. Napoli: Bibliopolis: 13-131;
Dilators, generalized Goodstein sequences, independence results: a survey.
In Contempiorary Mathematics, 65: 1-23; (con Paul Ruet) Non
Commutative Logic, I : the multiplicative fragment. In Annals of Pure and
Applied Logic, 2001: 29-64.

Claudio Bernardi has been Professor of Mathematics at Sapienza,

University of Rome, since 1987. His main interests include mathematical
education, algebraic logic, and the foundations of mathematics. He is the
Editor of the journal Archimede, member of the Scientific Commission of
the Unione Matematica Italiana. He is the author of Fixed Points and
Unfounded Chains, Annals of Pure and Applied Logic, (2001), 109/3:
163178; (2009). A topological approach to Yablos paradox, Notre Dame
Journal of Formal Logic, 50, Issue 3: 331338 (2009); Discussion on
Mathematical Intuition. In Cellucci C., Grosholz E., Ippoliti E., eds., Logic
and Knowledge, Cambridge Scholars Publishing, UK: 324-328 (2011).

Daniel G. Campos is Associate Professor of Philosophy at Brooklyn

College of The City University of New York. His articles have appeared in
Perspectives on Science, Synthese, Transactions of the Charles S. Peirce
Society, and Studies in Philosophy and Education among other journals.

Cesare Cozzo is Associate Professor of Logic at La Sapienza University of

Rome. His research has been mainly focused on the relations between the
philosophy of logic and mathematics and the theory of meaning. He wrote
several articles on analytic philosophy, the paradox of knowability, the
epistemic conception of truth and holism, which appeared in Erkenntnis,
Rivista di filosofia, Theoria, Topoi and other journals or collections of
essays. He also wrote three books: Teoria del significato e filosofia della
logica [Theory of Meaning and Philosophy of Logic], CLUEB, Bologna
258 Contributors

1994; Meaning and Argument, Almqvist & Wiksell, Stockholm, 1994;

Introduzione a Dummett [Introduction to Dummett] Laterza, Roma 2008.

Maria Rosaria Egidi is Emeritus Professor of Theoretical Philosophy at

Universit Roma Tre. Her main interests are logic, philosophy of language
and analytic philosophy. She is the author of Wittgenstein, Bergmann, and
Hochberg on Intentionality, in: Studies in the Philosophy of Herbert
Hochberg, ed. by E. Tegtmeier. Ontos Verlag, Frankfurt-Paris-Lancaster-
New Brunswick, 2012: 73-86; Meaning and Actions in Wittgenstein's Late
Perspective, in: Criss-Crossing a Philosophical Landscape: Essay on
Wittgenstein Themes Dedicated to Brian McGuinness, Rodopi,
Amsterdam, 1992, pp. 161-179; G.H. von Wright on Self-Determination
and Free Agency, in Philosophical Essays in Memoriam G.H. von Wright,
ed. by I. Niiniluoto and R. Vilkko, in Acta Philosophica Fennica, 77,
Helsinki 2005, pp. 105-14.

Maria Carla Galavotti is professor of Philosophy of science at the

University of Bologna and Research Associate of the Centre for
Philosophy of the Natural and Social Sciences of LSE. She is also life
member of the Center for the Philosophy of Science of the University of
Pittsburgh and of Clare Hall College, Cambridge. Her research focuses on
the foundations of probability, the nature of scientific explanation,
prediction, causality, and the role and structure of models in the natural
and social sciences. Among her publications the book Philosophical
Introduction to Probability (Stanford: CSLI, 2005).

Donald Gillies is emeritus professor of Philosophy of Science and

Mathematics at University College London. His research has been in the
general area of philosophy of science with a main specialization in logic,
foundations of probability and philosophy of mathematics. Since 2000, he
has been working on a new research programme concerned with the
application of philosophy of science to medicine. His publications include
the book: Artificial Intelligence and Scientific Method, Oxford University
Press, 1996.

