Professional Documents
Culture Documents
Editor
Keith Lehrer, University of Arizona, Tucson
Associate Editor
Stewart Cohen, Arizona State University, Tempe
The titles published in this series are listed at the end of this volume.
THE EPISTEMOLOGY
OF KEITH LEHRER
Edited by
ERIK J. OLSSON
University of Constance, Germany
ISBN 1-4020-1605-0
Cover art: MetaMe, Keith Lehrer, 2002, Oil painting, 16˝ × 20˝
PREF ACE. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. VB
INTRODUCTION ............................................ .
Erik J. Olsson: "The Epistemology of Keith Lehrer"
Chapter 1,
Ernest Sosa: "Epistemology: Does it Depend on Independence?" . . . . . . .. 23
Chapter 2
John Greco: "Why Not Reliabilism?" ............................. 31
Chapter 3
Jonathan L. Kvanvig: "Justification and Proper Basing" ............... 43
Chapter 4
Todd Stewart: "Lehrer on Knowledge and Causation" ................ 63
Chapter 5
Volker Halbach: "Can we Grasp Consistency?" ..................... 75
Chapter 6
Glenn Ross: "Reasonable Acceptance and the Lottery Paradox:
The Case for a More Credulous Consistency" ....................... 91
Chapter 7
Charles B. Cross: "Relational Coherence and Cumulative Reasoning" . .. 109
Chapter 8
Wolfgang Spohn: "Lehrer Meets Ranking Theory" .................. 129
Chapter 9
Carl G. Wagner: "Two Dogmas ofProbabilism" .................... 143
vi
TRUSTWORTHINESS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 153
Chapter 10
James Van Cleve: "Lehrer, Reid, and the First of All Principles" ....... 155
Chapter 11
G. J. Mattey: "Self-Trust and the Reasonableness of Acceptance" ...... 173
Chapter 12
Richard N. Manning: "The Dialectic Illusion ofa Vicious Bootstrap" ... 195
Chapter 13
Hans Rott: "Lehrer's Dynamic Theory of Knowledge" ............... 219
Chapter 14
Gordian Haas: "Some Remarks on the Definition of
Lehrer's Ultrasystem" ......................................... 243
Chapter 15
Jacob Rosenthal: "On Lehrer's Solution to the Gettier Problem" ....... 253
Chapter 16
John W. Bender: "Skepticism, Justification and the
Trustworthiness Argument" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 263
Chapter 17
Peter Klein: "Coherence, Knowledge and Skepticism" ............... 281
Chapter 18
David A. Truncellito: "The Ultrasystem and the Conditional Fallacy" ... 299
Chapter 19
Keith Lehrer: "Coherence, Circularity and Consistency:
Lehrer Replies" ............................................. 309
PREFACE
Erik J. Olsson
Vll
Introduction
Erik J. Olsson
University of Constance
Fido sees that there is a bone on the plate, but does Fido know that there
is a bone on the plate? David, a two-year-old, sees that the door to the
refrigerator is open, but does he know that it is open? Examples such as these
prompt very different reactions from philosophers. Some think it is obvious that
Fido and David know, and that they know in the same sense as adult humans do.
Others respond, equally emphatically, that they do not know, at least not in the
same way as adult humans. Philosophers ofthe latter inclination may grant that
someone who believes that Fido knows is allowed to use the term 'know' in any
way he or she wishes and, further, that there might be some point in defining a
concept of knowledge applicable also to Fido; but, they will urge, that concept
will not be of great interest if what we really care about is human knowledge in
its characteristic form.
Keith Lehrer belongs to the second category of epistemologists for whom
the mere possession of correct information is insufficient for human knowledge,
however reliable the source delivering the information may have been. In order
to know one must, in addition, recognize that the information one possesses is
correct. This additional demand for reasons internal to the subject
-characteristic of the position known as internalism-excludes poor old Fido,
and probably also David, neither of whom can plausibly be credited with the
conceptual resources required for such recognition, for "[t]hey lack any
conception of the distinction between veracity and correct information, on the
one hand, and deception and misinformation, on the other" (Lehrer, 2000, p. 11).
Why is it so important to recognize that one's sources are reliable? Why
does it not suffice that they actually are reliable, that the belief was caused in a
reliable way? Lehrer's answer, if! understand him correctly, is that the role of
knowledge in human reasoning is essential to its nature (ibid., p. 6), and one role
of knowledge concerns its employment in reasoning, e.g., in confirming some
hypotheses and refuting others. It is essential to knowledge that it enables us to
"reason about what is true or false, what is real and unreal" and to justify our
knowledge claims "in critical discussion and rational confrontation" (ibid., p.
11). Thus, knowledge, as Lehrer conceives it, is, in its essence, "inextricably
woven into reasoning, justification, confirmation, and refutation" (ibid., p. 6).
An externalist will certainly agree that knowledge plays an important role
in reasoning, but he or she will typically resist the conclusion that this role is
essential to its nature. He or she will concede that it is a good thing to have
reasons for one's beliefs-for instance, in convincing others that we
know-while insisting that having such reasons is not an ingredient in the very
concept of knowledge (see for instance Dretske, 1991). An externalist might
even grant that having reasons is part of the pre-systematic concept of
knowledge, and yet argue that there are good grounds, in this case, to depart
from it in favor of an allegedly more fruitful externalist conception. Several of
the papers in this volume address, directly or indirectly, the
internalism/externalism issue, e.g., the articles by Ernest Sosa, John Greco, and
Volker Halbach. Lehrer's view on justification and causation is discussed in the
papers by Jonathan L. Kvanvig and Todd Stewart. Highly relevant in this
connection is also the article by James Van Cleve.
The main purpose of this introduction is to survey the main ideas in
Lehrer's epistemology, so as to provide the necessary background against which
the other papers in this volume can be more readily appreciated. Another aim is
to point out what might be some difficulties in Lehrer's view. A majority of
these issues are explored in greater detail in the other contributions to this book,
and I have added references to guide the reader to the corresponding places. This
introduction, then, is also intended to serve as a conceptual and argumentative
map of the present book. Unless otherwise indicated, references to Lehrer are to
the 2000 edition of his Theory of Knowledge (abbreviated TK). Theory of
Knowledge, an extended and thoroughly revised version of Lehrer's early book
Knowledge, was published for the first time in 1990. Between the two editions
there are some interesting differences that will play a role later in this
introduction.
a false statement. The last clause is intended to take care of troublesome Gettier
examples, a topic I will return to later.
The particular interest of Lehrer's theory lies of course in its details.
Taking truth for granted, let us focus first on the condition of acceptance: for S
to know that p, S must accept that p. Why "acceptance" and not "belief', and
what might be the difference between the two? Acceptance, Lehrer writes, is an
attitude defined in terms of some purpose and involves an evaluation of whether
the attitude fulfills the purpose. Moreover the special kind of acceptance
relevant to knowledge is acceptance for the purpose of "attaining truth and
avoiding error with respect to the very thing one accepts" (p. 13): to accept that
p if and only ifp. Belief, on the other hand, is not defined in terms of a purpose.
Belief may happen to serve the purpose of attaining truth and avoiding error but
it is not defined in terms of that, or any other, purpose. We may, to take Lehrer's
example, believe that a loved one is safe because of the comfort of believing
this, and not because of an intrinsic interest in the truth of the matter. Belief,
Lehrer argues, is not the attitude characteristic of genuine knowledge;
acceptance is. Another feature of acceptance that will playa role later is that it
is a functional state, being characterized by the role it plays in thought, inference
and action.
Obviously much hinges on the third condition of personal justification.
In Lehrer's view, such justification amounts to coherence with a background
system. The relevant background system---called the evaluation system
-consists of three parts: the acceptance system, the preference system and the
reasoning system. The acceptance system is the core of the evaluation system
and is defined as the set of states of acceptance of S described by statements of
the form "S accepts that p" attributing to S just those things S accepts at t with
the objective of obtaining truth and avoiding error with respect to the content
accepted, that is, with respect to the content that p (p. 130). In the 1990' sedition
of TK, the background system was equated with the acceptance system.
Suppose, to take an example, that S accepts that Paris is the capital of
France. That might lead one to expect that the statement "Paris is the capital of
France" should be an element of S's acceptance system. But this, as we just saw,
is not how Lehrer defines the notion. Rather, the acceptance system contains the
statement "S accepts that Paris is the capital of France". So, when Lehrer writes
that in personal justification we must start with what we accept (p. 123), this
does not mean that we are allowed to take the truth of "Paris is the capital of
France" and other propositions we accept for granted. It means only that we may
take for granted that we accept those things, i.e., that we take a certain attitude,
that of acceptance, towards those propositions. Lehrer sometimes calls the set
of all propositions p such that "S accepts that p" is in the acceptance system the
content of that system, a practice I will follow here.
4 THE EPISTEMOLOGY OF KEITH LEHRER
2. TRUSTWORTHINESS
Where does self-trust or trustworthiness come in? Knowledge, as Lehrer
defines it, requires personal justification with the evaluation system of the
person. But as that system is defined preciously little will be personally justified.
Statements of the form ItS accepts that pIt constitute too meager a basis for
ERIK J. OLSSON 7
circumstances allow" (p. 140). For more on the intended interpretation of the
principle of trustworthiness, see the first chapter in Lehrer's Self-Trust.
There are corresponding principles of trustworthiness for the other parts
of the evaluations system. Thus I may accept that I am trustworthy in what I
prefer with the objective of preferring to accept somethingjust in case it is true.
By an argument paralleling the trustworthiness argument for acceptance I may
then conclude that I am reasonable in having a given preference. By the same
token, I may accept that I am trustworthy in how I reason, from which I may
conclude that I am reasonable in my reasoning to a given conclusion.
My acceptance of the principle of trustworthiness does not automatically
make my other acceptances justified; but it does make them reasonable and
hence in that way it contributes to their justification. This raises the question of
how to justify trustworthiness itself. Trustworthiness can hardly contribute to the
justification of other acceptances unless it is itself justified or at least
reasonable. Lehrer's basic answer is that trustworthiness applies not only to other
acceptances but also to itself. We recall that, for any proposition p which I
accept for the purpose of obtaining truth and avoiding error, I may reason from
the acceptance of my trustworthiness to the reasonableness of my accepting that
p. As a special case, I may reason from the acceptance of my trustworthiness to
the reasonableness of my accepting that I am trustworthy.
Lehrer has characterized the special role played by self-trust using an
analogy from Thomas Reid: ''just as light, in revealing the illuminated object,
at the same time reveals itself, so the principle, in rendering the acceptance of
other things reasonable, at the same time renders the acceptance of itself
reasonable" (p. 143). Reid's epistemology has had a profound influence on
Lehrer's theorizing about self-trust. On Lehrer's interpretation, one of Reid's
principles of common sense is applicable to all other principles including itself.
See Lehrer (1989) and, for a discussion of Lehrer's Reid interpretation, James
Van Cleve's article in the present volume.
As in the case of other acceptances, the trustworthiness argument falls
short of establishing that I am justified in my acceptance of (1). It only
establishes that I am reasonable in my acceptance (provided that the premises
are true). For me to be justified in my acceptance of (1), all objections to that
claim have to be answered or neutralized. In order to answer or neutralize such
objections I will need to refer to other things I accept. Hence, even in the case
of (1), background information is required in order for me to be personally
justified in accepting it.
The principle (1) does not justify itselfbut depends for its justification on
the background system of other things we accept. Therefore it would be
incorrect to call it a basic belief in the foundationalist sense. Lehrer has
suggested that (1) is more like a keystone in an arch. Without the keystone, the
ERIK J. OLSSON 9
arch would collapse; at the same time the keystone is supported by the other
stones in the arch.
Is the circularity involved in the argument from my acceptance of my
trustworthiness to the reasonableness of that acceptance a vicious one? Lehrer
has argued that it is not vicious by pointing out that his intention is not to use (T)
as a premise to prove something to a skeptic, but rather to use it for explanatory
purposes, the claim being that the principle of trustworthiness can be used for
explaining why it is reasonable for us to accept what we accept. Indeed, the shift
from justification to explanation makes it less obvious that the circle is vicious.
Lehrer even thinks that circularity in this case may be a virtue rather than a vice.
It is, he notices, preferable to leave as little as possible unexplained. Hence, an
explanation that does not only explain why other acceptances are reasonable but
also why its own acceptance is reasonable is better in this respect that an
explanation which accomplishes the former but not the latter.
Several contributors to this volume express dissatisfaction with Lehrer's
discussion of trustworthiness. As some point out, what Lehrer says about the
possible explanatory merits of self-trust seems irrelevant to the issue of
justification. Lehrer's view on the matter is examined in the contributions by
James Van Cleve, G. J. Mattey and Richard N. Manning. The papers by John W.
Bender, John Greco and Peter Klein contain related material. Manning, for
instance, contends that difficulties arise because of Lehrer's presupposition that
the principle of trustworthiness is something that needs to be argued for in the
first place. In Manning's view, by contrast, self-trust is a transcendental
condition on the possibility of our epistemic practice.
What has been said so far raises the following question: In what sense, if
any, is Lehrer's theory a coherence theory, as Lehrer claims it is (at least in
part)? If one takes it as essential that such a theory make use of a concept of
systematic or global coherence, then Lehrer's theory is clearly not a coherence
theory. For in Lehrer's view, "[c]oherence ... is not a global feature of the
system" (1997, p. 31). Rather, what he calls coherence, as we have seen, is a
relation between an evaluation system and a proposition. This relation,
moreover, "does not depend on global features of the system" (ibid.). This
notwithstanding, Lehrer has said that the content of the acceptance system
should be logically consistent, thus referring to the global feature of consistency
(1991, p. 131). Lehrer has also addressed the issue of consistency in his paper
"Reason and Consistency", reprinted as Chapter 6 in Metamind. The role of
10 THE EPISTEMOLOGY OF KEITH LEHRER
know, say, that I see a tree over there, unless I also accept my trustworthiness in
relevant matters. But is it psychologically realistic to suppose that acceptances
about other acceptances-in particular about their trustworthiness-are
necessary for knowledge? It seems that people often make perceptual claims
without ever having thought about the trustworthiness of their perceptual
faculties. Are we then forced to say that they do not know what they claim to
know?
While Lehrer concedes that some "unrealistic theory of belief maintaining
that all beliefs are occurrent states may yield the consequence that we lack such
beliefs [about trustworthiness]" (p. 202), he contends that his own theory of
acceptance avoids this conclusion. For according to the latter the mental state of
acceptance is a functional state, one that plays a role in thought, inference and
action. As a consequence, I may be said to accept that p at a time t even if my
acceptance ofp is not something I am contemplating at t. What is essential to my
acceptance of p is how I think, infer and act. This holds, in particular, for my
acceptance of my trustworthiness, which need not be present to my mind either.
As Lehrer puts it, "[ w]e think, infer, and act, in a way manifesting our trust in
what we accept" (ibid.). He concludes that "it is appropriate and not at all
unrealistic to suppose that, in addition to the other things we accept, we accept
our own trustworthiness and the reliability of it as well" (ibid.).
But does this clever defense against charges of lack of psychological
realism really work? Let us first ask this question: What would be the difference
between the functional role of my acceptingp only and the functional role of my
accepting, in addition, my trustworthiness in matters concerningp? Suppose that
I accept that p without accepting my trustworthiness in matters concerning p.
This means that I am inclined to think, infer and act in a certain way. More
precisely, my acceptance of p amounts to my being "inclined to assent to it
automatically, to draw inferences based on the assumption of it, and, in general,
to act as though it were true" (Lehrer, 1989, p. 270). For instance, suppose that
I am planning to take a walk. My acceptance that it is raining will then manifest
itself in my acting as though it were true, e.g., in my being inclined to bring my
umbrella. My acceptance that it is raining will also manifest itself in my being
inclined to infer, and thus to accept, further propositions as well, e.g., that the
ground will be wet. This new acceptance in turn will make me inclined to put on
my boots rather than my less sturdy Italian shoes, and so on.
Now add to my acceptance that p my further acceptance that I am
trustworthy in matters concerning p. My suspicion is that this addition
contributes nothing to how I will think, infer and act beyond what was already
part of my acceptance of p. To continue the example, my acceptance that it is
raining will manifest itself in my being inclined to bring my umbrella and infer
that the ground will be wet, and so on. Add to this my acceptance of my
trustworthiness in telling whether it is raining or not. This addition has no
12 THE EPISTEMOLOGY OF KEITH LEHRER
further effect on my dispositions to think, infer and act. I will still be inclined to
bring my umbrella and to infer that the ground will be wet, and no new
dispositions to think, infer and act seem to come forth. From the point of view
of how I think, infer and act, there is no difference between, on the one hand, my
accepting that p and, on the other hand, my accepting that p and, in addition, that
I am trustworthy regarding p. Rather, these two descriptions-"my accepting
that p" and "my accepting that p and, in addition, that I am trustworthy regarding
p"-are merely two different characterizations of one and the same functional
state, namely, that in which I accept that p.
The upshot of all this is to confront Lehrer with the following dilemma.
Without a functional theory of acceptance, his theory is indeed vulnerable to
charges of lack of psychological realism. But the same functional theory which
solves this problem so neatly has the unwelcome effect of trivia Ii zing the very
idea of trustwort hi ness, reducing trustworthy acceptance to mere acceptance. A
related objection is raised in John Greco's contribution to this volume.
owns a Ferrari than that no one does. No one in your class owns a
Ferrari. (My italics.)
Lehrer now maintains that the claimant, on the basis of her ultrasystem, will not
be able to answer or neutralize the ultracritic's objection.
As I understand it, Lehrer's analysis of the hard Ferrari example is
intended to illustrate the claim that "[t]he proper solution to these [Gettier]
problems may be obtained from the ultrajustification game by extending the role
of the critic in the game to include considerations of preferences and reasonings
that supplement the acceptance system in the evaluation system of the claimant"
(p. 160). Clearly, this particular example must be taken to illustrate
considerations of reasonings.