Emily Grosholz is professor of Philosophy at the Pennsylvania State

University and a member of SPHERE / CNRS / University of Paris Denis
Diderot Paris 7. She is the author of Representation and Productive
Ambiguity in Mathematics and the Sciences, Oxford University Press,
Oxford 2007; Cartesian Method and the Problem of Reduction, Oxford
University Press, Oxford 1991.
 From A Heuristic Point of View 259

Reuben Hersh is emeritus professor of Mathematics at University of New

Mexico. He is co-author, with Vera Jon-Steiner, of Loving and Hating
Mathematics. Challenging the Myths of Mathematical Life, Princeton
University Press, Princeton 2011; author of What Is Mathematics, Really?,
Oxford University Press, Oxford 1999; editor of 18 Unconventional
Essays on the Nature of Mathematics, Springer, New York 2006; co-
author, with Phil Davis, of The Mathematical Experience, Birkhuser,
Boston 1981, and Descartes Dream: The World According to Mathematics,
Harcourt Brace Jovanovich, San Diego 1986; and author and co-author of
papers on various branches of mathematics, and on mathematical
pedagogy, and on philosophy of mathematics.

Emiliano Ippoliti is Assistant Professor at Sapienza University of Rome.

His main interests and pubblications are in heuristic, informal logic, logic
of discovery, problem-solving, and economics (finance). Among his
publications the papers Between data and hypotheses (CSP, 2011) and
Generation of Hypotheses by Ampliation of Data (Spinger 2013).

Gabriele Lolli is Professor of Philosophy of mathematics at Scuola

Normale di Pisa since 2008. He specialized in mathematical logic at Yale
University under the supervision of prof. Abraham Robinson. His research
interests axiomatic set theory, Artificial Intelligence, Cognitive Science
and, and the history and philosophy of mathematics and logic. Among his
publications the book Guida alla teoria degli insiemi (Springer 2008).

Lorenzo Magnani is Professor of Philosophy of Science at the Department

of Humanities, University of Pavia, and runs the Computational
Philosophy Laboratory. He has been visiting professor at the Sun Yat-Sen
University in China (2006-2012) and visiting researcher at Carnegie
Mellon University (1992) McGill University (1992-93), University of
Waterloo, (1993) and at the Georgia Institute of Technology (1995 and
1998-99) and Weissman Distinguished Visiting Professor at Baruch
College, City University of New York (2003) His main interests are
philosophy of science, logic (abduction), cognitive science, artificial
intelligence and philosophy of medicine. He is the author of Abduction,
Reason, and Science. Processes of Discovery and Explanation (Kluwer
Academic/Plenum Publishers, New York, 2001) and Abductive Cognition.
The Epistemological and Eco-Cognitive Dimensions of Hypothetical
Reasoning (Springer Science+Business Media, Heidelberg/Berlin, 2009).
260 Contributors

Dag Prawitz is emeritus of Philosophy at Stockholm University. He has

worked mostly in logic and in philosophy of mathematics and language.
He is the author of the book Natural Deduction: A Proof-Theoretical
Study, Almqvist & Wiksell, Stockholm 1964 (republished by Dover
Publications, Mineola 2005). Among his papers one may mention
"Hauptsatz for higher order logic" in Journal of Symbolic Logic (1969)
and "Ideas and results in proof theory" in Proceedings of the 2.
Scandinavian Logic Symposium (1971). Two of his recent papers are
Logical Consequence from a Constructivist Point of View in The Oxford
Handbook of Philosophy of Mathematics and Logic (2005) and "Inference
and Knowledge" in The Logica Yearbook 2008.

Robert S.D. Thomas is professor of Mathematics at the University of

Manitoba in Winnipeg, Canada. His interests and work have included pure
mathematical research on braids and weaving, applied mathematical
research on elastic waves in solids, historical work issuing in his only
book (with J.L. Berggren), and philosophical work mostly editing the
journal Philosophia Mathematica. The book is Euclid's Phaenomena: A
Translation and Study of a Hellenistic Treatise in Spherical Astronomy,
Garland, New York 1996 (reprinted by the American Mathematical
Society and the London Mathematical Society, 2006).
 INDEX
abduction, 76, 149, 179, 180, 181,
184, 185, 186, 190, 191, 192,
193
abductive diagnoses, 188
abstraction, 149, 199, 203, 249, 251
accidents, 214
ad infinitum, 75, 169
additional serendipity, 37
aesthetic, 245, 246
agentbased framework, 184
agreement, 4, 7, 20, 74, 81, 85, 120,
140, 170, 200, 201, 202, 203,
241
Algebraic manipulation, 50
ALGOL, 31
algorithms, 131, 151, 183, 252
Aliseda, 180, 181, 182, 183, 189,
191
analogy, 13, 43, 121, 137, 139, 166,
190, 191, 215, 225, 235
analysis, 3, 5, 7, 14, 15, 18, 19, 30,
42, 45, 47, 81, 93, 96, 98, 102,
110, 111, 112, 141, 145, 147,
148, 149, 150, 151, 152, 153,
154, 155, 156, 158, 159, 160,
165, 175, 177, 182, 189, 193,
201, 204, 206, 207, 221, 228,
234, 235, 236
analytic, 3, 27, 28, 73, 74, 75, 77,
78, 80, 87, 89, 128, 130, 142,
148, 152, 158, 159, 163, 164,
165, 166, 175, 195, 196, 197,
257
analytic method, 73, 74, 75, 77, 78,
87, 89, 128, 142, 163, 164, 165,
166, 175, 195, 196, 197
analytic proofs, 75, 79, 158
analytical method, 46, 77, 168
Anselm of Canterbury, 241
anti-realist, 149
Archimedes, 96, 100
Aristotle, 89, 108, 112, 118, 120,
123, 124, 133, 175
Ars Conjectandi, 167, 177
artificial, 32, 50, 51, 176, 179, 180,
181, 182, 185, 187, 188, 190,
191, 196, 244
assertoric practice, 118, 119
assimilation, 185, 186, 241, 243,
244, 245, 246, 249, 250, 251,
252, 253
Atiyah, 97
automated deduction, 77
automated proof checking, 52
automated reasoning, 94
available evidence, 208
axiomatic method, 45, 73, 74, 77,
79, 89, 90, 134, 140, 164, 170,
176, 196, 200, 215
axiomatic proof, 128, 133, 136
axiomatic proofs, 75
axiomatic system, 41, 45, 46, 48,
164, 169
axiomatization, 170, 214, 247
axioms, 25, 26, 29, 34, 42, 44, 45,
46, 47, 48, 50, 64, 74, 75, 87, 90,
92, 127, 130, 131, 133, 134, 136,
138, 159, 163, 165, 166, 168,
169, 171, 176, 196, 197, 215,
251
Azzouni, 23
Babylonians, 88, 89
BaR, 219, 220
basic mathematical concepts, 92,
247
Bass, 216, 234, 235
Bayes, 204, 205
Bayesian, 204, 205
262
Bayesian approach, 204
begging the question, 82
Begriffsschrift, 30, 34, 38
Bewutseinzustnde, 13
biological evolution, 78, 180, 188,
197
bottom-up, 195, 196, 198, 201, 204,
207, 209, 213, 214, 215, 234,
235, 236
bottom-up approach, 195, 196, 198,
201, 207, 209, 234, 235
Bouleau, 218, 219, 220, 221, 223,
231, 237
Bourbaki, 29, 38, 97, 171, 196
Cantor, 102, 144, 253