Now, the ultracritic may instruct the claimant to give up a piece of
reasoning only if that reasoning is unsound. In the example, she instructs her to
give up her reasoning from the evidence about Mr. Nogot's driving a Ferrari etc.
to his owning a Ferrari. But is that reasoning really unsound? I think not. The
fact that the premises are true but the conclusion false in the example only serves
to disprove the deductive validity of the inference, but this is beside the point,
since the claimant is not committed to its deductive validity but at most to its
inductive validity. As an inductive inference it is sound. For the conclusion that
Mr. Nogot owns a Ferrari is highly probable on the evidence of his driving one,
showing papers stating that he owns one etc.
But perhaps Lehrer did not after all intend his new analysis of the hard
Ferrari example to illustrate the elimination of reasonings from the evaluation
system. Perhaps we should think of the hypothetical linking the evidence about
Nogotto his owning a Ferrari not as belonging to the reasoning system butto the
acceptance system (as in the 1990 edition of TK). But in that case it is unclear
what role the reasoning system plays in Lehrer's analysis of Gettier cases. Does
it perform any useful function at all?
Hence, it is not evident that Lehrer's new approach to the Gettier problem
amounts to an improvement in comparison to the old. Of course, some logicians
that were attracted to the 1990 version of Lehrer's theory primarily because of
its interesting relationship to belief revision theory will be quite happy with this
observation. Yet, Lehrer might be able to derive some support for his new
approach from Gordian Haas's article in this volume, in which Haas presents a
formal argument to the effect that the complicated elimination and replacement
operations occurring in the older approach are actually superfluous. Jacob
Rosenthal argues, in his contribution, that both solutions presented by Lehrer are
unsatisfactory. Undefeated justification and the Gettier problem are also
discussed in the papers by John Greco, Peter Klein and Wolfgang Spohn.
16 THE EPISTEMOLOGY OF KEITH LEHRER
such a way that the explicatum can be used in most cases in which the
explicandum has so far been used, but the similarity need not be very close. The
explicatum, besides being exact, should also be fruitful in the sense of being
useful for formulating universal statements (empirical laws and logical
theorems). Finally, the explicatum should be as simple as the other more
important desiderata permit.
Lehrer's characterization of critical epistemology conveys the impression
that he is essentially a common sense philosopher in the tradition of G. E.
Moore. Like G. E. Moore, he repudiates the idea that epistemology should
engage in radical doubt concerning common sense knowledge claims. Yet being
based on a general view as to the nature of philosophical and scientific
definitions, Lehrer's approach is more sophisticated than Moore's position,
which, lacking a more solid methodological foundation, amounts to ruling out
skepticism on the sole grounds that it conflicts with common sense.
It is not difficult to appreciate that from the point of view of Lehrer's
critical epistemology, even in the sophisticated form referring to Carnap's notion
of explication, radical skepticism can be dismissed for methodological reasons
alone. For a radical skeptic will want to define knowledge so that most or all of
our common sense knowledge claims turn out false, e.g., by insisting that
anything worthy of the name must be such that there is no logical possibility of
error. Hence, any such attempt will score very poorly as regards Carnap' s
criterion of similarity to the explicandum. To compensate for this, it would have
to do extremely well in other respects, i.e., regarding exactness, fruitfulness and
simplicity. But even ifit can be given an exact and simple definition, a skeptical
conception of knowledge will be a complete failure with respect to the important
desideratum of fruitfulness: a concept under which nothing, or very little, falls
will be utterly useless for articulating laws and theories. We would not accept
a definition of water that makes nothing qualify as water; and we should be
equally discontent with a definition of knowledge that makes nothing qualify as
such. Almost any other proposal one could think of will fare better.
The (sophisticated) common sense position is presented in the
introductory chapter of TK, but its influence seems to decline as the book
develops, and in Chapter 9 radical skepticism makes a surprising comeback as
an epistemological threat that must be "answered". Lehrer's reasoning in
defense of what is, in effect, a moderate skepticism is not easy to follow, and I
will not be able to cover all details here. Nonetheless, the main idea seems to be
the following. We cannot rule out the possibility that we may be in error
regarding what we accept because of the general fallibility of our faculties and
because we may be deceived by a Cartesian demon. However, our fallibility
does not imply a lack of personal justification on our part in accepting what we
do accept. For we may still be able to answer or neutralize all objections on the
basis of our acceptance system, or so Lehrer claims. This is accomplished by our
18 THE EPISTEMOLOGY OF KEITH LEHRER
I have great sympathy for Lehrer, the sophisticated common sense philosopher,
who has a simple and, in my view, entirely satisfactory answer to skepticism
based on some general methodological considerations concerning the nature of
philosophical and scientific definitions. Lehrer's answer to skepticism is
critically assessed in the contributions by John W. Bender and Peter Klein.
There is also the issue of whether Lehrer's theory implies radical
skepticism concerning facts of acceptance. Robert Shope (1978) has argued that
Lehrer, in his account of undefeated justification, commits the "conditional
fallacy". A similar objection is raised by Peter Klein in this volume. Suppose
that I accept some proposition p which is actually false. While p's falsity
prevents me from knowing that p, it should not prevent me from knowing that
1 accept that p. But Lehrer's theory seems to have just this consequence: if p is
false, 1 cannot know even that 1 accept that p. The reason is that "I accept that
p" will not be in the ultrasystem, making it impossible for me to answer or
neutralize the objection that 1 do not accept that p. Lehrer, as a matter of fact,
has addressed this problem in the latest edition of TK, in which the ultrasystem
is required to acknowledge the existence of the eliminated states of acceptance,
preference and reasoning in the original evaluation system (p. 160). The
conditional fallacy is discussed in David A. Truncellito's contribution to this
volume. Truncellito maintains that the objection rests on a misunderstanding of
Lehrer's account.
The purpose of this introduction has been merely to give a brief overview
of Lehrer's theory of knowledge and justification, and to indicate what might be
possible troubles. It goes without saying that there is much more to be said about
Lehrer's epistemology. My perhaps most serious omission concerns his and Carl
G. Wagner's celebrated theory of rational consensus which they have developed
in a series of papers and an influential book, Rational Consensus in Science and
Society. Wagner's paper in the present volume discusses some probabilistic
issues of importance to this other, no less important, aspect of Lehrer's
epistemological theorizing. 3
ENDNOTES
1 Lehrer writes on p. 10 in TK: "We shall be concerned with an analysis that will be useful for
explaining how people know that the input (the reports and representations) they receive from
other people, their own senses, and reason is correct information rather than error and
misinformation. A person may receive a representation that p as input without knowing that the
representation is correct and, therefore, without knowing that p."
2 For reasons that that will emerge later, the ultrasystem is also required to acknowledge the
existence ofthe eliminated states of acceptance, preference and reasoning in the original evaluation
system (p. 160).
3 Thanks are due to Gordian Haas for his comments on an earlier version of this introduction.
20 THE EPISTEMOLOGY OF KEITH LEHRER
REFERENCES
Carnap, R. (1950), Logical Foundations ofProbability, The University of Chicago Press, Chicago.
Dretske, F. (1991), "Two Conceptions of Knowledge: Rational vs. Reliable Belief', Grazer
Philosophische Studien, Vol. 40: 15-30.
Lehrer, K. (1974), Knowledge, Clarendon Press, Oxford.
Lehrer, K. (1975), "Reason and Consistency", in Analysis and Metaphysics: Essays in Honor of
R, M Chisholm, Lehrer, K. (ed.), Dordrecht.
Lehrer, K. (1989), "Reply to My Critics", in The Current State ofthe Coherence Theory, Bender,
J. W. (ed.), Philosophical Studies Series, Vol. 44, Kluwer, Dordrecht.
Lehrer, K. (I 990a), Metamind, Clarendon Press, Oxford.
Lehrer, K. (1990b), Theory of Knowledge, First Edition, Westview Press, Boulder.
Lehrer, K. (1991), "Reply to Mylan Engel", Grazer Philosophische Studien, Vol. 40: 131-133.
Lehrer, K. (1997), Self-Trust: A Study of Reason, Knowledge and Autonomy, Clarendon Press,
Oxford.
Lehrer, K. (2000), Theory of Knowledge, Second Edition, Westview Press, Boulder.
Lehrer, K, and Wagner, C. G. (1981), Rational Consensus in Science and SOCiety, Reidel,
Dordrecht.
Olsson, E. 1. (1998), "Competing for Acceptance: Lehrer's Rule and the Paradoxes of
Justification", Theoria, Vol. 64: 34-54.
Olsson, E. J. (1999), "Cohering With", Erkenntnis, Vol. 50, Nos. 2-3: 273-291.
Rott, H. (2001), Change, Choice and Inference, Oxford Logic Guides No. 42, Oxford University
Press.
Shope, R. (1978), 'The Conditional Fallacy in Contemporary Philosophy", Journal ofPhilosophy,
Vol. 75: 397-413.
EXTERNALISM VS. INTERNALISM
Chapter 1
Ernest Sosa
Brown University and Rutgers University
That does not affect the truth or falsehood of what we say when we say that in
that group what matters to whether the counterpart of S is tall is the
distribution of heights in the group. Here we are of course relying on what we
mean by 'tall' and on whatever contextual constraints there may be on our
present proper and truthful use of such terminology. So it is important to
distinguish our question of whether tallness-relative-to-G supervenes on the
distribution of heights alone, from the question of whether the correctness of
applications of 'tall (for group G)' in group G is entailed simply by the
distribution of heights. (Obviously it is not entailed simply by that; such
correctness is also crucially dependent on the standards operative in group G
for the use of the relevant terminology, which need not coincide with our
standards. )
Our question should also be distinguished, moreover, from the question
whether our attributions of 'tall (for group G)' have their correctness
determined simply by the distribution of heights. Here again, the answer to
our correctness-of-attribution question is clearly in the negative, since here
again a crucial factor determinative of such correctness will be the standards
and meanings for the predicate among the relevant set of speakers, those the
correctness of whose attributions is at issue. But this further factor is surely
not determinative of which of us is actually tall; it is determinative rather of
which of us is now truthfully and properly said to be "tall" (which in turn is
distinct from the question of which of us is now truthfully and properly said
to be tall.)
this would give that theory an advantage in competition with its rival theories
of that epistemic status. In that case, it would seem that our foundationalism
of enough clarity and distinctness would share the supposed advantages of
coherentism.
ENDNOTES
I This paper originated as a talk in a Symposium in honor of Keith Lehrer held as part of the 2001
Pacific Division meetings of the American Philosophical Association. I remember meeting Keith
at an APA "smoker," as they were called in the early sixties. Keith was then a professor in the
legendary Wayne State department, I a graduate student at Pitt. Little did we know that our paths
would cross so often in the years to follow, always to my benefit, as Keith realized his brilliant
potential, with many publications on issues of main interest to us both, especially in epistemology.
Though we started out far apart, over the ensuing decades our relevant views have converged
steadily. To dwell on agreement would be boring, however, so here I will discuss one main
remaining disagreement.
2 Self-Trust: A Study of Reason, Knowledge, and Autonomy (Oxford University Press, 1997), p.
62.
3 Ibid., p. 65.
4 Ibid., p. 67.
30 EPISTEMOLOGY AND INDEPENDENCE
John Greco
Fordham University
When I was a graduate student I would have gone to great lengths to talk
with Keith Lehrer about epistemology. In fact, I did! Hearing that Lehrer would
be talking at my alma mater, Georgetown University, I drove down to
Washington D.C. from Providence, listened to his talk, and got myself invited
to dinner. Much to the Georgetown students' horror, and probably Lehrer's as
well, I proceeded to monopolize his time with my questions and commentary.
(Sorry about that, Keith.) In any case, it is a great honor for me now to be
invited to engage Lehrer more formally and more appropriately. I very much
appreciate this opportunity.
In his Theory ofKnowledge Lehrer tells us that his coherentist theory can
be regarded as a form of reliabilism. 1 His reasoning is that, on his view,
reliability is a necessary condition of both undefeated justification and
knowledge. More specifically, personal justification requires that one accept
both of the following principles:
3. Gettier problems show that true reliable acceptance is not sufficient for
knowledge. However, coherence requires further conditions, and these
work so as to solve Gettier problems.
In the next section I will consider Lehrer's first argument rather briefly.
I will then tum to the opacity objection in a more extended way. Here my
argument will be this: Depending on how we interpret Lehrer's conditions for
coherence, we must say either a) that Mr. Truetemp satisfies those conditions
and therefore knows, or b) that in ordinary cases of perception, people do not
satisfy those conditions and therefore do not know. If (a), then the counter-
examples employed by the opacity objection fail to distinguish Lehrer's
coherentism from non-coherentist versions of reliabilism. If (b), then Lehrer's
coherentism has unacceptable skeptical results. I will end by considering
Lehrer's treatment ofGettier problems. Here I will argue that a non-coherentist
version of agent reliabilism can adopt a strategy very similar to Lehrer's.
JOHN GRECO 33
Lehrer has long endorsed a broadly Sellarsian theme: that knowledge properly
so-called involves abilities to articulate and give one's reasons, and to defend
one's knowledge and reasons against relevant objections. In the second edition
of his Theory of Knowledge, for example, Lehrer writes:
This same theme is the basis for objections to both foundationalism and
externalism. Thus Lehrer objects to a foundationalist account of
perceptual beliefs on the following grounds:
The problem is now this: The present objection to externalism, and the
motivation for turning to a coherentist version of reliabilism, is that knowledge
requires background information for the purposes of reasoning, justification,
confirmation, and refutation. But now, it turns out, the background information
that coherence provides need not be available for use in these activities. It
would seem that Sellarsian themes regarding the role of knowledge in science
and practical wisdom cannot be used to motivate Lehrer's coherentism over non-
coherentist reliabilism.
I will now argue that there is no way to interpret Lehrer's theory so that
all of these conditions are fulfilled. More specifically, on some interpretations
of Lehrer's theory, Mr. Truetemp knows. On other interpretations, ordinary
cases of perception do not count as knowledge.
First, it is not clear why Mr. Truetemp does not know on Lehrer's
account. Specifically, it is not clear why Truetemp is not in the same position,
vis-a.-vis coherence, as knowers are in ordinary cases. Consider, for example,
Lehrer's remarks concerning knowledge and skepticism.
In general, Lehrer thinks, we can invoke our trustworthiness to reason that some
acceptance is reasonable.
I accept that p with the objective of accepting that p just in case it is true.
The relevant question is, why can't Mr. Truetemp reason in exactly the same
way? Put differently, why isn't the reasonableness of Mr. Truetemp's
acceptance about the temperature explained in exactly the same way?
Moreover, why can't Mr. Truetemp invoke principle (TR) to reason even
further, and to conclude that his acceptances about the temperature are not only
trustworthy, but reliable?
Again, consider how, according to Lehrer, the dialectic with the skeptic
proceeds in ordinary cases.
Critic (or skeptic): Let us admit that your are intellectually trustworthy and
intellectually virtuous, as you claim. You are, nevertheless, in error
because such trustworthiness and virtue fail to achieve their purpose. What
you accept in this trustworthy and virtuous way is not reliably connected
with the truth ....
What should the claimant reply? The reply must be that the critic or
skeptic is wrong! ...
Again, why can't Mr. Truetemp follow the same dialectic? If he can, then
Truetemp knows on Lehrer's account, and so the Truetemp case does not
distinguish non-coherentist reliabilism from Lehrer's coherentism.
I assume that Lehrer thinks that Mr. Truetemp cannot follow the same
dialectic. However, it is not clear why. One reason that suggests itself is that
Mr. Truetemp accepts his thoughts about the temperature "unreflectively." But
what does this mean? One thing it might mean is that Truetemp has no further
acceptances regarding his trustworthiness and reliability in this regard, and so
does not have the requisite materials in his evaluation system to sustain a
relevant dialectic. However, it is not at all clear that Mr. Truetemp does not
have the relevant acceptances in his system. Remember, Lehrer defines
acceptances in terms of their functionality in action and thought. But according
to the example, Mr. Truetemp does act and think as if he thinks that he is
trustworthy and reliable regarding his ability to determine the temperature. He
acts and thinks more or less the way that normal people do regarding their ability
to determine simple mathematical truths, or regarding their ability to determine
colors in good light. That is, he acts and thinks as if he thinks that he is
trustworthy and reliable, although he makes no conscious judgements in this
regard. Moreover, Mr. Truetemp is reliable in the relevant regard, and therefore
his acceptances are true and convert personal justification to undefeated
justification and knowledge.
38 WHY NOT RELIABILISM?
Lehrer might object that Mr. Truetemp cannot explain his ability to tell
the temperature in any detailed way; i.e. in a way more specific than referring
to principles (T) and (TR). But neither could most people defend their ability
to tell colors in any detailed way. Most people, I assume, would simply insist
that they can tell, or else very quickly get into a quite inadequate (and probably
false) explanation as to how they can tell.
Let me be clear: The point I am making here is not that Mr. Truetemp
knows. On the contrary, I think that he does not know, and below I will offer
my own explanation regarding why he does not know. My point, rather, is that
Lehrer's coherence theory is faced with the dilemma set out above. Depending
on how one interprets Lehrer's conditions for coherence and knowledge, either
a) Truetemp has the requisite true acceptances in his evaluation system, in which
case he knows, or b) he does not, in which case neither do ordinary perceivers,
and so they don't know.
Let us consider one more reply on Lehrer's behalf, however. Lehrer
might argue that there are true objections (or competitors) in the Truetemp case,
which defeat Truetemp's personal justification, and which are not present in
cases of ordinary perception. For example, Lehrer might argue, Truetemp
cannot answer the following objecton: (0) Someone has implanted a device in
your brain, and that is why you believe what you do about the temperature.