Cauchy, 95, 97, 218
causal relationships, 205
Cellucci, 91
certainty of mathematics, 41, 42
ceteris paribus, 221
chimera, 107, 108, 109, 110, 117,
200
chimeras, 108, 200
choice of the variables, 231, 236
Chrysippus, 80, 82
closed world view, 164, 165, 171,
190
coherence, 109, 111, 132
cohomology theory, 152, 155, 157
Columbus, 24, 33
Comedy, 103
complete knowledge of the
universe, 200
Completeness, 55, 63, 65, 66, 67,
68, 69, 70
completeness of knowledge, 200
complex spatial networks, 199
complexity of data, 199
computer science, 23, 24, 25, 30,
31, 32, 33, 41, 51, 52
computer-assisted proofs, 51
concavity hypothesis, 221
concentration, 223, 224, 228, 238,
249
consequence of failure, 217
Index
consistency, 27, 28, 29, 42, 55, 64,
65, 69, 79, 109, 158
constructivism, 195, 202, 243, 254
constructivist approach, 195
context, 10, 13, 14, 15, 19, 23, 25,
37, 55, 78, 79, 82, 83, 90, 92,
114, 115, 121, 151, 154, 159,
185, 190, 195, 197, 198, 199,
200, 204, 206, 207, 208, 209,
242, 255
convexity for losses, 222
Copernicus, 121
Correspondence, 111
Cozzo, 257
creative, 10, 18, 68, 181, 182, 183,
184, 193
creative abduction, 181, 182
cultural evolution, 180, 197
cultural niche, 188
Dante, 103, 112
Daston, 169, 170, 178
Data, 198, 199, 211
Dawid, 205, 206, 209, 210
decision-making, 203, 215
Dedekind, 38, 55, 59, 60, 61, 62, 63,
65, 66, 67, 70, 131, 136, 138,
195, 253
Dedekinds, 55
deduction, 76, 77, 94, 96, 108, 140,
149, 165, 169, 171, 182, 183,
184, 191, 198, 200
deductive inference, 73, 81, 82, 83,
84, 85, 86, 87, 215
Deductive proof, 131, 138, 140
deductive proofs, 73, 74, 75, 77, 84,
87, 88, 90
deductive validity, 73
deductivism, 95
definition, 23, 27, 28, 31, 42, 46, 47,
49, 55, 60, 66, 68, 74, 81, 100,
111, 112, 135, 155, 157, 166,
200, 208, 229, 236, 249, 250,
251, 253
definitions, 29, 31, 41, 42, 44, 45,
46, 47, 60, 61, 93, 97, 158, 165,
199, 241, 250, 252
 From A Heuristic Point of View
density, 17, 218, 228
derivative, 46, 101
Descartes, 130, 149, 153, 251, 253,
259
Detelefsen, 28
Devlin, 44, 52, 144, 180, 192, 251,
254
Dewey, 53, 137, 143, 144, 145, 146,
164, 254
Dickens, 252
diminishing marginal utility, 221
Diophantus, 128
Dirichlet, 130, 152, 154, 195
discontinuity, 102, 223, 224, 238
Discourse on Method, 149
discourses, 147, 148
discovery, 10, 23, 24, 29, 33, 37, 38,
43, 67, 68, 74, 92, 107, 125, 135,
139, 147, 149, 150, 155, 163,
164, 165, 176, 183, 197, 198,
199, 213, 234, 259
distributed cognition, 179, 180, 183,
184, 186
distribution
mild approach, 218
wild approach, 218
Dummett, 83, 119, 123, 258
effectiveness of mathematics, 213
elementary geometry, 47, 131, 183
Elements, 48, 137, 177, 178
elliptic curve, 129, 135, 152, 153,
154
endogenous, 229
epistemic situations, 114
epistemology, 108, 110, 120, 149,
192, 195, 198, 241, 248
essentialism, 108
essentialist epistemology, 108
established mathematics, 127, 130,
131, 132, 133, 134, 136, 137,
138, 139, 140, 141, 142, 143
Euclid, 36, 37, 45, 48, 133, 137,