Notice, however, that (0) is misleading: (0) counts as an objection on Lehrer's
view, since it is less reasonable for Truetemp to accept what he does about the
temperature on the assumption that (0) is true than it is on the assumption that
(0) is false on the basis ofTruetemp's evaluation system. However, Truetemp
is perfectly reliable in matters regarding the temperature, and so the objection
is misleading insofar as it suggests a reason against Truetemp's belief. But now
consider that there will also be misleading objections available against S's
acceptances in ordinary cases of perception. For example: (0') You are reliable
in your perceptual beliefs in conditions C but not C', and it is possible that you
are presently in conditions C'. Or consider: (0") There are experts in
philosophy and science who believe that apples are not red and that the sky is
not blue. How are these sorts of objections to be answered in ordinary cases of
perception? Why should we think that such objections always can be answered?
For example, consider that C' in (0') might describe conditions about which S
is wholly ignorant, or perhaps conditions that S lacks the conceptual capacities
even to consider. Similarly, S might have no idea why anyone would accept that
(0") is true, and no idea how to reply to such considerations if she did have an
idea of them.
Perhaps S can neutralize (0') and (0") by referring to principle (T)
and/or principle (TR) above. Or maybe she can neutralize the objections in
some other way. But if she can, why can't Mr. Truetemp neutralize (0) in the
same way? Again, depending on how Lehrer's theory is supposed to handle
JOHN GRECO 39
misleading objections, it would seem that either a) Truetemp has the requisite
true acceptances in his evaluation system to neutralize such objections, in which
case he knows, or b) he does not, in which case neither do ordinary perceivers,
and so they don't know. Either way, the Truetemp case does not serve Lehrer's
purposes in the opacity objection against externalism and reliabilism.
3. GETTlER PROBLEMS
Another equally important reason why reliabilism will not suffice [for the
sort ofjustification required for knowledge] is that global reliability might
be irrelevant locally. Consider again the case of Mr. Goodsumer, who
sums reliably but not in the particular case. He is not trustworthy in the
way he sums in this case. Trustworthiness must be connected with truth
in order for personal justification to convert to knowledge in the particular
case. (224)
This is a very nice idea. In Gettier cases, the person has true acceptance but it
is only a matter of luck that the acceptance is true. In cases of knowledge, the
person has true acceptance because she is trustworthy (or virtuous) in the way
that she accepts what she does. This idea does not give the advantage to
coherentism over non-coherentist reliabilism, however, because the reliabilist
can say the same thing. More specifically, the agent reliabilist can say the same
thing: in cases of knowledge, S accepts what is true because she is reliable. 4
This move is particularly attractive from the perspective of agent
reliabilism, since in general credit is closely tied to agent reliability. For
example, we give an athlete credit when we think that she accomplishes her feat
because she has great abilities. According to Aristotle, actions deserving moral
credit "proceed from a firm and unchangeable character."s The present
40 WHY NOT RELIABILISM?
ENDNOTES
IKeith Lehrer, Theory 0/ Knowledge, 2nd edition (Boulder, CO: Westview Press, 2000), p. 172.
Below, all page numbers in the text refer to this volume.
2. See my "Agent Reliabilism," Philosophical Perspectives. J3, Epistemology, James Tomberlin,
ed. (Atascadero. CA: Ridgeview Press, 1999); and Putting Skeptics in Their Place (Cambridge:
Cambridge University Press, 2000).
3. See Lehrer's "Replies" in The Current State 0/ the Coherence Theory, John Bender, ed.
(Dordrecht: Kluwer Academic Publishers, 1989); and Theory o/Knowledge, esp. pp. 39-40.
4. I argue for this account ofknowledge in "Knowledge as Credit for True Belief," in Intellectual
Virtue. Perspectives from Ethics and Epistemology, Michael DePaul and Linda Zagzebski, eds.
(Oxford: Oxford University Press, 2002).
JOHN GRECO 41
5. Nicomachean Ethics, II.4. Joel Feinberg suggests that all attributions of moral credit imply that
the action in question proceeds from character. This is probably too strong. Nevertheless, there
is a special sort of moral credit, with which Aristotle is concerned, that does imply this. See .Toe I
Feinberg, DOing and Deserving: Essays in the Theory of Responsibility (Princeton: Princeton
University Press, 1970), p. 126.
6. Doing and Deserving, p. 171. Emphasis added.
7 More needs to be said, of course. I try to say some of it in "Knowledge as Credit for True
Belief." On my own view, abnormality and manipulation do not always undermine positive
epistemic status. But they do in cases where they undermine credit.
x. I would like to thank Keith Lehrer and Ernest Sosa for helpful conversations on relevant issues.
Chapter 3
Jonathan L. Kvanvig
University of Missouri
Some thirty or so years ago, Keith Lehrer attacked the idea that causation
has much to do with knowledge or justification with the case of the gypsy
lawyer, and has more recently endorsed the same kind of attack with the case of
the racist scientist. 2 These cases threaten not only causal theories of knowledge
but also theories of knowledge or justification which require that one's evidence
be at least a partial cause of one's belief. They threaten, that is, the view that
causation is at the heart of the distinction between propositional justification, the
justification one has for the content of one's belief, and doxastic justification,
the justification which attaches to the believing itself. When justification
attaches to the believing itself rather than merely to the content of what is
believed, it is because one holds the belief in question on the basis of the
evidence. When justification only attaches to propositional contents, there is a
failure of such basing, e.g., one may have the evidence but believe for different
reasons.
This distinction is important for at least two reasons. First, only
doxastically justified beliefs are candidates for knowledge, on any theory which
requires justification for knowledge. Propositional justification is a step in the
right direction, but if one's believing itself is not justified, one cannot have met
the justificatory requirements for knowledge. Second, only doxastically justified
beliefs satisfy any purely intellectual requirement to believe claims that are
justified, for any such requirement will surely include the requirement that we
hold such beliefs for the right reason, i.e., on the basis of that which justifies
them. So a proper account of the distinction between doxastic and propositional
justification is important for a complete theory of justification and for a
complete theory of knowledge.
43
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 43-62.
© 2003 Kluwer Academic Publishers.
44 JUSTIFICATION AND PROPER BASING
1. LEHRER'S EXAMPLES
We can begin with Lehrer's examples. The first case concerns a gypsy
lawyer of a client accused of eight murders. The lawyer consults the Tarot
cards, and they say that the client is guilty of committing all but the eighth
murder. The lawyer believes what the cards say, and his conviction of the
innocence of his client regarding the eighth murder leads him to reconsider the
JONATHAN L. KV ANVIG 45
evidence, which he now comes to see conclusively establishes that his client is
innocent of the eighth murder. Lehrer then goes on to claim:
He freely admits, however, that the evidence which he claims shows that
he knows his client to be innocent of that crime is not what convinced him
of the innocence of his client, and, indeed, would not convince him now
were he not already convinced by the cards ... His conviction could not be
increased by his consideration of the evidence because he was already
completely convinced. On the other hand, were his faith in the cards to
collapse, then emotional factors which influence others would sway him
too. Therefore the evidence which completely justifies his belief does not
explain why he believes as he does, his faith in the cards explains that, and
the evidence in no way supports ... or partially explains why he believes as
he does. 4
Lehrer thus claims that the lawyer knows that his client is innocent even though
the evidence which justifies his belief does not either prompt his original
acquisition of the belief, nor does the evidence lend increased confidence in the
belief once it is discovered, nor does the evidence at present sustain the belief.
This last point is true in virtue of the fact that, if the influence of the cards were
removed, emotional factors would hold sway and the lawyer would no longer
believe that his client was innocent. Nonetheless the lawyer now knows and
justifiably believes that his client is innocent of the eighth murder.
More recently, Lehrer has forwarded the case of Ms. Prejudice:
Imagine the case of Ms. Prejudice, who out of prejudice against a race
believes that the members of the race who have a certain disease get the
disease because of their genetic makeup. Of course, she, being very racist,
believes this is a sign oftheir racial inferiority, and she is totally convinced
because of her racism that the disease is the result of the genetic
constitution of the race. Now imagine that Ms. Prejudice becomes a
medical student and learns, to her pleasure, of the medical evidence that
supports her prejudiced conviction. She becomes, however, a medical
expert of the highest quality, fully capable of separating her prejudices
from her scientific studies. As luck would have it, she becomes part of a
research team assigned the task of checking on the genetic basis of the
disease .... She wants to make sure it [the disease] really is [genetically
caused], and she will force them [her co-investigators] to investigate with
the greatest care every reason for doubting that the disease is genetically
caused. She wants to make absolutely sure that she cannot be charged
with concluding on the basis of the scientific evidence that the disease is
genetically caused because of prejudice. Of course, her belief that the
disease is genetically caused is the result of her still very intense prejudice,
but her scientific evaluation ofthe evidence in favor of this belief must be
rigorously tested.
46 JUSTIFICATION AND PROPER BASING
Every objection to the claim is considered and refuted by the team, all
of whom, except for Ms. Prejudice, who plays devil's advocate, are
completely without prejudice. They all conclude that the scientific
evidence shows conclusively that the disease is genetically caused, as Ms.
Prejudice has believed all along. After the investigation, she knows that
the disease is genetically caused. She has the same evidence ... for
believing this as the other members of her team, and if they know that the
disease is genetically caused, so does she. But her belief is the product of
an improperly functioning system of racial prejudice. 5
Just as in the case of the gypsy lawyer, the racist scientist initially comes to a
belief on the basis of suspect motivations. According to Lehrer, these
motivations remain the basis for the belief even after learning of the evidence
that confirms the belief. Still, Lehrer claims, the racist scientist has knowledge
(and, by implication, justification) if her colleagues do. After all, they have
worked on the project together, knowing the intellectual character and prejudices
of each member of the team. It is obvious that, if asked who on the team knows
and who doesn't, they'd be perplexed at the idea of having to draw such a
distinction. They'd surely say that either everyone knows or no one does.
Since I will be focusing on the issue of the proper construal ofthe basing
relation, it will be useful to recast the discussion in terms of it. The lawyer and
the scientist each have adequate evidence for thinking what they do, but they do
not initially base their beliefs on that evidence. Later on, each comes to see that
the evidence confirms their belief, even though the evidence is not even a partial
cause of their beliefs. Since there is nonetheless a distinction between merely
having sufficient evidence for a belief and having that evidence justifY one's
believing of the claim in question, we can characterize the two cases as follows.
The lawyer and the scientist come to be doxastically justified in believing what
they were originally not doxastically justified in believing, and this claim
implies that they satisfY whatever basing relation between evidence and belief
is appropriate for capturing the distinction between doxastic justification and
propositional justification. Their beliefs are nonetheless not caused, not even
partially, by their awareness of the evidence; their deficient intellectual
characters result in still-true regrettable causal stories about their beliefs. Thus,
if we accept Lehrer's account, a causal account of the basing relation is
mistaken.
2. AUDI'S DEFENSE
investigation by the principals involved, and have attempted to save the heart of
the causal theory by giving up the claim that any actual causal relations are
required. Instead, they have attempted to salvage some causality in knowledge
by finding a true counterfactual involving a causal relation between evidence
and belief which is true and hence is compatible with these admissions, claiming
that this causality-embedded counterfactual is necessary for knowledge. 6 I have
argued against such theories elsewhere/ but will bypass the issues involved
here. For such emendations of a causal requirement on the basing relation do
not amount to a defense of a causal theory of basing. Instead, they agree with
Lehrer that a causal requirement is mistaken, so they pose no threat to the
conclusion Lehrer wishes to draw from his examples.
The other direction is to argue against the claim that the players in
Lehrer's examples come to have doxastically justified beliefs and hence against
the claim that they have knowledge. Robert Audi presents the only sustained
defense in the literature of this position, and his discussion focuses on the case
of the gypsy lawyer. He says,
Audi here worries about a certain sort of epistemic luck in believing a proposi-
tion. This sort of luck is present when one is indiscriminate with respect to the
truth. In the case of the lawyer, his reliance on the cards makes him
indiscriminate with respect to the claim that his client is innocent. As Audi
claims, "given his faith in the cards, he would have believed p even if it had been
false, indeed, even if, on the basis of the cards, it had not been rendered so much
as objectively likely (to any degree) to be true." This feature of the gypsy lawyer
case runs counter to what Audi sees as a reliable indicator connection es-
tablished by the presence of personal justification. That connection is that
48 JUSTIFICATION AND PROPER BASING
personal justification implies that if the proposition in question were not true,
the person in question would not believe it. Audi thus claims that because the
lawyer's belief is just a matter of this sort of epistemic luck, or "good fortune"
as he puts it, it is wrong to think that the lawyer really knows, or justifiably
believes, that his client is innocent of the eighth murder.
Audi seems to recognize some weakness in this defense of the causal
requirement, however. Consider his formulation of the reliable indicator
connection:
number offake barns in a landscape in which we happen to focus on the one real
barn. Suppose, though, that they were only giddily considering the fun they'd
have doing so, but decide at the last minute not to play such ajoke on us. We'd
have knowledge in such a case, but be lucky to have it. Or suppose that they
decide to play the joke on us, but we serendipitously happen to take a different
road and happen to look at a different landscape with no fake barns on it.
Perhaps a black cat crossed the road, and our superstitions led us to turn to the
right rather than the left in order to avoid the horror of having the cat cross the
fork in the road that we traveled. Then we'd still have knowledge, but be lucky
to have it.
Audi might insist that these are not the kinds of luck that knowledge rules
out, whereas the kind he is speaking of is that kind. To defend such a position,
we would need an account of the distinction between these two kinds of luck,
and Audi recognizes that no such account will be forthcoming by appeal to the
reliable indicator connection (since it can only be said to hold in the most
common cases). In addition, progress on the dispute between Audi and Lehrer
is not going to be made by relying on some delineation of the various kinds of
luck that might infect belief. As I see it, the only way to determine what kind of
luck knowledge rules out is to forward an acceptable account of knowledge, and
then classify instances ofluck in terms of that account. That is, I see no grounds
for thinking that we could sort instances of luck into kinds apart from our
interest in the nature of knowledge, and have one of these kinds be just the right
kind to solve the Gettier problem. In the absence of the plausibility of this
approach to the Gettier problem, appeals to the concept of luck will end up as
unconvincing as relevant alternative responses to skepticism, which claim that
skepticism is false because one need only be immune from error in relevant
alternatives, among which are not found skeptical alternatives. For these
reasons, Audi's arguments based on the concepts ofluck and reliable indication
against Lehrer's examples must be determined to be unsuccessful.
his awareness of his inadequate motivations. The fact which must be explained
is this progress.
The explanation I will argue for is that the progress achieved is that sort
attaching to the performance of the action in question which, when it obtains,
justifies the performance of that action. Of course, this explanation cannot be
employed by the causal theory, because the reasoning process above does not
have any causal impact on the particular behavior with which we are concerned.
Call the reasoning process in question "R". The occurrence of R is compatible
with the following truths: (a) if Jim did not have his irrational desire, he would
no longer be running for Congress; (b) even if R had not occurred, Jim would
still be running for Congress; and (c) if R had occurred and Jim's irrational
desire disappeared, Jim would no longer run for Congress. Moreover, if we
were to find a way to control for the effect of Jim's irrational desire, R would
have no probabilistic effect on Jim's behavior. Surely it is reasonable to assume
that R will affect his overall behavior at some point, for that is exactly what the
reasoning process shows Jim's intentions to be; but it need not have any im-
mediate causal impact on his running for Congress. We might capture this point
by calling the reasoning process a meta-motivational or meta-causal reasoning
process. If the reasoning process is a motivational one, i.e., if it is even
minimally causally efficacious with respect to the action in question, it would
be quite confused; for it includes the recognition that Jim cannot, simply by wil-
ling it or thinking about it, alter his motivation at present. I shall assume,
though, that Jim is not so confused.
What needs to be explained, to repeat, is the progress Jim makes in the
above case. What I shall argue is that any of the explanations open to a causal
theory do not sufficiently explain this progress, and thus that the causal theory
is shown to be inadequate by cases involving meta-motivational reasoning
processes. In order to defend these claims, we need some understanding of the
components of the justificatory ideal so we will be able to ascertain the variety
of alternative explanations open to the causal theory. This ideal requires, first,
that all actions, beliefs, desires, intentions, etc., be fully justified, i.e., all actions
be justified ones to perform, and all mental states be justified states in which to
be. Further, it requires that one justifiably perform all of one's actions, and
justifiably be in all the relevant mental states in question. Finally, this ideal re-
quires that the person in question have certain character traits which I shall refer
to loosely using the notion of being fully rational. The first two requirements
of the ideal relate, first, to the actions and states themselves, and then to the
relation between the person and the actions and states. This final requirement
relates to the person alone: he/she must be fully, or ideally, rational as well. So,
we are looking for an explanation in one of these three areas for the progress that
Jim has made in reconciling his running with his inadequate motivations.
JONA THAN L. KV ANVIG 53
The causal theorist might attempt to account for the case by claiming that
the only difference is a future difference. So, for example, the causal theorist
might hold that, given R, it is now true that in the future, Jim will justifiably run
for Congress (perhaps when he comes up for re-election); whereas withoutR, he
will not. On this option, no present difference exists except for the fact that it
is now true that the future will be different than it otherwise would have been.
54 JUSTIFICATION AND PROPER BASING
behavior with his motivations. And yet his distinctive progress does not occur
until stage three. Hence, neither of these internal states can be the only
explanation of the difference in Jim's state before constructing R and after con-
structing R.
The final internal state to which a causal theorist might appeal is Jim's
belief that he should (continue to) run for Congress. A causal theorist might
claim that, before constructing R, this belief was not doxastically justified, for
it was sustained by Jim's desire to prove his critics wrong; after constructing R,
R comes (at least in part) to sustain that belief. Hence, the progress Jim makes
is to be accounted for by noting that, before constructing R, Jim may have had
ajustification for thinking that he should continue to run, but he did notjustifi-
ably believe that claim. After constructing R, he comes to justifiably believe
that he should continue the campaign.
The difficulty with this attempt is that there is no reason to think that R
must sustain in part the belief in question. The irrational desire to prove his
critics wrong may be responsible both for his behavior and for his belief that he
should continue the campaign. Jim may know this sad fact about himself, and
try to sever the causal connection between his desire and this belief in addition
to trying to sever the causal connection between his desire and his running for
Congress. Nonetheless, it is still intuitively obvious that Jim has made progress;
thus, appeal to the belief in question will not explain this progress.
Moreover, there is something unsatisfying about the general approach
suggested here, that Jim's progress can be explained by citing some additional
mental states that are justified. Mere numbers of unrelated beliefs, desires,
intentions, or other mental states would add to Jim's overall epistemic condition,
but would not explain the unique kind of progress Jim has made. Furthermore,
the mere addition of further justified mental states about the particular issue of
running for progress won't explain his progress either. For note that such
changes are bound to occur between stages one and two, for the simple reason
that new experience can be counted on to provide additional evidence for new
mental attitudes. Yet, the sense of progress that Jim makes between stages two
and three is simply not there in the transition from stage one, where he is
unaware of his inadequate motivations, to stage two, where he becomes aware
of them. With this new awareness comes a host of new mental attitudes about
himself and his situation, attitudes which we may presume to be justified. Some
progress in terms ofjustification or rationality has been achieved because of this
change, but it is not the distinctive kind achieved in the transition from stage two
to stage three. This fact suggests that it is not in virtue of addingj Llstified mental
states that explains the progress Jim has made.
So, it would seem, appeal to internal states cannot salvage the causal
theory from the case of the ill-motivated politician. Let us turn then to a
different area to see if it offers more hope for the causal theory, for if internal
56 JUSTIFICATION AND PROPER BASING
states cannot explain what needs to be explained, perhaps external features, i.e.,
that which counts as overt behavior by Jim, can.
These external features include the acts, omissions, and aspects of each
which characterize or might characterize Jim at present. One such external
feature which cannot be of any use to the causal theorist is the act of running for
Congress itself. That act had a justification for it before Jim discovered his
motivations for performing it, and presumably the act itself is also justified after
Jim constructs R. Further, the causal theorist cannot hold that Jim justifiably
performs the act either before or after constructing R, for in neither stage does
that which justifies the act causally sustain his so acting.
A causal theorist might claim, though, that the act in question comes to
have a greater justification, or perhaps a justification all things considered (of
which Jim is aware), after constructing R. The appeal to a greater justification,
though, does not explain the advance toward the justificatory ideal. For one can
acquire a greater justification for believing, e.g., that all ravens are black just by
seeing another black raven, without making any such advance. The intuitively
obvious point about Jim is that he has made that sort of progress, and since
greater justification can occur without such progress, citing it in this case does
not explain that progress. Nor does the appeal to justification all things
considered explain the progress. The only way for this appeal to explain the
progress would be for the act to fail to be justified, all things considered, before
Jim undergoes the reasoning process in question, for otherwise there would be
no difference (on these grounds) between the first and second stages. Yet, when
Jim was not aware of his poor motivations for running, he had ajustification, all
things considered, for his campaign; so if we are to accept this explanation, we
must hold that after finding out about his poor motivations, Jim loses this
justification, and then regains it after engaging in the reasoning process
described. This explanation is inadequate precisely because the process from
lack of awareness to reconciliation through the reasoning process involves
progress toward an ideal, whereas the proposed explanation leaves Jim at the
same level after the total process as before. Hence, this attempt fails to free the
causal theory of the counterexample of the ill-motivated politician.
Could the causal theorist claim that Jim has greater justification than he
had before for running for Congress? I think the answer is no. On the proposal
above, Jim loses justification for running when he becomes aware of his
inadequate motivations. This information defeats whatever justified his running
in the first place. The force of the reasoning process is, then, to override the
defeating information. When such overriding occurs, the force of the original
JONA THAN L. KV ANVIG 57
say that an omission of a prima facie justified act is justified requires the
justification of some other action which implies that the primafacie justification
for the omission is overridden.
Thus, the difference in levels of rationality before and after having
reconciled his bad motivations with continuing to run cannot be explained
merely by claiming that certain omissions are justified. That may be true, but
if it is, it is only true in virtue of Jim's justifiably doing something else. One
obvious option here is that it is justified in virtue of its understood relationship
to Jim's running for Congress in spite of his inadequate motivations, but that
explanation is not open to the causal theory. For it would first have to be
granted that Jim s running is itself personally justified since its impersonal
I
justification is no different from stage one to stage three. Without appeal to this
action and its justification, however, it is hard to see where to find an action
whose justification confers justification on the omission in question. So we
must conclude that this attempt at an explanation is only as good as some other
one as yet forthcoming.
The only remaining alternative is to claim that some of Jim's actions, or
some aspects of his action, are justified while some others are not. The causal
theorist might hold that Jim's running is not justified, though his attempting to
help his constituents is. Or he might claim that Jim's running is not justified
though his acting so as to alter his motivational structure is.
This approach does not work either. For the aspects of the action, or the
different acts (however one chooses to individuate actions), are inextricably
linked. Perhaps some of the elements are primafacie unjustified whereas others
are prima facie justified. Given the inextricable linkage that occurs, however,
none of the elements can achieve actualjustificatory status without all the others
achieving the same status. There may be possible circumstances in which the
linkage is dissolved so that, say, Jim's running is not justified and yet his
attempting to help his constituents is. But it would be quite regrettable if this
possibility were taken to imply that the justificatory status actually diverges. For
the only way to make sense of the transition from prima facie justification to
actual justification is with reference to the entire set of elements of the actual
circumstances--to say that a prima facie justified action is actually justified is
to say that there is nothing else in the actual circumstances that overrides the
primafacie justification. And to say that a prima facie unjustified action is not
on the whole justified is to say that no other action has a primafacie justification
strong enough to override the prima facie lack of justification in such a way as
to justify the second action. But that is equivalent to requiring that the jus-
tificatory status of the elements stand or fall together, given that they are all
parts of the same situation. Thus, the causal theorist cannot hold that Jim's
playing is not justified whereas his using his talents is, and hence we must look
JONA THAN L. KV ANVIG 59
The final option open to the causal theorist is to claim that, whereas no
progress is made in the areas considered above, progress is made in that Jim,
himself is more rational after constructing R than before. It is, on this option,
progress regarding the rationality of the person in question (rather than his
(present or future) mental states, acts or omissions) which explains Jim's
progress.
In order to evaluate this attempt to rescue a causal theory, we must
consider what it is for a person to be rational. When we claim that a person is
rational, there are two things we might be claiming: first, we may be saying
something about the collection of actual beliefs, actions, etc., of the person and
noting that the collection is constituted by a sufficiently high degree of rational
beliefs, actions, etc., to warrant calling the person a rational one; or,
alternatively, we may be claiming that a positive evaluation applies to the
character of that person. The first option is ruled out by the above discussion
of the internal and external features of the case ofthe ill-motivated politician,
for all the same points can be made about the presence of rational belief and
action as were made about the presence of justified belief. So let us concentrate
on the second option. On it, to say that a person is rational is to say something
about the way in which that person determines what to do and how to do it, what
to believe, what and how to change, or what and how not to change. In other
words, we are saying something about the dispositions of the person in question
to proceed rationally or justifiably in forming and holding beliefs, choosing
actions, etc.
To return to the case of Jim, our ill-motivated politician, the causal theory
is claiming that Jim has better dispositions with regard to his actions, beliefs,
desires, etc., after constructing R than he had before constructing R. There is
something to be said in favor of the view that, through the process of
self-discovery, Jim himself becomes more rational. Perhaps prior to learning of
his poor motivations, he was disposed to form beliefs and perform actions in line
with poor motivations; after becoming aware of his poor motivations, he is less
inclined to do so. After learning that his irrational desire can prompt his
behavior, perhaps he is less likely to form a belief or perform an action when all
it has in its favor is that it satisfies such an irrational desire.
This difference is not sufficient for the causal theorist. In the case of the
politician, we have three separate stages: the first stage is where he is unaware
of his poor motivations, the second is where he becomes aware of his poor
60 JUSTIFICA nON AND PROPER BASING
motivations, and the third stage involves his reconciliation of his poor
motivations with his continuing to play professionally. Any advance in the
rationality of Jim's character, as clarified in the last paragraph, occurs in the
second stage, for his awareness of his poor motivations and the role his irrational
desire can play in his behavior have already been perceived prior to the
reconciliation in question. The progress Jim makes for which we are seeking an
explanation, however, occurs most obviously at stage three. Hence, this
explanation fails to account for the data.
It may be thought that Jim adds an intention in the third stage, so that this
stage involves two distinctive elements: first, the reconciliation about the
particular bit of behavior under question and second, an intention to alter his
motivational structure if he can. It might then be claimed that the progress to be
explained is that the additional intention is justified, and given its presence Jim
is both better off now and will be better off in the future.
The intention in question could just as easily have been added during the
second stage, however, and if it were added at that point, it would still be the
case that Jim makes progress toward ideal rationality in the third stage. Thus,
the case does not depend on the additional intention at the third stage. Further,
even if the added intention were central to the third stage, that would not explain
the progress in question, for, as we saw earlier, appeals to desires, intentions,
and increased self-awareness on their own do not explain the distinctive advance
characteristic of stage three, as opposed to stage two, which involves increased
self-awareness and new mental attitudes which can be presumed to bejustified.
It appears, then, that the causal theorist has no resources whatsoever by which
to explain the progress Jim makes toward ideal rationality.
It is important to notice that none of my arguments against the causal view
here presupposes a deterministic theory of causation. These arguments work
just as well against the proposal that Jim's reasoning provides some degree of
causal support, however minute. It is just as possible for this reasoning to occur
and have no probabilistic causal impact on his behavior as it is for it to occur
without having a deterministic impact. A denial of this possibility could be
sustained only by arguing that reasons must always be causes, but such a claim
wreaks havoc with the idea that Jim's running failed initially to be personally
justified. For in stage one, Jim has reasons for what he is doing (he gives these
reasons to himself and to others when asked), and hence if reasons are always
causally active in some way, these reasons would have to be partial causes even
in stage one. Such a maneuver would make a defense of the causal view even
more difficult than it is on the assumption that reasons need not always be
causes, for if reasons are always causes, then the causal view has no interesting
story to tell to distinguish doxastic from propositional justification.
No matter how causal theorists wish to view such a problem, however, the
point remains that nothing argued here presumes a non-probabilistic account of
JONATHAN L. KVANVIG 61
causation. So there is nowhere for a causal theorist to turn for rescue from the
case of the ill-motivated politician.
8. CONCLUSION
The most natural account ofthis case is just this: before constructing R,
Jim did not possess the kind of rationality which implies that he is justifiably
running for Congress; but after constructing R, his running is rational in a way
which implies that it is perfectly justified. The advance Jim makes is to be
explained by his moving from performing a justified action to justifiably
performing that action, from performing an impersonally justified action to
performing a personally justified one.
Such a conclusion gives us an account of the nature of justification that
covers the variety of things to which it applies and one which fits well with
Lehrer's intuitions about the gypsy lawyer and the racist scientist. The only
relevant difference between the racist scientist and the ill-motivated politician,
in this view, is that the scientist has no remorse for her racism whereas the
politician has regrets for his motivational structure. In that way, the politician
is better offthan the scientist. Still, the relevant beliefs and actions are on par,
being justified by the reasons available to each person, contrary to the demands
of the causal account of basing.
One may worry here that in accepting the non-causal account of the case
given above, we have lost the obvious point that there is something lacking
about Jim and his relation to his Congressional campaign. That is not so. We
can readily grant that Jim has not achieved ideal rationality, even if it turns out
that the lack of rationality in his running for Congress is the last vestige of ir-
rationality or lack of justification remaining among any of his acts, beliefs,
intentions, desires, valuations, etc. In order to be ideally rational, Jim must have
a proper character; and having a proper character requires being disposed toward
rationality and justification in the arenas of action, belief, desire, intention,
valuation, etc. Resolving his inadequate motivations with his continued running
surely does not eliminate the character flaw he possesses, even if this event is
part of the process toward character perfection.
I close with one last word of speculation on the attractions of the causal
theory. Perhaps the more fundamental kinds of knowledge, such as perceptual
knowledge, are kinds for which meta-motivational reasoning processes such as
those discussed here are not possible. Perhaps, that is, perceptual beliefs rely
essentially for their justification on a causal element in perception, and no
circuitous route through a meta-motivational reasoning process could make up
for a faulty causal story about such beliefs. Given the central place that
perceptual knowledge plays in the construction and defense of theories of
62 JUSTIFlCA nON AND PROPER BASING
ENDNOTES
I. This essay is written in honor of Lehrer's luminary career in philosophy, out of deep respect for
his intellectual achievements and gratitude for his acquaintance and gracious assistance throughout
my career. It is a privilege to know him, and I dedicate this essay to a splendid human being and
brilliant philosopher.
2 Keith Lehrer, Knowledge, (Oxford: Oxford University Press. 1974), pp. 124-125; "Proper
Function versus Systematic Coherence," in Jonathan L. Kvanvig, ed., Warrant in Contemporary
Epistemology, (Lanham, Maryland: Rowman & Littlefield Publishers, Inc., 1996), pp. 25-46.
3 The common attitude is represented in discussion of the issue by John Pollock, who footnotes
his claim that the basing relation is in part a causal relation as follows: "Lehrer has argued against
this, but I do not find his counterexample persuasive," (Contemporary Theories of Knowledge,
(Totowa, NJ: Rowman and Littlefield, 1986), p. 81). He offers no argument, supplies no
discussion.
4 Keith Lehrer, Knowledge, pp. 124-125.
5 Keith Lehrer, 'Proper Function versus Systematic Coherence,' pp. 33-34.
6 See, for example, Marshall Swain, Reasons and Knowledge, (Ithaca: Cornell University Press,
1981), chapter 3.
7 See "Swain on the Basing Relation," Analysis, Vol. 45, No.3, June 1985, pp. 153-158. Lory
Lemke criticizes my arguments in "Kvanvig and Swain on the Basing Relation," Analysis, Vol. 46,
No.3, June 1986, pp. 138-144; I reply to his objections in "On Lemke's Defense of a Causal
Basing Requirement," Analysis, Vol. 47, No.3, June 1987, pp. 162-167.
8 Robert Audi, "The Causal Structure of Indirect Justification," Journal of Philosophy, 1983, p.
406.
9 ibid., p. 407.
10 For a developed defense of the counterfactual theory of causation, see David Lewis,
Countelfactuals, (Oxford: Basil Blackwell, 1973).
II See, e.g., Jaegwon Kim, "Causes and Counterfactuals," in Ernest Sosa, ed., Causation and
Conditionals, (Oxford: Oxford University Press, 1975), pp. 192-195.
Chapter 4
T odd Stewart
University ofArizona
I find the case a bit underdescribed by Lehrer. Two critical pieces of the
example are left unstated: (1) whether or not Mr. Raco, when asked about his
belief regarding the susceptibility of a particular minority to some disorder, lists
off the relevant scientific studies as his reasons, and (2) if Mr. Raco himself
takes these scientific studies to be his reasons of ifhe simply gives these reasons
to others to stop them from attacking his racism. I think that the case is meant
to be one in which Mr. Raco does in fact respond to challenges by citing the
scientific studies and in which he does sincerely take his reasons to be those
studies. Mr. Raco does not realize that the root of his belief is his racism. What
Lehrer intends is to put forward a case where the agent honestly takes his
reasons to be ones which are not the actual cause of the belief or acceptance?
The important thing to notice about the example is that Lehrer does
clearly claim that Mr. Raco knows and is completely justified in believing that
members of the race in question are susceptible to the disease. A brief sketch
of the relevant pieces of Lehrer's theory is warranted at this point. Knowledge
according to Lehrer is, roughly, completely justified true belief (acceptance).3
A belief is completely justified when it is both personally and verifically
justified. 4 Lehrer uses complete justification as a technical term, so it is
important that he explicitly claims that Mr. Raco is completely justified. This
means that Lehrer thinks that Mr. Raco is both personally and verifically
TODD STEW ART 65
From this, we can infer that since Lehrer is a coherence theorist and further
thinks that any correct theory excludes causal factors as irrelevant to whether or
not a belief is justified, Lehrer's own theory must exclude these causal factors,
since otherwise Lehrer's own theory would (a) not be a coherence theory, and
(b) be incorrect. This is such an obvious inference that I suspect that
Koppelberg would endorse it. Again he never actually addresses the issue of
whether or not Mr. Raco is justified according to Lehrer's theory since he is
more interested in showing that Mr. Raco is not justified according to the
various notions of justification that Koppelberg thinks to be correct. 6
Koppelberg ends his paper by making similar general remarks about
causation and coherence theories. He writes, "A genuine coherence theory
claims that epistemic justification consists in adequately specified inferential
relations among belief contents--causal relations among belief states do not
matter.,,7 Again, the inference to the claim that Lehrer's theory, insofar as it is
a coherence theory, will entail that a belief may be justified even if it is not
appropriately related to its justifying reasons is obvious. So, I think it fair to
claim that Koppelberg would agree that Lehrer's theory is committed to the view
that appropriate causation is not necessary for a belief's being justified. Of
course, Koppelberg would use this to claim that Lehrer's theory is false, but this
is not important at present.
Interestingly enough, the arguments later in this paper will show that
Koppelberg (and perhaps Lehrer as well if he sees coherentism as being
logically incompatible with giving a role to causation) is simply incorrect to
suppose that coherence theories must by their nature exclude the causal origin
66 KNOWLEDGE AND CAUSATION
Mr. Raco can neutralize this competitor by claiming something like the
following:
Proposition n can neutralize c, assuming that c does in fact compete with the
racist belief, because the conjunction of c and n does not compete with p given
Mr. Raco's beliefs and it is at least as reasonable for Mr. Raco to accept this
conjunction as to accept c alone given his acceptance system.13 For, we can
stipulate that Mr. Raco does in fact accept various things that do appropriately
support n. Hence, c is neutralized and Mr. Raco is personally justified in his
belief despite the evidentially-illicit causation of his belief.
So far, we have seen that Lehrer's theory does entail that Mr. Raco is
personally justified. But, is Mr. Raco verifically justified? Unfortunately, I
think that Lehrer's theory as it stands does not deliver any verdict on this issue.