140, 155, 177, 178, 246, 260
Euler, 92, 95, 152
existential quantifier, 68
Expected Utility Theory, 221
263
experimental, 16, 17, 91, 94, 95, 96,
97, 142, 175, 183, 199
experimental mathematics, 91, 142
explanation, 37, 43, 46, 80, 82, 83,
92, 100, 117, 135, 150, 193, 203,
205, 207, 208
external representations, 183, 184,
186, 187, 188
fat tails, 218, 219, 220
Feferman, 143, 253, 254
Feigl, 197, 203, 210
Ferguson, 233, 237
Fermat, 88, 128, 145, 146, 147, 151,
152, 153, 154, 155, 156, 157,
161, 162, 167, 169, 172, 175
Fermats Last Theorem, 88, 128,
145, 147, 151, 152, 153, 154,
155, 156, 157, 161, 162
fiction, 137, 244
fides quaerens intellectum, 241
financial innovations, 233, 234
first-order arithmetical sentence, 58,
59, 62, 63, 66
first-order arithmetical sentences,
58, 59, 61, 63
first-order language, 49
Florenskij, 91, 97, 98, 99, 100, 102,
103, 104
following-a-rule, 3, 6, 7, 11, 12, 14
formal language, 31, 36, 58, 59,
136, 149, 150, 206
formalism, 25, 26, 27, 28, 36, 50,
159, 253
formalists, 94
formalization, 44, 51, 136, 158
Formalization, 49
Formulaire de Mathmatiques, 26
FORTRAN, 31
foundation, 9, 23, 24, 25, 29, 30, 34,
35, 42, 61, 132, 138, 140, 166,
168
foundational quest, 23, 24, 25
foundationalism, 23
foundationalist, 4, 10, 18, 163, 165,
169, 175, 177
foundations, 258
264
four colour theorem, 51
Four Colour Theorem, 94
Fourier, 128, 131, 138
fractal, 213, 217, 223, 224, 225, 235
free variables, 55, 56
Frege, 9, 12, 21, 24, 25, 26, 27, 30,
31, 33, 34, 35, 36, 38, 39, 108,
118, 123, 164, 183
function of damages, 217
fundamental probability set, 176
fundamental variables, 213, 214
Gabbay, 190, 191, 192, 193
Galen, 107
Galileo, 108, 214, 246
Galois, 130, 134, 154, 162
games of chance, 163, 166, 167,
175, 176
Gauss, 92, 102, 219, 249, 254
geometrical proofs, 182, 184
Geometrie, 130, 137, 171, 253
Gillies, 127, 258
Gdel, 24, 26, 27, 28, 29, 36, 38, 39,
55, 63, 66, 70, 74, 75, 90, 92, 93,
104, 156, 157, 158, 160, 162,
164, 184, 191, 196, 237, 248,
254, 256
Goldbach, 95
Gorgias, 120
Grosholz, 258
Grothendieck, 130, 140, 145, 154,
155, 157, 161, 162
Grundgesetze der Arithmetik, 9, 21
Hale, 27, 28, 39
Hempel, 199, 210
Hennig, 195, 201, 202, 203, 204,
210
Hersh, 43, 44, 53, 91, 104, 128, 136,
139, 140, 143, 144, 146, 178,
250, 253, 254, 259
heuristic, 76, 92, 93, 94, 107, 109,
110, 117, 129, 139, 142, 163,
165, 166, 168, 170, 171, 173,
174, 175, 176, 177, 181, 184,
190, 200, 213, 215, 217, 230,
235, 259
heuristic phase, 76
Index
hierarchy of types, 32
high school, 47, 48, 49, 50, 51, 100
Hilbert, 24, 25, 26, 27, 28, 32, 34,
35, 38, 39, 55, 64, 65, 66, 67, 69,
71, 109, 131, 132, 137, 164, 169,
171, 195, 253, 257
Hintikka, 182, 183, 184, 192
Hippocrates of Kos, 196
Homer, 112
Horwich, 112, 123
Hume, 27, 135, 145
Husserl, 244, 246
Huygens, 89, 163, 166, 167, 168,
169, 171, 172, 173, 174, 175,
178
Ideal Mathematician, 44, 136
idealism, 13, 255
ignorance preservation, 191
Iliad, 112