Further, this failure can be traced to the fact that Lehrer's theory actually takes
no stand at all on whether appropriate causation by reasons is necessary for a
belief to be completely justified or an instance of knowledge. Here we should
separate Lehrer from his theory; clearly, Lehrer thinks that causation is not
required for justification. But, surprisingly, Lehrer's theory as it stands does not
commit itself on this question.
Let us now turn to the argument for this claim. An agent's verific
justification system is what remains of her personal j ustification system but with
68 KNOWLEDGE AND CAUSA nON
all false beliefs deleted. If belief in a given proposition can be justified on the
basis of what remains (and these propositions can all be verifically justified as
well, and so on until a special kind of circle emerges), then that belief is
verifically justified. Thus, if we wanted to show that Mr. Raco is not verifically
justified in his racist belief, we would need to locate false beliefs that will be
used in support of the belief in question (either directly or in the handling of
competitors suggested by the skeptic). And, this is the heart of the problem:
Lehrer needs to explicitly claim that certain propositions are true or false in
order for his theory to entail that Mr. Raco has or lacks verific justification. But,
Lehrer's theory as it stands simply does not do this. Because of this, it is
impossible to say whether Mr. Raco's racist belief is verifically justified
according to the theory.
To see this, suppose that the skeptic makes the following claim during the
ultra-justification game: 14
c. Mr. Raco, the reasons you give for your belief that the
minority suffers from a disease are good reasons but they
are not your reasons because they have no causal influence
on your belief. As this is the case, despite the fact that they
are good reasons, since they are not your reasons, they do
not justify your bel ief.
reasons while still giving what is a good reason. Causation is actually part of the
content of certain reasons claims, and hence relevant to their truth or falsity. So,
when Wendy claims that she accepts a certain mathematical proposition on the
basis of Mr. Mathright's testimony, a necessary condition for the truth of her
claim is that Mr. Mathright's testimony itself or her belief that there was such
testimony was in some part causally responsible for producing or sustaining the
belief in the mathematical proposition. Because this causal component is false
in Wendy's case (and further we may assume that Wendy believes it to be false
as well - this must be added so that Wendy counts as actually having lied),
Wendy has lied about her reasons for accepting the mathematical proposition. I?
This, I hope, helps to make sense of c above; causation can and does enter the
picture where one's own reasons are concerned. The skeptic then claims that
only one's own reasons are relevant to justification. This is the skeptic's
objection to Mr. Raco.
So, what can Mr. Raco say in response to c? He could try to beat this
competitor by saying the following:
But, here it matters very much whether b is true or false. One can only
beat or neutralize a competitor in the ultra-justification game with a proposition
if that proposition is true. Otherwise, the skeptic is allowed to delete the belief
and make any other necessary adjustments to the agent's doxastic system in light
of this deletion. So, if b is true, and Mr. Raco responds to c with b, then b beats
c and Mr. Raco is verifically as well as personally justified in his belief
(assuming that b is in fact both personally and verifically justified as well).
Hence, he knows thatthe minority suffers from some peculiar and otherwise rare
disease. However, if b is false, then the skeptic can say this, excluding b from
use in the ultra-justification game. This means that b would not beat c, and
therefore Mr. Raco would not have knowledge according to Lehrer's theory as
it is not the case that all competitors have been beaten or neutralized. 19 Thus, we
can appreciate the critical importance of the truth or falsity of b in Lehrer's
theory.
70 KNOWLEDGE AND CAUSA nON
The problem is that Lehrer's theory does not actually tell us whether b is
true or false. The core of Lehrer's theory-all of Lehrer's definitions, formal
principles, etC.-is compatible with conflicting judgements about Mr. Raco; if
b is true, then Mr. Raco is verifically justified in his racist belief, but if b is false
then Mr. Raco is not verifically justified. One way of better seeing this is by
noticing that one could adjust Lehrer's theory to require appropriate causation
merely by understanding all reasons claims as about one's own reasons. Mr.
Raco's own reasons are not the scientific studies but rather his racism (if we
count this as a reason at all). Therefore, when Mr. Raco replies to the skeptic
with b and the skeptic rebuts him and deletes the proposition from Mr. Raco's
acceptance system, Mr. Raco is left without good reasons for his racist belief.
Thus, this belief will fail to be verifically justified and will not be an instance
of knowledge. To secure this result, however, no changes were made anywhere
to Lehrer's stated theory of knowledge and justification. No definitions were
adjusted, nor were any principles rewritten. Even the spirit of Lehrer's theory
is preserved here; only our conception of how we are related to a fact, as
encoded in the statement of our reasons for acceptance, and truth and falsity (as
the externalist bridge between personal justification and knowledge) are relevant
to justification. 20 This in itself helps make clear that Lehrer's theory simply
does not entail a verdict except in light of the truth or falsity of various
propositions which the theory itself does not entail.
The deeper significance of this sort of concern is that Lehrer's theory is
incomplete as it stands. In particular, it will only deliver verdicts about certain
kinds of cases when it is supplemented by specific claims as to the truth or
falsity of certain propositions that might be used to beat or neutralize various
skeptical objections. For, an agent can be personally justified in accepting that
a proposition beats or neutralizes a competitor or objection and yet not be
verifically justified in believing this. This reveals a more troubling problem for
Lehrer's theory: all of the definitions and principles do little to actuaIly pin
down exactly when an agent is verifically justified. These principles wiIl need
to be supplemented with various contentful claims about what is true and false
before it will be possible to derive results about verific justification. 21
This leaves us with the question as to whether it matters that Lehrer's
theory is incomplete. On the one hand, the incompleteness of the theory means
that it may be adjusted in all sorts of ways without changing the existing theory
itself. So, as I argued above, Lehrer's theory can be made to account for the
intuition that Mr. Raco's racist beliefis not an instance of knowledge by adding
a claim about the falsity of a certain proposition that will be employed by the
agent when she attempts to neutralize or beat the skeptic's objections in the
ultra-justification game. And, it might be considered a considerable virtue of a
theory that it remains silent about, yet compatible with, all sorts of contentious
claims about the relationship of causation to justification, etc. At the very least,
TODD STEWART 71
the theory is weaker than it perhaps appears, but hence more flexible because of
this.
In some sense, this is as it should be since Lehrer's theory is fallibilist and
allows that an agent can be personally justified and yet lack knowledge. But, it
also makes clear that the externalist components of Lehrer's theory- namely
the creation of an ultra-system by the deletion of all false beliefs-play such an
important role that we can only assess knowledge claims according to Lehrer's
theory if we have an indication of exactly what propositions are true and false
and hence which propositions an agent may use to neutralize or beat objections,
remembering that all false propositions will be deleted by the skeptic in the
ultra-justification game. Lehrer's treatment of the Mr. Raco case makes clear
what Lehrer thinks, namely that knowledge is possible without causation by
justifying reasons. But, Lehrer would need to explicitly add claims about the
truth or falsity of various kinds of propositions likely to be employed to meet the
skeptic's objections in the ultra-justification game to his theory for it to entail
this result.
On the other hand, this incompleteness makes Lehrer's theory very
difficult to assess, and might even make the theory nearly unfalsifiable. Testing
our intuitions against the theory as it stands may be impossible in some cases,
as with Mr. Raco. Critics will have to be extremely careful to make sure that
Lehrer's theory really does deliver certain results before they can present
compelling counterexamples. Because of this, it might be best to focus on the
general structural features of the theory and refrain from the usual sort of
counterexamples until the theory is modified so as to have clear consequences
about various kinds of cases, but this severely limits our ability to determine
whether or not Lehrer's theory is a good one. The flexibility of the theory is
here purchased at the cost of our ability to test the theory in various respects.
I have argued that Keith Lehrer's theory of knowledge does not in fact
have some of the consequences that Lehrer and others have taken it to have. In
particular, Lehrer's theory as it stands remains silent about the relationship of
causation to justification. Indeed, it seems possible to easily accommodate the
view that causation is necessary for justification by simply understanding
reasons claims as always being partly causal. Causation would enter the picture
here because it is explicitly represented in reasons claims, and improper
causation can make such reasons claims false. Of course, Lehrer's theory is also
compatible with the view that justifying reasons need not be causal. Whether
the incompleteness of Lehrer's theory is a virtue or a vice, and both the
flexibility and difficulty of assessment this incompleteness reveals, I leave to the
reader to decide. 22
72 KNOWLEDGE AND CAUSATION
ENDNOTES
I Lehrer (1990), pp. 169-70. The Mr. Raco case also appears in Lehrer's more recent (2000)
(, In some sense, Koppelberg never takes Lehrer's theory of justification very seriously at all.
Koppelberg never assesses Lehrer's theory but rather simply argues that Mr. Raco is not justified
according to various other conceptions of justification. Insofar as there is any direct contact
between Koppelberg and Lehrer, it is that Koppelberg would deny that Mr. Raco has knowledge
of the proposition belief in which is sustained or caused by prejudice. See fn. 21, p. 460.
7 Koppelberg, p. 459.
8 One might argue that what this shows is that Lehrer's theory is not really a coherence theory after
all. This taxonomic maneuver strikes me as unsatisfying, though. Lehrer's theory strikes me as
an exemplar of a certain type of coherence theory (with certain externalist components of the sort
needed to solve the Gettier problem), and it would do injustice to our taxonomy to classify it
otherwise. In particular, I think that we will always have to add non-coherence factors - most
obviously truth - to our account of knowledge, and so as long as one's basic theory of justification
is coherentist, one should be classified as a coherentist. This does seem true of Lehrer's theory.
9 Lehrer, p. 126.
10 For convenience, I have dropped reference to a specific time t t!'om the definitions.
13 It is not actually clear to me that c as I state it does compete with Mr. Raco's racist belief. But,
if it does not, the skeptic can presumably work c into something that is a competitor that can be
neutralized (or perhaps beaten) with something approximately like n. For example, the skeptic
might claim that it unreasonable to hold a belief if it is not causally responsive to evidence. We
could then understand n as rebutting this further claim which would act as a competitor to the
racist belief.
14 In the ultra-justification game, the skeptic is allowed to simply delete any false propositions (and
all propositions that logically imply this proposition) used in the defense of some other
proposition, disqualifying such stricken propositions from use as justifiers. This is core of the
difference between the justification and ultra-justification games. See Lehrer, pp. 141-44.
15 Read the "because" in this sentence as both causal and justificatory.
16 I wish to remain neutral on how it is that testimony justifies what the content of what the testifier
says (viz., whether testimony proceeds on the basis of perception and induction, or whether it has
some sort of intrinsic a priori authority as a basic source of belief, etc.). I use testimony here only
because it helped make for a clear example.
17 I am being intentionally vague here about exactly what is to count as a reason. Is it Wendy's
belief that Mr. Mathright testified, or Mr. Mathright's testimony, or something else entirely? I do
not think that my argument depends on any specific construal of a reason, so I leave it to the reader
to supply their own preferred notion. What I am suggesting, however, is that one could very easily
TODD STEWART 73
build an explicit causal component into the truth conditions for at least some of these conceptions
ofa reason.
18 I think we need to treat this proposition as beating the skeptic's objection rather than neutralizing
it because one cannot consistently believe both c and b. Second, I will admit that I am a bit
concerned that it is far from likely that any average person, even one with a bit of philosophical
savvy, would respond to the skeptic's challenge in this way. I fear that perhaps Lehrer's account
may exclude many intuitive cases of knowledge. But, this is an argument for another time.
19 Interestingly enough, one might try to explain differences in intuition (which a small poll has
revealed do seem to diverge) about the Mr. Raco case by focusing on one's attitude towards b. If
one thinks that b is true, then one will likely have the intuition that Mr. Raco does have knowledge.
On the other hand, if one sees b as false, then one will be inclined to deny that Mr. Raco has
knowledge.
20 See Lehrer (1990), p. 153.
21 I suspect that this problem is not limited to various causal principles of justification. For
example, the skeptic might challenge certain people on the grounds that their justifications are too
short. Instead of presenting any content-dependent reasons for accepting thatp, suppose the agent
simply argues that she is trustworthy in what she accepts and hence, given that she accepts that p,
p is likely to be true (this argument will work for any proposition which the agent accepts).
Suppose that the skeptic objects that not everything that the agent believes can be justified with
this universal argument. Suppose that the agent replies by simply reapplying the simple argument
again: since the agent accepts that his justifications are long enough to justify his beliefs, and what
he accepts he accepts in a trustworthy manner, it is probably true that his arguments are long
enough to justify his belief. Is this an effective way of beating the skeptic's objection? This will
depend where verific justification is concerned on whether or not the agent's reply is in fact true.
But, does Lehrer's theory as it stands tell us whether or not we can use this simple universal
argument to completely justify all our beliefs? This is not clear to me.
22. Thanks to Keith Lehrer for his support and comments on drafts of this paper. Thanks also to
Erik Olsson and to Paul Thorn for their extremely helpful suggestions.
74 KNOWLEDGE AND CAUSATION
REFERENCES
Koppelberg, Dirk. "Justification and Causation." Erkenntnis 50, 447-462 (1990).
Lehrer, Keith. Theory of Knowledge. Boulder, co: Westview Press, 1990.
Lehrer, Keith. Theory of Knowledge, Second Edition. Boulder, CO: Westview Press, 2000.
Chapter 5
Volker Halbach
University of Constance
In this paper, I will scrutinize a family of arguments which are supposed to show
that consistency is not accessible in an epistemically relevant sense. If these ar-
guments were sound, then consistency would be on a par with external facts:
neither are directly accessible to us. External factors are not directly accessible
because they need to be mediated in some way. Consistency would be inaccessi-
ble, according to these arguments, because proving or determining consistency
in the relevant cases exceeds our intellectual capabilities. Thus, if consistency
were actually inaccessible, then consistency could hardly playa role in an inter-
nalist account of epistemology. Traditionally (for instance, in Schlick's account
(1934)), however, consistency has been seen as a main ingredient of coherence.
Modern epistemologists, like Bonjour (1985), have also used consistency as a
criterion of coherence, and they have even used consistency in order to define
coherence.
The arguments against the accessibility of consistency jeopardizes this
central role of consistency in any internalist account of epistemic justification. I
Thus, epistemic justification cannot imply the consistency of the belief system.
If consistency is not directly accessible, then it cannot be used as a criterion for
consistency on an internalist account.
Keith Lehrer's epistemology is not internalist. Internalist justification,
however, is an important ingredient of Lehrers's account since internalist justi-
fication is a necessary condition for 'full' justification. Because of the Gettier
problem, external factors enter the definition of one form of epistemic justifica-
tion on Lehrer's account. Lehrer (1990b), for instance, distinguishes between
personal and undefeated justification. External factors form part of the defi-
nition of the latter. Undefeated justification is then used in the definition of
75
E.J. Olsson (ed.), The Epistemology of Keith Lehrer, 75-87.
© 2003 Kluwer Academic Publishers.
76 CAN WE GRASP CONSISTENCY?
Thus justification implies the consistency of the acceptance system. 3 The quot-
ed passage does not imply that we have a direct grasp of the consistency of our
acceptance system, and Lehrer avoided, as far as I can see, such strong claims.
The reason for avoiding such a commitment may be, in part, the argument I will
discuss in this paper.
Implicitly, however, consistency is also central for Lehrer's theory, for
an inconsistent theory logically refutes any given belief. One would like to
conclude that an inconsistent belief system therefore also defeats any belief. It
seems hard to see how Lehrer can avoid that consequence, although he must
deny this when claiming that beliefs can be justified on the basis of an incon-
sistent belief system. Therefore, one has to be, at least, very careful about in-
consistency. The reader is sent to the above mentioned papers (Lehrer 1990a)
and (Lehrer 1999) for a discussion of this topic. At any rate, it seems interesting
to see to what extent consistency can be used as a criterion for justification and
coherence in a framework like Lehrer's.
In the following, I will discuss the arguments against the accessibility of
consistency in a general setting in order to see whether consistency can be used
as a criterion for epistemic justification. I will not explore what consequences
are implied for particular theories like Lehrer's or Bonjour's.
The point of this paper is modest: I will show that some of these argu-
ments have been accepted rashly and that their scope is more limited than is
often assumed. I hope that this is interesting enough. For these arguments have
had a major impact, and they have influenced many epistemologists.
Thus, I will examine whether the arguments actually demonstrate that
consistency is not accessible and, in this respect, is much like an external factor.
The problem is vague, of course. In the first place, it has to be specified whether
one is dealing with the consistency of arbitrary sets of beliefs or only with the
consistency of one or more specific sets like one's belief system, (i.e., the set
of beliefs one accepts). Moreover, I have to specify for whom consistency is or
VOLKER HALBACH 77
1. A SIMPLE ARGUMENT
The following is the most radical argument I will consider. According to this
argument, consistency cannot be checked by human beings for more than a very
limited number of beliefs. Thus, not even reference to the complexity of truth
tables, or similar metalogical aspects, are envoked.
In his presentation ofChemiak's (1986) account, Hooker (1994) writes:
It is not simply that human beings have difficulty in determining the con-
sistency of large sets of sentences. It is simply beyond the powers of any
possible computational device to determine the consistency of a large set
of sentences.
I agree that human beings are not able to determine the consistency of arbitrary
large sets of sentences for their consistency. But in order to attain justification
a subject only has to prove that its own belief system is consistent. Thus, Korn-
blith seems to think that humans are not able to establish the consistency of a
set of sentences that is as large as our belief systems.
However, it is not in general impossible for human beings to establish the
consistency of certain large and even infinite sets. Kornblith's claim taken as
a universal statement is simply false. It is not generally beyond the powers of
any possible computational device to determine the consistency of large sets of
sentences.
Logicians prove consistency of infinite sets of axioms all the time. Peano
Arithmetic PA, for instance, is given by infinitely many axioms and cannot be
axiomatized by a finite set of axioms. Nevertheless, it does not exceed the ca-
pabilities of the average logician to show the consistency of PA by providing a
model, or by proving its consistency in Gentzen's style (cut elimination). Fur-
ther examples can be found in abundance. In general, a high cardinality of an
axiom set does not prevent logicians from providing consistency proofs.
In general, consistency proofs do not require that we juggle all sentences
of a set simultaneously in consciousness; this is not necessary for carrying out
consistency proofs. In a consistency argument, we do not have to use the sen-
tences of the set in question, which is impossible; rather we only have to talk
about them. The latter is no problem; we can say that all sentences in a set have
a certain property. In particular, we can ascribe certain formal properties to all
sentences in a set and finally arrive at the conclusion that a contradiction cannot
be derived from these sentences.
Therefore, the argument does not show, by any means, that we cannot
determine the consistency of sets with more than ten sentences.
There are, at least, three results in metamathematics that have been em-
ployed in order to prove the inaccessibility of consistency. The arguments fall
into three classes:
• Complexity of truth tables. This point concerns consistency only with re-
spect to propositional logic. Propositional logic is decidable; a straight-
forward method for checking it is provided by the truth table method,
which is taught in almost any introductory course to logic. There are 2n
many different assignments of truth values to n propositional variables.
Thus, a propositional sentence with, for instance, 12 propositional vari-
ables yields a truth table with 212 = 4096 many lines. It is easily seen
that even very simple sentences are hard to check for consistency with
the truth table method and practically cannot be checked for consistency
by the truth table method.
How conclusive are the three arguments? Do they show that consistency
is like an external factor in being out of our control and in being (in the relevant
cases) inaccessible to us?
counts of belief revision deal only with propositional logic. Thus "consistency"
is to be understood in terms of propositional logic. Since propositional logic is
decidable, the check for consistency can be carried out in principle (although it
may require fairly complex computations).
But ifthis picture of how new beliefs are incorporated in our belief system
is applied to predicate logic, a fundamental difficulty arises. Epistemic agents
do not know how to obey the command "check first whether the new belief A
is consistent with the old stock S of beliefs!" According to Church's Theorem,
there is no general recursive method to carry out this check.
So there might be a general problem for the applicability of formal ac-
counts of belief revision. But I do not take it as a conclusive argument against
the claim that consistency is not accessible in a sense relevant for epistemology.
It can be granted that consistency of very complex belief systems can be shown,
and that the complexity of truth tables and Church's Theorem do not contradict
this observation. In fact, consistency proofs of even very strong systems can
be very simple in terms of length of proofs. Thus, I conclude that the above
arguments cannot disprove conclusively that consistency is accessible.
Here is an interesting twist. What tools do we have at our disposal in
order to establish the consistency of our belief system?
Obviously, we must carry out the consistency proof of our belief system
on the basis of what we believe, that is, on the basis of our belief system. For we
can use in our reasoning only assumptions we already believe. The consistency
of the belief system must be established within that very belief system.
Godel's Second Incompleteness Theorem seems to tell us that such a
consistency proof cannot work, according to the slogan "No system can prove
its own consistency." So, proving consistency of one's belief system does not
produce problems because the proofs are too complex or because there is no
uniform procedure to check consistency, but rather because the resources for
the proof are limited in the case we are interested in, namely where we try to
prove consistency of a belief system within that very same belief system. In all
interesting cases, the consistency of a belief system can only be established in a
more comprehensive belief system, or so it seems.
In the discussion of the argument based on Church's Theorem, I claimed
that consistency proofs can be carried out although we lack a decision procedure
for consistency. These consistency proofs, however, require means that go be-
yond the resources available in the system for which consistency is shown. The
consistency of Peano Arithmetic, for instance, can easily be shown within set
84 CAN WE GRASP CONSISTENCY?
those who claim that consistency is not attainable because of Gbdel 's theorem;
they have to provide versions of the theorem that yield their claim that consis-
tency is not within our reach.
3. CONCLUSION
I have shown that the arguments that are purported to prove that consistency is as
little accessible to us as purely external facts are not conclusive. Of course, con-
sistency is not accessible, if 'accessible' is understood in a very strong way, for
we do not have general procedures for determining whether an arbitrary given
set of sentences is consistent. But this kind of accessibility is not required for
most epistemological purposes. For those purposes, we only need to establish
the consistency of our own belief system or the consistency of our belief system
with another belief.
Although I have shown that the discussed arguments are not conclusive, I
do not claim that consistency actually is accessible in a relevant sense. Church's
Theorem and Gbdel's Second Theorem will make it hard to give a general ac-
count of how justification is obtained if consistency is to be used as a partial
criterion for justification. The arena is open again.
Acknowledgements. The paper was presented to the Belgian Society for Logic
and Philosophy of Science, Brussels, Belgium, on 23 January 1999. The helpful
suggestions of the audience are gratefully acknowledged. The work on this
paper was carried out while the author was a member of the research group
Logic in Philosophy financed by the Deutsche Forschungsgemeinschaft. I thank
David McCarty and the members of the research group and, in particular, Erik
Olsson for useful suggestions and discussions.
ENDNOTES
1 Moreover, these arguments have also been employed against extemalist coherence theories. See
systems.
3 According to this quote, an inconsistent acceptance system does not justify any belief. In the
more recent paper (1999), Lehrer explicitly maintains on p. 252 that beliefs can be justified on
the basis of inconsistent belief systems. I quote another passage where he seems to suggest that
inconsistency might not be fatal to our efforts to obtain justified beliefs. He writes in (l990a,
p. 166):
Hence one who values consistency as an end in itself should recognize that a person with
equally pure intellectual concerns may reasonably accept an inconsistent set of sentences.
What he accepts may be suited to the objectives of obtaining new information and eschew-
ing error.
86 CAN WE GRASP CONSISTENCY?
4 At this place it makes a difference how belief is defined. If, linguistically speaking, the limited
performance of a subject is neglected and only his or her competence is considered, then a person
may believe even very long sentences and in fact sentences of arbitrary length. However, if there
is no limit imposed on the perfonnance of a subject, then there is no limit on the complexity of
truth tables that can be calculated.
5 There is also another system called Robinson's arithmetic which is usually designated by R.
6 Just how valuable the consistency proofs for Q are is a different problem. In the light of Go del's
Second Incompleteness Theorem, the value of these consistency proofs may be doubted. I will
return to this point below.
VOLKER HALBACH 87
REFERENCES
Bonjour, Laurence. 1985. The Structure C?f Empirical Knowledge. Cambridge, Massachusetts:
Harvard University Press.
Chemiak, Christopher. 1986. Minimal Rationality. Cambridge, Massachusetts: MIT Press.
Feferman, Solomon. 1960. "Arithmetization of metamathematics in a general setting." Funda-
menta Mathematicae XLIX:35-91.
Harman, Gilbert. 1973. Thought. Princeton: Princeton University Press.
Hooker, Cliff A. 1994. "Idealization, Naturalism, and Rationality: Some Lessons from Minimal
Rationality." Synthese 99:181-231.
Komblith, Hilary. 1989. "The Unattainability of Goherence." In The Current State C?fthe Coher-
ence Theory, edited by John Bender, 207-214. Dordrecht: Kluwer.
Lehrer, Keith. 1990a. "Reason and Consistency." In Metamind, 148-166. Oxford: Clarendon
Press .
.- - . 1990b. Theory of Knowledge. Boulder: Westview Press.
- - - . 1991. "Reply to Mylan Engel." Grazer Philosophische Studien 40:131-133.
- - - . 1999. "Justification, Coherence and Knowledge." Erkenntnis 50:243-258.
Schlick, Moritz. 1934. "Uber das Fundament der Erkenntnis." Erkenntnis 4:79-99.
COHERENCE AND PERSONAL JUSTIFICATION
Chapter 6
Glenn Ross
Franklin and Marshall College
1. ADILEMMA
In a fair lottery with very many tickets, and only one winning ticket, I
should be very confident that my ticket would lose. My reasons for being highly
confident would seem to make it epistemically permissible for me to accept that
91
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 91-107.
© 2003 Kluwer Academic Publishers.
92 THE LOTTERY PARADOX
my ticket will lose. Yet, since my reasons for accepting that my ticket will lose
are qualitatively identical to my reasons for accepting that any other ticket will
lose, I do not have any epistemic reason to prefer accepting that my ticket will
lose to accepting that some other ticket will lose. It seems epistemically wrong
and arbitrary to accept that which I recognize I have no better reason for
accepting than that which I choose not to accept. So, either I should not accept,
of any ticket, that it will lose, or I should accept, of each and every ticket, that
it wi\l lose. To refuse to accept of any ticket that it will lose in a very large
lottery seems immoderately skeptical. To accept that a given ticket will lose is
attractively less cautious. Yet, if it would be arbitrary and epistemically
unreasonable to accept that one ticket will lose and not to accept the same of the
other tickets, then one cannot accept of one ticket that it will lose without being
obliged to accept the same of all the rest. If I accept of each ticket that it will
lose, but also accept that some ticket will win, my acceptances will be self-
recognizably inconsistent. That seems just as bad. ,
Keith Lehrer 1 and many others (e.g., Mark Kaplan 2 , John Pollocko, Sharon
Ryan4, and Dana K. Nelkin 5) adopt the epistemically cautious resolution and
insist that considerations of consistency demand that one should uniformly
withhold judgment on each proposition that a particular lottery ticket will lose.
Richard Fole/ and Peter Klein7 opt for the much more credulous
recommendation that one can be recognizably inconsistent: accepting of each
ticket that it will lose while also accepting that some ticket will win. I propose
that we can slip through the horns by adopting a consistently credulous
approach 8 : one can rationally accept that one's ticket will lose, while not
accepting that of many of the others, even though one is no less confident that
they too willlose. 9
Both the skeptical and the inconsistency solutions presuppose a principle
of symmetry: that if! have equally good reason to accept of any ticket that it will
lose as I have to accept that any other ticket will lose, I should adopt the same
attitude uniformly. I should either accept of each ticket that it will lose or not
accept of any ticket that it will lose.
Naturally, statements beat their competition for reasonab Ie acceptance when they
are more reasonable to accept.
The point of the game is not to refute the skeptic, but to exhibit those elements
of one's acceptance system that justify the original claim.
Still, a statement need not beat all of its competitors to merit reasonable
acceptance. It is sufficient to neutralize the competition. The notion of
neutralization is defined thusly:
The reason that the claimant's response neutralizes the skeptical challenge is that
it is as reasonable to accept the conjunction that I am not now dreaming even
though people sometimes dream that they see snakes as it is to accept that people
sometimes dream that they see snakes. Though a conjunction is less probable
than its conjunct, the additional informational content of a conjunction can make
it as reasonable to accept as the conjunct inasmuch as it can have greater
epistemic utility.
Now we are in a position to consider how Lehrer utilizes his theory to
handle the lottery paradox:
The very last claim, however, demands an argument. Why should we think that
the skeptic's claim could not be neutralized? Could it not be that the conjunctive
claim that both the number one ticket and the number two ticket have not won,
while less probable than the claim that the number one ticket has not won, is at
least as reasonable in virtue of its additional informational content? It is not
obvious that the rate of increase in the utility, as we conjoin these lottery
GLENN ROSS 95
statements, will never be as great as the rate of decrease in the probability. So,
it is not obvious that the skeptic's objection cannot be neutralized.
Whether or not Lehrer's theory is adequate to the task of giving him the
result he seeks for resolving the lottery paradox, the fundamental question is
whether he is seeking the right result. Do considerations of consistency preclude
the rational acceptance of any lottery proposition? Let us turn to an argument
that withholding on such lottery statements is analogous to withholding on many
other statements about the future, statements that we have adequate inductive
reasons to accept.
The fact that my lottery ticket is very probably a loser constitutes a good
reason for accepting that it will lose as long as merely probabilistic grounds can
suffice for reasonable acceptance that my ticket is not going to win. Such
probabilistic reasons are adequate for reasonable acceptance if lottery situations
are analogous to situations in which one has merely probabilistic, but adequate,
grounds for various empirical acceptances. There do seem to be many lottery-
like empirical claims about that which is unobserved, where our evidence,
though statistical, seems perfectly adequate for reasonable acceptance. Ifso, then
by analogy, it can be reasonable to accept that one's lottery ticket will lose.
Jonathan Vogel has suggested that lottery-like empirical statements
abound. One is the proposition that a meteorite will not hit the place where I am
sitting right now. I have no better reason to accept this than I have to accept of
any other place of comparable size on the surface of the globe that a meteorite
will not hit there at some particular time. Moreover, ifI consider a long enough
span of time, I have high confidence that a meteorite will hit somewhere and
sometime. Yet, even with no better reason to deny that the meteorite will now hit
here than I have to deny that it will hit somewhere else at some other time, I
confidently assert and accept that a meteorite will not hit me here and now.
Likewise it is reasonable for me to accept that a satellite due to fall out of orbit
will not hit my house, though pieces of it are expected to hit somewhere. Further,
it is reasonable for me to accept that I will not be burned to death by a pyroclastic
flow of hot ash from Mount Rainier while I spend a few days at a conference in
Seattle, though I fully expect that eventually such events will occur around active
volcanoes. I can reasonably accept an earthquake will not kill me during my next
visit to Los Angeles, despite my having good reason to expect there will be such
earthquakes, and resultant fatalities, sometime or other. 13
It would seem immoderately cautious and skeptical to deny that I can
reasonably accept that a meteorite will not hit me, or that the satellite falling out
96 THE LOTTERY PARADOX
of orbit will not hit my house, or that I will not be killed by a volcano while in
Seattle or an earthquake while in Los Angeles.
The skepticism will be extreme if we endorse a restricted principle of
closure for reasonable acceptance. Suppose I am subject to epistemic criticism
for failing to accept what I recognize to follow, in a deductively immediate
manner, from that which I accept. Then, if it is unreasonable to accept that a
meteorite will not imminently crush me, it is unreasonable to accept anything
that entails this negative fact, including any fact about my future life. Yet, even
if we do not accept such a limited principle of closure, it still seems
immoderately skeptical and cautious to maintain that an epistemically scrupulous
individual should refrain from accepting that a meteorite will not hit here and
now, despite reasons that should make one highly confident that this proposition
is true.
Yet we should also not follow Foley and Klein and contend that it is
rational to accept of each ticket that it will lose, despite the recognition that these
acceptances are not logically compatible with accepting that one of the tickets
will win. We should resist this position, as Kaplan and Pollock have argued,
because if recognized inconsistency does not leave one vulnerable to epistemic
criticism, then there is no dialectical point in providing a reductio ad absurdum
of someone else's position. That reductios do have epistemic weight shows that
recognized inconsistency is an epistemological vice.
other hand, it is false that I would still accept that the Bulls had won even ifthey
had not. For, had the Bulls not won, I very well might have read an accurate
newspaper account of the game and consequently have accepted that they did not
win. 15
Yet, whether an accepted statement passes or fails DeRose's subjunctive
conditional test cannot make the difference between whether or not it is rational
to accept it. Consider a revised newspaper case proposed by DeRose, in which
one knows that each day there is one, and only one, copy of the newspaper in
which all of the sports scores are inverted. Using DeRose's criterion as a
requirement for rational acceptance on this case yields the consequence that one
can rationally accept that the Bulls won on the basis of the reported score in the
paper. Yet, one cannot rationally accept that one's paper is not the defective
copy. Nonetheless, one's grounds for accepting that the Bulls won is the score
that is reported in one's copy of the paper. It would seem that if one cannot
reasonably accept that one does not have a non-defective paper, and one
recognizes that fact, then one cannot reasonably accept that the Bulls won. Since
it is epistemically reasonable to accept in these circumstances that the Bulls won,
I conclude that DeRose's test, though possibly demarcating cases of knowledge
and assertibility, will not work as a criterion for demarcating cases of rational
acceptance.
Another unique feature of the lottery case that might be exploited to
provide a disanalogy with the lottery-like cases is the fact that in the lottery case
one knows that there is a winning ticket. Since one knows that there is a winning
ticket of the lottery, one knows that there is some objective probability of one's
winning. So one might contend that though it may be reasonable to accept that
one will probably lose, it is not reasonable to accept that one will lose. On the
other hand, despite the probability that not all parts of the satellite will burn up
before reaching the earth's surface, one cannot take it for granted that some parts
will reach the ground. 16 Since one does not know that any parts of the satellite
will hit the ground, one cannot be certain that there is any objective probability
of my house being hit. So, perhaps this difference could allow one to claim that
in the lottery case, one cannot rationally accept that one's ticket is a loser, while
in the case of the satellite, one can reasonably accept that one's house will not
be hit.
That this disanalogy is not adequate to the task at hand is apparent,
however, if we consider a revised lottery case. Suppose that one ticket among the
million or so tickets is designated the "No Winner" ticket and is assigned to no
one, so that if that ticket is selected, no ticket wins. If it is reasonable for you to
accept that the satellite will not hit your house because you cannot be sure that
any part of the satellite will crash anywhere, then it is similarly reasonable for me
to accept that my ticket is not a winner. For, one cannot take it for granted that
any ticket is a winner, and it is very unlikely that any particular ticket will win.
Yet, since one knows that one ticket will be selected, then according to this line,
98 THE LOTTERY PARADOX
it is still unreasonable for me to accept that my ticket was not selected. So, I can
reasonably accept that my ticket will lose, but not reasonably accept that it was
not selected. This cannot be right. For, I know, by an immediate inference, that
if my ticket is not the winner then it was not selected, since I know that it is not
the No Winner ticket. So, if it is reasonable to accept that my ticket will lose,
then it is reasonable to accept that my ticket was not selected. It follows, that
whether or not I know that some lottery ticket will lose cannot matter in the
reasonability of accepting whether a particular ticket will lose.
Good analogies cut both ways. If you are convinced you do not have good
reason to accept that your lottery ticket is a loser, you could use the analogy to
lottery-like cases to argue that you similarly have no good reason to accept that
your house will fail to be hit by the satellite. Dana K Nelkin, in seeking to find
a difference between lottery situations and cases in which inductive reasoning
based on perception is justified, argues that the feature of lottery inferences that
makes them epistemically unjustified lies precisely in the purely statistical nature
of the evidence. Thus, on Nelkin's account, there is no difference between the
lottery situations and the lottery-like situations, so long as one's evidence in both
cases is purely statistical. Nelkin argues that statistical inferences are only
acceptable when grounded in presuppositions of a causal explanation of the
statistical evidence.