Incompleteness Theorem, 26, 27,
55, 70
inducers, 190
induction, 43, 60, 84, 86, 87, 92, 96,
103, 136, 137, 139, 149, 155,
166, 186, 190, 198, 215
inference, 18, 34, 35, 73, 76, 77, 78,
79, 80, 81, 82, 83, 84, 85, 86,
108, 173, 174, 181, 183, 186,
193, 198, 241
infinitesimals, 141
innovation, 231, 232, 233
institution, 13, 15, 19, 20
intellectual virtues, 120, 121, 122
internalized representations, 187
intolerance, 117
intuition, 29, 34, 35, 36, 61, 74, 94,
97, 108, 133, 183, 192, 200, 224,
254
intuitionists, 94
Ippoliti, 259
irrealism, 249
justification, 74, 102, 116, 135, 137,
147, 148, 166, 191, 197, 198,
199, 203, 207, 242, 250
juxtaposition, 153
Kahneman, 221, 223, 238
 From A Heuristic Point of View
Kant, 34, 35, 108, 164
Klein, 94, 104, 195
Knight, 216, 238
Knowledge assimilation, 185
Kolmogorov, 118, 123, 169, 178
Kowalski, 185, 186, 192
Kreisel, 248, 254
Kripke, 12, 150
Lakatos, 92, 94, 95, 104
Lakoff, 243, 250, 252, 254, 255
language-game, 6, 12
language-games, 5, 7, 8, 10, 13
Leibniz, 148, 155
Leng, 103, 104, 247, 249, 255
levels of reality, 201
linear logic, 68, 257
LISP, 30, 31
location-problem, 114
logical empiricism, 197, 201
logical hierarchies, 61
logicism, 25, 26, 27, 28, 31, 36
logicists, 94
Logicomix, 24, 38
MacFarlane, 118, 123
Macintyre, 148, 155, 156, 158, 159,
160, 161, 162
Mancosu, 127, 145
Mandelbrot, 218, 219, 220, 223,
224, 225, 235, 238
manipulations, 9, 10, 189, 205, 208
margin of safety, 227
mathematical activity, 41, 43, 44
mathematical discourse, 147, 151,
156, 251
mathematical experience, 44, 164,
170, 176
mathematical logic, 5, 23, 24, 25,
29, 30, 31, 32, 33, 35, 37, 41, 42,
43, 44, 47, 51, 62, 148, 150, 158
Mathematical Logic, 23, 24, 25, 30,
38, 39, 41
mathematical modelling, 201, 202,
203
mathematical models, 103, 199, 203
mathematical object, 236
265
mathematical objects, 9, 74, 92, 93,
135, 150, 151, 155, 201, 203,
213, 236, 237, 244, 247, 248,
255, 256
mathematical practice, 92, 127, 131,
135, 164, 168, 176
mathematical proof, 4, 15, 16, 17,
19, 94, 133, 142
Mathematical propositions, 8
mathematization, 213, 214, 215,
216, 221, 232, 233, 235, 236,
237
meaning, 257
medical practice, 196
medicine, 258
Meditations, 149
memorylessness, 189
menschliches Frwahrhalten, 12
mental models, 127, 135, 182, 183,
250
metaphor, 99, 166, 190, 215, 252
Miranda, 30, 32
Models, 127, 145, 198, 199, 203,
205, 211
modular form, 152, 153, 154
modus ponens, 134
multimodal, 186
multiplicity of representations, 200
NASA, 114
natural, 4, 6, 8, 9, 35, 37, 45, 49, 50,
55, 56, 58, 59, 60, 61, 62, 65, 75,
80, 83, 86, 87, 92, 93, 94, 95, 96,
97, 111, 131, 150, 151, 158, 179,
180, 181, 182, 187, 188, 190,
195, 196, 201, 214, 223, 225,
226, 229, 250, 251, 253
Natural Deduction, 260
natural kinds, 150
natural logic, 80, 97, 179, 180
natural mathematics, 179, 180, 197
natural selection, 179, 180, 188
naturalistic view, 197, 201
neo-empiricism, 91
neo-logicism, 27, 28
neutrinos, 114
266 Index

new knowledge, 74, 77, 80, 81, 82, philosophical turn, 214
107, 110, 132, 236 philosophy of mathematics, 3, 4, 5,
new metaphysics, 200, 201 7, 8, 14, 19, 26, 27, 35, 37, 61,
Newton, 89, 96 73, 74, 91, 92, 135, 147, 149,
Nicomachean Ethics, 120, 123 164, 242, 245, 254, 256, 258,
Noether, 196 259, 260
nominalists, 251 physical proof, 100
non-ampliative, 78, 81, 82 Piaget, 243
non-deductive, 75, 76, 77, 78, 80, Plato, 89, 112, 120, 124, 175
166, 175, 191, 196, 197, 198, Platonic, 79, 81
215 platonism, 248, 249, 251
non-reflective, 80 Platonism, 9, 93, 253, 255
non-standard semantics, 252 plausibility, 75, 80, 83, 87, 109,
notion of context, 207 113, 114, 118, 121, 137, 196,
notion of risk, 213, 215, 216, 225, 197, 200
229, 231, 235 Plausibility, 109, 113, 114
novel features, 215 Plausible, 103, 105, 138, 146
novelty, 226, 232 pluralism, 195, 200, 205
Nez, 250, 252, 255 Polya, 53, 92, 105, 130, 139, 140,
objectivity, 108, 180, 200, 241, 248, 142, 144, 145, 146, 254
249, 250, 251, 253, 254, 255 polynomials, 153, 159
observer-independent reality, 201, pragmatism, 137, 180, 205
202 pragmatist, 198, 200, 201
Odyssey, 112 Prawitz, 73, 75, 85, 87, 109, 260
Oliver Twist, 252 Praxis, 11
ontological questions, 248 prediction, 203, 205, 208, 239
ontology, 241 predictors, 190
irrelevant, 137, 139, 158, 186, primitive analytic proposition, 27
252 Principia Mathematica, 17, 26, 39,
Ontology, 252 71
open systems, 46, 96, 165 probabilistic empiricism, 201
open world view, 164, 190, 192 probability distribution, 217, 219
paradox, 11, 12, 20, 31, 36, 81, 94, probability of failure, 217, 228
223, 257 problem of knowledge, 109, 111,
paradoxes, 42 120
particularization, 166, 173, 175 problem of truth, 111
Pascal, 167, 169, 172, 175 problem-solving, 129, 138, 153, 165,
Paulson, 216 166, 171, 172, 173, 174, 175,
Peano, 24, 25, 26, 29, 34, 35, 36, 37, 196, 201, 213, 215, 230, 259
38, 58, 131, 136, 138, 149, 155, PROLOG, 30, 31
156, 159, 160 proof, 3, 6, 10, 13, 14, 15, 16, 17,
Pearl, 205, 210, 211 18, 19, 20, 32, 33, 34, 35, 37, 42,
Peirce, 116, 118, 120, 123, 177, 43, 44, 46, 47, 49, 51, 52, 56, 57,
180, 181, 188, 192, 193, 257 58, 59, 62, 63, 64, 65, 66, 67, 69,
personal reality, 201 70, 75, 76, 77, 80, 84, 85, 87, 88,
perspicuity, 17, 18, 160 89, 90, 93, 94, 95, 99, 100, 101,
 From A Heuristic Point of View
127, 128, 129, 130, 131, 133,
134, 135, 136, 137, 138, 139,
140, 141, 142, 143, 147, 148,
149, 151, 152, 153, 154, 155,
156, 157, 159, 160, 165, 185,
189, 192, 247, 257, 260
Proto-Investigations, 3, 4, 5, 6, 7,
13, 14
provability, 63, 68, 69, 70, 109, 111
Putnam, 92, 94, 95, 105, 144, 150
Pythagoras theorem, 88
Quine, 127, 150
Random Walk, 224
realism, 13, 93, 202, 248, 254
realisms, 249
recursion theory, 42, 45
reductio ad absurdum, 37
reductio proofs, 248
Reduction, 174, 258
redundancies, 226
redundancy, 226, 227, 228
reference, 4, 7, 9, 12, 17, 20, 47, 97,
109, 117, 131, 147, 148, 149,
150, 151, 153, 155, 156, 159,
160, 206, 207, 208, 221, 249
referent, 150, 151
reflective, 79, 80, 83, 90
Reflective inferences, 79, 80