The intuitive costs of this account are evident when Nelkin considers a
case proposed by Gilbert Harman. I? If Mary will be in Trenton only if she wins
the lottery, and in New York otherwise, then if one cannot rationally accept that
Mary will not win the lottery, it would seem that one could not rationally accept
that Mary will be in New York. Nelkin responds by conceding this implausible
consequence, and responds that such situations are relatively rare.
Nelkin's position is immoderately skeptical. If it is not rationally
permissible to accept that your house will not be hit by a satellite or a meteorite,
on purely statistical grounds, then if a limited closure of recognized immediate
inferences holds, then it would seem to be correspondingly impermissible to
accept that which presupposes that we will not be hit by a meteorite, and then it
is difficult to imagine much of anything in our future about which one could
form a reasonable expectation.
Moreover, even if limited closure fails, apparently justified statistical
inferences are not all that rare, and not confined to the future. If you were to tell
me that you randomly flipped twenty heads in a row yesterday. I would infer that
you were flipping for a long time. I do so on the basis of statistical
considerations, and not because I accept that there is a causal connection between
flipping a coin for a long time and getting twenty heads in a row. It would seem
that such purely statistical inferences are both common and reasonable.
Using a generalization of Keith Lehrer's Racehorse Paradox,18 John
Pollock has provided a technical argument to show how one can turn any
statistical syllogistic inference from high probability into a lottery-like form. 19
GLENN ROSS 99
In the lottery, this throws us back onto our two horns: either we do not accept of
any particular ticket that it will lose, or we accept of all tickets that they will lose
despite our recognizing the inconsistency with our accepting that some ticket will
win.
Gilbert Harman considers a via media:
... To say one can infer this of any ticket is not to say one can infer
it of all. Given that one has inferred ticket number 1 will not win,
then one must suppose the odds against ticket number 2 are no
longer 999,999 to 1, but only 999,998 to 1. And after one infers
100 THE LOTTERY PARADOX
ticket number 2 won't win, one must change the odds on ticket
number 3 to 999,997 to 1, and so on. If one could get to ticket
number 999,999, one would have to suppose the odds were even,
1 to 1, so at that point the hypothesis that this ticket will not win
would be no better than the hypothesis that it will win, and one
could infer no further. (Presumably one would have to have
stopped before this point.)21
There is thus no inconsistency in accepting that some tickets will lose while not
accepting that of others. Yet, that is not to say that such arbitrary choices are
epistemically reasonable. Indeed, they will not be, if the symmetry principle
holds.
Harman seems to endorse a symmetry principle for epistemic acceptance,
for he takes non-arbitrariness to be a mark that distinguishes theoretical from
practical choices:
Suppose, with Harman, that Jim can rationally infer that t1 through
t999,000, say, will lose. Jim is also rational in believing that one of
t1 through 1,000,000 will not lose. It would seem to follow that Jim
could rationally infer a logical consequence of these beliefs,
namely, that one oft999,001-tl,000,000 will not lose. But this is
strongly counterintuitive. If Jim were rational in believing that one
of those 1,000 tickets will win, then, depending on the order of his
inferences, he should either try to get his hands on one of those
1,000 tickets or feel fortunate to be holding one of them already!
But there is no reason for Jim to do either of these things. Thus,
Harman's argument fails ... 24
is to obtain truth and avoid error.,,26 If truth is our ultimate aim, why can we not
have purely epistemic reasons to reject symmetry and accept some lottery
propositions?
Are there philosophers who question symmetry for what appear to be
purely epistemological reasons? It would seem so. In "Sellars on Induction
Reconsidered," Lehrer displays a new interpretation of Wilfrid Sellars's view of
empirical acceptance rules, a view that violates a non-arbitrariness condition that
is implied by the Symmetry Principle. Here is how Lehrer describes Sellars's
view:
Thus, according to the new interpretation, this way to combine the objectives of
accepting true observation statements and at the same time maintaining
explanatory coherence is to accept the percentage ofthe observation statements,
provided it is greater than 50%, that corresponds to the percentage of members
ofK that are known to be B in the total population. For example, if we know that
3/4 K are Band K satisfies the pertinent relevance conditions, then we accept
75% of the statements of the form' a] is B. ,27
While Lehrer maintains his own commitment to symmetry, he does allow
that Sellars's position is "justified by systematic epistemic objectives and is in
no way ad hoc ... ,,28 I agree. Sellars's position demonstrates that rejection of
symmetry can be grounded in purely epistemological, not pragmatic,
considerations.
In rejecting symmetry, we need not endorse a position akin to Sellars's.
If we did, we would license accepting of all but one ticket that it lose! That is
obviously too permissive. Yet, is there not good reason to accept of a few tickets
that they will lose, while not getting carried away? The only barrier in our way
is the Symmetry Principle.
GLENN ROSS 103
The claimant has won this round by neutralizing the skeptic's challenge. The
claimant's original claim, 'The number one ticket has not won', neutralizes the
skeptic's challenge, inasmuch as the conjunction of their two claims is at least
as reasonable as the claimant's original claim. Of course it could happen that this
conjunction of the objection with the original claim does not neutralize the
objection. This will happen if the increase in informational content due to
conjoining does not outweigh the decrease in probability. For example, if the
lottery is not very large, then the probability of these two tickets losing could be
substantially less than the probability of one ticket losing. In a three-ticket
lottery, the probability is thereby halved. It follows that in a small lottery one is
not personally justified in accepting that one holds a losing ticket. That is as it
should be.
The game could continue with the claimant winning round after round by
neutralizing the opposition:
Skeptic: The number two through number n tickets (for some large
n) have not won.
Claimant: It is more reasonable for me to accept that the number
one ticket has not won than to accept that the number two through
n tickets have not won.
On our first supposition, it is no less reasonable to accept that ticket one will lose
on the assumption that ticket two through m will all lose than on the assumption
that they will not all lose. Consequently, this objection of the skeptic does not
compete with the original claim and is thus not a legal move for the skeptic in the
justification game.
We might worry about what happens on precise boundaries. We should
not. Given the limitations of human reasoners to make very fine epistemic
distinctions, the supposition of one ticket's losing will only make the relevant
epistemic difference if the lottery is too small for one to reasonably accept that
a particular ticket will lose. In a very large lottery, if we can successfully judge
that an objection by the skeptic is indeed a competitor, then we will be able to
neutralize or beat it. If one cannot judge whether an objection competes, then it
should not defeat one's personal justification. 30
Notice that up to this point, the Symmetry Principle has not come into
play. So, let us consider these moves:
Skeptic: You do not accept that the number two ticket has not won
but you have no better reason to accept that the number one ticket
has not won.
Claimant: The Symmetry Principle is false.
GLENN ROSS 105
ENDNOTES
I Keith Lehrer, Metamind, Oxford, Clarendon Press, 1990, pp. 235-6 and Theory of Knowledge,
particularly pp. 136-7 and Decision Theory as Philosophy, Cambridge: Cambridge University,
1996, pp. 139-40.
lJohn Pollock, Cognitive Carpentry: a Blueprintfor How to Build a Person, Cambridge, Mass.:
MIT/Bradford, 1995, chapter 2.
4Sharon Ryan, 'The Epistemic Virtues of Consistency," Synthese 109 (1996), pp. 121-41,
particularly p. 126. Ryan is working with a strong notion of justified belie( i. e., whatever
justification knowledge requires. In the lottery paradox I am considering, the relevant concept of
justification ranges from a weak notion of epistemic rational permissibility to a concept of
justification no stronger than Lehrer's notion of personal justification.
S DanaK. Nelkin, "The Lottery Paradox, Knowledge, and Rationality," Philosophical Review 109
(2000) pp. 373-409.
6Richard Foley, Working Without a Net, New York: Oxford, 1993, pp. 164-5.
7Peter Klein, "The Virtues of Inconsistency," Monist 68 (1985), pp. 105-35, particularly p 108.
Note that Klein, like Ryan, is presuming a notion ofjustified beliefthat is required for knowledge.
R I borrow the labels' skeptical' and 'credu lous' for the two positions one can take on symmetry
from John Pollock, "Justification and Defeat," Artificial Intelligence 67 (1994), 377-407, who in
turn acknowledges borrowing from D. S. Touretzky, J. F. Horty, and R. H. Thomason, who speak
of credulous and skeptical reasoners.
9 Bayesians who reject the notion of acceptance in favor of quantitative measures of confidence
would reject all three positions. I will not rehearse the standard replies to such Bayesians.
10 Theory of Knowledge, p. 118.
II Theory of Knowledge, p. 125. These definitions of competition, neutralization, and personal
justification are essentially preserved in Lehrer, Self Trust: A Study of Reason, Knowledge, and
Autonomy, Oxford: Clarendon Press, 1997, p. 30.
12 Theory of Knowledge, p. 130.
13 See Jonathan Vogel in "Are There Counterexamples to the Closure Principle':>" in Doubting,
Michael D. Roth and Glenn Ross (eds.), pp. 13-27, for several such cases where we would
ordinarily claim to have knowledge though they appear to be analogous to lottery situations.
14 Keith DeRose, "Knowledge, Assertion and Lotteries," Australasian Journal of Philosophy 74
15 For this account to succeed, we must reject Robert Stalnaker's principle of conditional excluded
middle for counterfactuals.
16 It is just this disanalogy that Sharon Ryan exploits in (1996).
65-85. Also see Pollock's "A Solution to the Problem ofInduction," Noiis 18 (1984), pp. 423-461
and "Justification and Defeat," Artificial Intelligence 67 (1994), pp. 377-407.
20 Pollock (1991), pp. 80-1.
21 Change in View, p. 71.
22 Change in View, p. 68.
23 Pollock (1994), p. 383.
24 Nelkin (2000), pp. 377-8.
25 Epistemological deontologists might also demur from such a defense of symmetry, seeing
considerations of high probability and coherent acceptance as trumping aprima/acie reasonability
behind symmetry.
26 Theory o/Knowledge, p. 121.
27 Lehrer (1983), p. 471.
28 Lehrer (1983), p. 469.
29 Note that this flexibility to accommodate more credulous intuitions on the lottery is not shared
by Pollock's (1994) notion of justification, inasmuch as his "Principle of Collective Defeat" (p.
383) incorporates a symmetry principle.
30 The argument can be fully spelled out as follows: Suppose the skeptic offers this challenge:
Skeptic: The number two through m tickets have not won.
There are three cases to consider:
(a) The expected epistemic utility of accepting that ticket one will
lose is less than the expected epistemic utility of accepting that all
of the tickets one through m will lose. In this case, the statement
that ticket number one will lose neutralizes the challenge, as in
the first two rounds above.
(b) The expected epistemic utility of accepting that ticket one will
lose is equal to the expected epistemic utility of accepting that all
of the tickets 1 through m will lose. In this case, the statement that
tickets numbered two through m will all lose does not compete
with the original claim ofthe claimant. It is no less reasonable to
accept that ticket one will lose on the assumption that ticket two
through m will all lose than on the assumption that they will not
all lose.
©) The expected epistemic utility of accepting that ticket one will
lose is greater than the expected epistemic utility of accepting that
tickets one through m will lose. There are two sub-cases:
(c-i) The expected utility of accepting that ticket one will lose is greater than the
expected epistemic utility of accepting that tickets two through m will lose. In this
case, the statement that ticket number one will lose beats the challenge, as in the
above case for large n.
(c-ii) The expected utility of accepting that ticket one will lose is less than or
equal to the expected utility of accepting that tickets two through m will lose.
Given the limitations in our abilities to make fine discriminations in the expected
utility for the acceptance of statements, these cases will not arise unless the lottery
is small enough for one ticket to make such a difference. So, in a very large
lottery, these cases will not arise for human reasoners. If we can successfully
GLENN ROSS 107
judge that an objection by the skeptic is indeed a competitor, then we will be able
to neutralize or beat it. If we cannot judge whether an objection competes, then
it does not defeat our personal justification.
Chapter 7
Charles B. Cross*
University of Georgia
According to Keith Lehrer's (1974, 1986, 1988, and 2000) theory of knowledge,
coherence is the basis of epistemic justification. In Lehrer's theory, coherence
is a relation between a proposition and what Lehrer (2000: 170) defines as an
evaluation system:
Thus, on Lehrer's account, p coheres with X of Sat t if and only if all objections
to p are answered or neutralized for S on X at t. Lehrer (2000: 170) defines the
notions of objection, answering, and neutralizing as follows:
In this example, the claim Wendy is lying is an objection to the claim Feldman's
book is in print because it is more reasonable for the subject to accept Feldman's
book is in print on the assumption Wendy is not lying than on the assumption
Wendy is lying. Since it is more reasonable for the subject to accept The book
has been published than to accept Wendy is lying, the claim Feldman's book is in
print answers the objection Wendy is lying. But is every objection to Feldman's
book is in print answered? Consider this example, again adapted (with updated
terminology) from the example in Lehrer 1988:342:
Imagine the skeptic persists and says, "Well, you know people
sometimes lie about when books are in print." Now this skeptical
innuendo does count as an objection to the claim Keith believes in
a sort of indirect way. It would be more reasonable for Keith to ac-
cept that Feldman's book is in print on the assumption that people
do not ever lie about when books are in print than on the assump-
tion that they do sometimes lie. Moreover, it is quite reasonable
for Keith to accept that people do sometimes lie about these mat-
ters. But this skeptical innuendo, though it cannot be answered,
can be neutralized by conjoining the reply that Wendy is not lying.
Of course, the reasonableness of accepting the latter depends on
Keith's background information about Wendy.
The claim Wendy is not lying neutralizes the claim People sometimes lie about
when books are in print as an objection to Feldman's book is in print because
People sometimes lie about when books are in print, but Wendy is not lying is
not an objection to Feldman's book is in print, and because it is as reasonable to
accept People sometimes lie, but Wendy is not lying as to accept People some-
times lie by itself, given the subject's background evaluation system. If every
objection to Feldman's book is in print is either answered (like the claim Wendy
is lying) or neutralized (like the claim People sometimes lie about when books
112 RELATIONAL COHERENCE AND CUMULATIVE REASONING
are in print), then the subject is justified in accepting that Feldman's book is in
print.
This example illustrates how natural it would be to interpret Lehrer's nec-
essary and sufficient condition for justification relative to an evaluation system
as a sufficient condition for the permissibility of an inference. The subject Keith
in the previous example who asks, "Am I justified in believing that Feldman's
book is in print?", may be using certain background information and skeptical
alternatives to evaluate the epistemological status of a belief he already holds,
or he may be considering whether to accept Feldman's book is in print as a new
belief on the basis of that background information. But to consider whether to
accept a given claim on the basis of given background information is to consider
whether to infer that claim from the background information. This would be an
inductive inference, of course, since the claim Feldman's book is in print does
not follow deductively from the subject's background information, but the fact
that the subject would be justified in believing this claim surely indicates that
it would be a good inductive inference. What are the consequences of reinter-
preting Lehrer's account of justification relative to an evaluation system as an
account of inductive inference? In order to answer this question we will first
need an introduction to the theory of inductive inference in its most general
form: the theory of cumulative reasoning.
2. CUMULATIVE REASONING
Reflexivity If p E X then X f- p.
But most of the reasoning subjects actually do on a day-to-day basis (such as in-
ference to the best explanation) is inductive. Despite its nondeductive character,
it is possible to theorize in a proof-theoretic way about inductive reasoning, and
formal approaches to inductive reasoning abound in the artificial intelligence
literature. 4 Formally, the most conspicuous way in which inductive inference
differs from deductive inference is the fact that inductive inference is nonmono-
tonic. Where 'X ~ p' means that p is an inductive consequence of X (under
some given conception of inductive inference represented by '~'), it is possible
for it to be the case that X ~ p and X r;;:; Y even though Y If p. For example,
CHARLES B. CROSS 113
Knowing that the match was struck and that oxygen was present, we can rea-
sonably infer that the match lit. But if our evidence is expanded to include
information that the match was wet, then the inference to the conclusion that
the match lit is not licensed given the expanded evidence.
If in the above list of conditions Monotonicity is replaced by the weaker
condition of Cautious Monotony, the result is a version of what Gabbay (1985)
calls cumulative inference:
Reflexivity If p E X then X ~ p.
Makinson (1989) and Kraus, Lehmann, and Magidor (1990) present semantical
approaches to cumulative inference based in part on Shoham's (1988) notion
of preferential entailment. By investigating how Lehrer's notion of relational
coherence can be interpreted as an account of inductive reasoning we will be
developing an alternative to the received preferential model semantics for cu-
mulative inference.
evaluation systems associated with some fixed S and t. Now, different evalua-
tion systems in Lehrer's theory can share the same acceptance system, hence a
subject's evaluation system at a given time is not strictly a function of his or her
acceptance system. But in the coherence-based theory of inductive consequence
to be explored below, the inductive consequences associated with a given eval-
uation system will be a function of the acceptance system component alone.
Accordingly, the theory of inductive reasoning presented below may appear to
assume that a subject's reasoning and preference systems are a function of his
or her acceptance system. How can this be reconciled with the spirit of Lehrer's
theory? The conflict is only apparent. When two evaluation systems incorporate
the same acceptance system but not the same preference and reasoning systems,
this means that the two evaluation systems in question are associated with dif-
ferent relations of comparative reasonableness relative to the given acceptance
system. In the theory of inductive consequence developed below, the inductive
consequences of X are determined in part by comparative reasonableness rela-
tive to X. When two evaluation systems incorporate the same acceptance sys-
tem but not the same preference and reasoning systems, this means that the two
evaluation systems in question are associated with different notions of inductive
consequence. Lehrer's theory of evaluation systems is therefore, potentially, a
theory of the dynamics of the inductive inference concept-a theory of induc-
tive logic revision. We shall not offer a dynamic theory here. The theory to be
presented below will concern the statics of coherence-based inductive inference.