Regelfolgen, 4, 5, 7, 8, 11, 12, 14,
19
regress, 83, 134
reifications, 250
reliability engineering, 226
Remes, 182, 192
resilience, 229, 230
Resnik, 247, 255
reverse mathematics, 45, 53
re-writing, 148, 159, 160
Ribet, 88, 129, 152, 153, 162
Riemann, 95, 150, 153, 154, 195
rigorous, 42, 105, 129, 132, 137,
138, 139, 140, 179, 180, 181,
191
rigour, 50, 108, 189, 200
risk
co-evolutionary view, 229, 230
267
Evolutionary point of view, 225
Fractal point of view, 223
heuristic point of view, 213, 230
orthodox approach, 217
Probabilistic point of view, 217
Psychologistic point of view, 221
risk management, 215, 216, 225,
230, 233, 237, 239
risk-free, 216
robotics, 147
robustness, 225
Rota, 43, 44, 134
rule-governed activities, 3, 7, 9, 12
Russell, 9, 10, 19, 24, 25, 26, 30,
31, 32, 33, 35, 36, 39, 74, 89,
150, 183, 243, 256
Russian mathematics, 100
satisfiability, 55, 56, 64, 69, 70
scaling, 219, 220, 223, 224, 238
Schnemann, 213, 214, 238
scientific revolution, 108, 214
scientific structures, 200
securitization, 235
self-correction, 188
self-evident, 89, 133, 163, 165, 169
semantic, 127, 134, 136, 182, 184,
185, 189
semantic tableaux, 182, 184, 189
semiotic anchors, 183, 193
serendipity, 23, 24, 25, 29, 33, 37
Serendipity, 23, 33, 39
Shapiro, 248, 253, 255
skeptical expert, 136
social construct, 203
social constructivism, 94, 243
social reality, 201, 255
Solly, 23
Sornette, 229, 230, 239
Soros, 234, 235, 239
Stability, 219
stable representations, 203
statistical data, 208
statistics, 195, 201, 203, 216, 236
Steiner, 259
stress, 45, 48, 181, 184, 227, 229,
230, 251
268
stressor, 229
subjective probability, 204
Sudoku, 84
superpositions, 148
Suppes, 195, 198, 199, 200, 201,
205, 210, 211
syllogistics, 134
syntactic, 82, 127, 134, 136, 185
tail of the distribution, 218, 219
Taleb, 223, 224, 225, 226, 228, 238,
239
Taniyama-Shimura conjecture, 88,
129, 152, 153, 154, 157
Tarski, 26, 33, 39, 118, 124, 158,
211
technology, 37, 131, 143, 229, 232
Thagard, 186, 193
theorem prover, 32, 51
theorem-proving, 165
Thomae, 9
Thomas, 260
truth, 1, 9, 51, 56, 65, 68, 74, 79, 80,
83, 84, 86, 88, 93, 94, 107, 108,
109, 110, 111, 112, 113, 114,
115, 116, 117, 118, 119, 120,
121, 122, 123, 128, 132, 133,
134, 135, 136, 137, 138, 149,
150, 152, 181, 186, 188, 197,
200, 203, 257
truth preservation, 83
Index
Turing, 30, 39, 190, 193, 253
Tversky, 221, 223, 238
Tymoczko, 91, 94, 104, 105, 142
type theory, 32, 36, 155
bereinstimmung, 20
undecidable proposition, 26
universal quantifier, 43, 49, 68
usefulness of deductive rules, 198
validation, 92, 95, 96, 197
validity, 19, 35, 57, 63, 68, 70, 73,
84, 95
VaR, 217, 218, 219, 220, 233
vindication, 4, 197, 203
volatility, 218, 224
Walpole, 23
warranted assertibility, 127, 132,
137, 138, 139
Weierstrass, 35
Whitehead, 26, 30
Wiles, 88, 128, 129, 130, 133, 134,
136, 138, 140, 146, 147, 148,
151, 152, 153, 154, 155, 156,
157, 159, 160, 162
Wittgenstein, 3, 4, 5, 6, 7, 8, 9, 10,
11, 12, 13, 14, 15, 16, 17, 18, 19,
20, 21, 137, 184, 244
Woods, 190, 191, 192, 193
Wright, 15, 21, 27, 28, 39, 118, 124
Zermelo, 32, 131, 133, 138