Since we are taking the variables S and t as fixed, Lehrer's seven-place
relation of comparative reasonableness ("it is more reasonable for S to accept
that p on the assumption that q than to accept r on the assumption that s on the
basis of X at t") will become a five-place relation ("it is more reasonable to
accept p on the assumption q than to accept r on the assumption s on the basis
of X,,).5 These modifications allow us to define 'X ~ p' as in C1 below and
work with C2-C6 instead of Lehrer's D2-D6. We will assume a background
deductive logic whose consequence relation is represented by the straight turn-
stile 'f-'. 6 We will assume that this background logic is a compact, consistent
extension of classical tmth-functionallogic and that an expressively complete
set of boolean connectives is available in the language:
Given C 1-C6, the key to the logic of 'f-v' will of course be the logic of the com-
parative reasonableness relation. Lehrer (2000:144) states that he takes com-
parative reasonableness as undefined, but he identifies both probability (Lehrer
2000:144) and epistemic utility (Lehrer 2000:146) as factors in the determina-
tion of the degree of reasonableness of a claim. Lehrer (2000: 146) even provides
an expected utility equation for calculating degrees of reasonableness. Our ap-
proach to comparative reasonableness will be entirely nonquantitative. 7 We will
treat comparative reasonableness relative to a set X of sentences as a binary re-
lation >- x relating one pair (p, q) of sentences to another such pair. A locution
of the form 'It is more reasonable to accept p on the assumption of q than to
accept r on the assumption of s on the basis of X' will thus be abbreviated
'(p, q) >- x (r, s)'. With comparative reasonableness understood qualitatively,
the following all seem plausible as principles of comparative reasonableness:
(p,q) ~x (r,s) ifandonlyif (p,q) 'l-x (r,s) and (r,s) 'l-x (p,q).
The following result follows immediately by Nontriviality, Irreflexivity, Asym-
metry, and Negative Transitivity:
That is, ~ x is reflexive, symmetric, and transitive for any fixed X. The fact that
~ x is an equivalence relation makes it acceptable to interpret' (p, q) ~ x (r, s)'
as meaning it is as reasonable to accept p given q as to accept r given s on the
basis of X. Where q and s are both logical truths, '(p, q) ~ x (r, s)' means
p is as reasonable as r on X, and it is this notion of equal reasonableness that
we appeal to in C6. The fact that ~x is an equivalence relation also yields the
following useful substitution principle.
CHARLES B. CROSS 117
Theorem 2 If (PI , qI) '::::.X (p2, q2) and (rIl 81) '::::.x (r2' 82), then (PI, qr) >- x
(rI' 8r) iff(p2, q2) >- X (r2' 82).
Proof: Assume that (PI, qI) '::::.x (p2, q2) and (rI, 8r) '::::.x (r2, 82). Now for re-
ductio suppose that (PI, qr) >- X (rI' 8r), but (p2, q2) >I- X (r2' 82). By hypothe-
sis, (PI, qI) >I- X (p2, q2) and (p2, q2) >I- X (r2' 82) and (r2' 82) >I- X (rI' 81). By
two applications of Negative Transitivity, (PI, qr) >I- X (rI, 81), which is con-
trary to hypothesis. So by reductio if (PI, qI) >- X (rr, 81) then (p2, q2) >- X
(r2' 82)' An exactly similar argument shows that if (p2, q2) >- X (r2, 82) then
(PI, qI) >- X (rI' 8r). (QED)
Equivalence has this direct consequence:
Proof: Suppose that X f-v P and P -1r q. If X r then X f-v q follows imme-
diately, so suppose X j,L. Then every objection to P on X is either answered or
neutralized. Let 0 be an obj ection to q on X. Then (q, -'0) >- X (q, 0), but since
P -1r q it follows by Equivalence that (p, -'0) >- X (p, 0). Hence 0 is an objection
to p, so 0 is either answered or neutralized as an objection to P on X. Suppose
that 0 is answered as an objection to P on X. Then (p, T) >- x (0, T), but since
P -1r q, it follows by Equivalence that (q, T) >- x (0, T). Hence 0 is answered
as an objection to q on X. Alternatively, suppose that n neutralizes 0 as an ob-
jection to P onX. Then (p, -,(o&n)) >l-x (p, o&n) and (o&n, T) '::::.x (0, T).
Since P -1r q, it follows by Equivalence that (q, -, (0 & n)) >I- x (q, 0 & n), so
n neutralizes 0 as an objection to q. So 0 is either answered or neutralized as an
objection to q on X. Generalizing on 0, it follows that X f-v q. (QED)
Superiority expresses the idea that the logical consequences of a consistent
set X answer all of their respective objections. The role of Superiority in
the logic of coherence is to ensure that the principle of Supraclassicality holds
for f-v:
Inconsistency If X J.L and {p, q} f--- and {p} J.L and {q} J.L, then (p, -,q) >- x
(p, q).
Top/Bottom If X J.L and {p} J.L, then (T, p) >- x Cl, p).
Bottom/Top If X J.L, then el,.1) >- x (.1, T).
Exclusive Objection for.1 If (.1, -,p) >- x (.1,p), then f--- p.
Exclusive Equivalence If X J.L and (.1, T) ~x (p, T), then {p} f---.
Proof: Suppose that X ¥. For reductio, suppose that there is a p such that
X f-v p and X f-v -'p.
Case 1: Suppose that I- p. Then -,p -II- ..l, so by Theorem 3, X f-v ..l.
Since X ¥, it follows that every objection to ..l is either answered or neu-
tralized on X. By Bottom/Top and Equivalence, (..l, -, T) 'r x (..l, T), so
T is an objection to ..l on X. Since ¥ -, T, we have by Top/Bottom that
(T, T) 'r x (..l, T), so by Asymmetry, (..l, T) >f x (T, T). By Exclusive
Objection for ..l, every objection to ..l on X is logically equivalent to T, so
by Equivalence, (..l, T) >f x (0, T) for every objection 0 to ..l on X. Hence
..l does not answer any of its objections on X. Let n neutralize 0 as an ob-
jection to ..l on X. Then 0 is an objection to ..l on X and 0 & n is not an
objection to ..l on X and (0 & n, T) r:::=.x (0, T). Exclusive Objection for ..l
implies that 0 -II- T, so by Equivalence we have that T & n is not an objection
to ..l on X, i.e. (..l, -,(T & n)) >f x (..l, T & n), and T & n is as reasonable
as T on X, i.e. (T & n, T) r:::=.x (T, T). But T & n -II- n, so by Equivalence,
(i) (..l, -,n) >f x (..l, n) and (ii) (n, T) r:::=.x (T, T). By Top Level, (ii) implies
n -II- T, from which it follows by (i) and Equivalence that (..l,..l) >f x (..l, T),
which contradicts Top/Bottom. So a contradiction follows from the hypothesis
that I- p.
Case 2: Suppose that I- -'p. This hypothesis leads to a contradiction by
an argument similar to Case 1.
Case 3: Suppose that ¥ p and ¥ -'p. By Inconsistency, (p, -,-,p) 'r x
(p, -,p) and (-,p, -,p) 'r X (-,p, p). Hence -,p is an objection to p on X and p is
an objection to -'p on X. Since X f-v p and X f-v -,p and X ¥, -,p is either
answered or neutralized as an objection to p on X, and p is either answered
or neutralized as an objection to -'p on X. Asymmetry implies that p and -,p
CHARLES B. CROSS 121
Proof: Assume Invariance under Reformulation and suppose that X U {p} r-- r
and X U {p} f- q and X U { q} f- p. If X U { q} f-, then X U { q} r-- r by C 1, so
suppose that X U {q} J.L. Since X U {p} f- q and X U {q} f- p, it follows that
Cn(X U {p}) = Cn(X U {q}). Let 0 be an objection to r on Xu {q}. Then
(r, -'0) >-- xu{ q} (r, 0). By Invariance under Reformulation, (r, -'0) >-- xu{p}
CHARLES B. CROSS 123
Distribution for Disjunction If X U {p} j,L and X U {q} j,L, then for all r,
s, t, and u, (r,s) >-xu{pvq} (t,u) only if (r,s) >-xu{p} (t,u) and
(r,s) >-xu{q} (t,u).
Proof: Assume Distribution for Disjunction, and suppose that X U {p} j,L and
X U {q} j,L and X U {p} f"v r and X U {q} f"v r. Let 0 be an objection to r on
Xu {p V q}. Then (r, -'0) >- xu{pVq} (r,o). By Distribution for Disjunction,
(r, -'0) >- xU{p} (r,o) and (r, -'0) >- xU{q} (r,o). Hence 0 is an objection to r
on both X U {p} and X U {q}. (QED)
The question of how to secure Answering and Neutralizing for Disjunctions via
a reasonable principle of comparative reasonability remains an open problem.
Exclusive Objectiou for T If X J.L and (T, ...,p) )- x (T, p), then {p} f-.
4. CONCLUSION
ENDNOTES
* I am pleased and honored to participate in celebrating the career of so brilliant a philosopher
as Keith Lehrer. Lehrer's status as a leading figure in the field is richly deserved, and, without a
doubt, his work will continue to stand the test of time.
1 Cumulativity was defined by Gabbay (1985), but see also Kraus et al. 1990, Makinson 1989,
and Makinson 1994.
2 But see Lehrer 2000:144-146 for a discussion of relative reasonableness, probability, and ex-
pected value.
3 In this and later sections, the variables p, q, r, and s range over statements in some given
language, and the variables X and Y with and without subscripts range over sets of statements.
The expression 'X f- p' means that p can be inferred from X in the inference system defined
by'f-'.
4 See Makinson 1994 for a survey.
5 In parallel with Lehrer's definitions, "it is more reasonable to accept p than to accept r on the
basis of X" will be assumed to mean "it is more reasonable to accept p on the assumption T than
to accept r on the assumption T on the basis of X", where T is some fixed logical truth. 1. is
defined as -, T.
6 'p -jf- q' will mean that {p} f- q and {q} f- p. Where X is a set of sentences, 'X f- ' will mean
that X f- p and X f- -,p for some p.
7 In Lehrer's theory, comparative reasonableness is conditional as well as comparative ("it is more
reasonable to accept p on the assumption q than to accept r on the assumption s"). For a well
worked-out example of a nonquantitative theory of unconditional comparative reasonableness,
see Chisholm and Keirn 1972.
8 It might be argued that distinctions of comparative reasonableness can be made in such cases.
If this is right, then our adoption of Nontriviality should be considered a simplifying assumption
that reflects a decision not to address comparative reasonableness for inconsistent X.
9 See Cross 1990 for an application of a version of this result to the question of the tenability of
the Ramsey test for conditionals.
CHARLES B. CROSS 127
REFERENCES
Chisholm, R. and R. G. Keirn. 1972. "A system ofepistemic logic." Ratio 14:99-115.
Cross, C. 1990. "Belief revision, nonmonotonic reasoning, and the Ramsey test." In Knowledge
Representation and Defeasible Reasoning, edited by H. Kyburg, R. Loui, and G. Carlson,
223-244. Dordrecht: Kluwer.
Gabbay, D. 1985. "Theoretical foundations for non-monotonic reasoning in expert systems."
In Logics and Models of Concurrent Systems, edited by K. R. Apt, 439-457. Berlin:
Springer-Verlag.
Kraus, S., D. Lehmann, and M. Magidor. 1990. "Nonmonotonic reasoning, preferential models,
and cumulative logics." Artificial Intelligence 44:167-207.
Lehrer, K. 1974. Knowledge. Oxford: Oxford University Press.
~~-. 1986. "The coherence theory of knowledge." Philosophical Topics 14:5-25.
~~-. 1988. "Metaknowledge: Undefeated justification." Synthese 74:329-347.
~~-. 2000. Theory ofKnowledge. 2nd ed. Boulder: Westview Press.
Makinson, D. 1989. "General theory of cumulative inference." In Lecture Notes in Artificial
Intelligence 346: Nonmonotonic Reasoning, edited by M. Reinfrank, 1. de Kleer, M. L.
Ginsberg, and E. Sandewall, 1-18. Berlin: Springer-Verlag.
Makinson, D. 1994. "General patterns in nonmonotonic reasoning." In Handbook of Logic in
Artificial Intelligence and Logic Programming, Volume 3: Nonmonotonic Reasoning and
Uncertain Reasoning, edited by D. M. Gabbay, C. 1. Hogger, and 1. A. Robinson, 35-110.
Oxford: Oxford University Press.
Shoham, Y. 1988. Reasoning about Change. Cambridge: MIT Press.
Chapter 8
Wolfgang Spohn
University of Konstanz
Meets what? Ranking theory is, as far as I know, the only existing theory suited
for underpinning Keith Lehrer's account of knowledge and justification. If this
is true, it's high time to bring both together. This is what I shall do in this paper.!
However, the result of defining Lehrer's primitive notions in terms of
ranking theory will be disappointing: justified acceptance will, depending on the
interpretation, either have an unintelligible structure or reduce to mere
acceptance, and in the latter interpretation knowledge will reduce to true belief.
Of course, this result will require a discussion of who should be disappointed.
So, the plan of the paper is simple: In section 1 I shall briefly state what
is required for underpinning Lehrer's account and why most of the familiar
theories fail to do so. In section 2 I shall briefly motivate and introduce ranking
theory. Basing Lehrer's account on it will be entirely straightforward. Section 3
proves the above-mentioned results. Section 4, finally, discusses the possible
conclusions.
pp.145ff.) and also used for a solution of the lottery paradox. However, I am
doubtful because epistemic decision theory has remained a promise that has
never been redeemed in a satisfying way in the last 35 years.
The basic difficulty, I believe, is this: Probability theory may claim, in a
way, to offer a complete epistemology. If so, it is hard to complement it or to
merge it with other epistemological ideas like acceptance, epistemic decisions,
or whatever, and radical probabilism as Jeffrey (1992) has defended it seems
unavoidable.
Hence, we should put probability theory aside and rather look at theories
dealing directly with acceptance or belief. A large variety of such theories-such
as default logic, AGM belief revision theory, Pollock's and other accounts of
defeasible and non-monotonic reasoning, etc.-has been developed in the past
25 years. Maybe they provide an account of Lehrer's preference system as well.
Alas, they don't. At least, I claim this with confidence with respect to AGM
belief revision theory. There it is shown that the behavior of belief revisions is
equivalent to the behavior of so-called entrenchment relations. 5 These could
indeed fill the role of Lehrer's unconditional preference relation. Maybe
entrenchment relations can be generalized so as to capture the special case A IC
>- BIC of Lehrer's preference relation which refers twice to the same condition.
However, Lehrer requires the full conditional relation, which cannot be
accounted for in AGM belief revision theory. I suspect that essentially the same
is true of all accounts of defeasible reasoning implicitly or explicitly appealing
to some kind of epistemic ordering.
There is only one theory that is about belief or acceptance and provides
a sufficiently powerful preference system: ranking theory. That's why I said it
is the only existing theory suited for underpinning Lehrer's account. What does
it look like?
2. RANKING THEORY
The basics are quickly told. Originally, ranking theory was developed 6 in
order to overcome essential restrictions of AGM belief revision theory. As it
turns out, AGM theory generally accounts only for one step of belief revision
and thereafter returns to a static picture. But, of course, a full dynamics has to
account for several or iterated belief changes. The problem has been around since
Harper (1976), and there have been quite a number of attempts to solve it within
the confines of AGM theorizing.? However, I find these proposals inferior to the
one provided by ranking theory.
Iterated belief revision is not our concern here. However, there exists a
close connection between iterated belief change and full conditional epistemic
preference. 8 It is for this reason that ranking theory, though addressed to the
former, can also provide for the latter. So let's take a look.
132 LEHRER MEETS RANKING THEORY
Hence, K(A) > 0 says that A is disbelieved (to some degree), and K(A) > 0 says
that A is believed. 9 K(A) = 0 only expresses that A is not disbelieved and leaves
open the possibility that A is not disbelieved as well. Since there is no point here
in distinguishing between belief and acceptance, we thus have
proposition is more strongly disbelieved than any other. More substantial is the
law ofdisjunction that K(AuB) = min {K(A), K(B)}. Clearly, the disjunction AuB
cannot be more firmly disbelieved than either of its disjuncts. Nor can it be less
firmly disbelieved than both disjuncts, since this would entail the absurdity that
given AuB both, A and B, are disbelieved, though AuB is not. An immediate
consequence is the law ofnegation that either K(A) = 0 or K(A) = 0 (or both). A
and A cannot be both disbelieved. Perhaps the most important law is the law of
conjunction that K(AnB) = K(A) + K(BIA) which follows trivially from the
definition of conditional ranks. It says that in order to arrive at the degree of
disbelief in AnB one has to sum up the degree of disbelief in A and the
additional degree of disbelief in B given A. This, I believe, agrees with intuition.
There is a surprisingly well working translation from probabilistic into
ranking terms which almost automatically generates a large number of ranking
theorems from probability theorems. This applies also to the account of belief
change. The basic rule for probabilistic belief change is simple conditionalization
according to which one moves to the probabilities conditional on the information
received. This is generalized by Jeffrey's conditionalization lO which is
WOLFGANG SPOHN 133
unrestrictedly performable and thus defines a full dynamics within the realm of
strictly positive probability measures. In the corresponding way, such
conditionalization with respect to ranks offers a full dynamics of belief or
acceptance. 11
This informal hint at belief revision may suffice. However, we should
formally introduce belief contraction because Lehrer makes explicit use of it in
what he calls the ultrasystem. Belief contraction is the operation of giving up
some belief without adding new ones. It is extensively discussed in AGM belief
revision theory because it is interchangeable with belief revision.!2 Within
ranking theory it is easily defined as well (and turns out then to have all the
properties described in AGM theory!3):
One may be tempted to say that Lehrer's account is wedded to ranking theory.
The wedding is unhappy, however. On the basis of Definition 4, Lehrer's
account ofjustification is translated in a straightforward way, and the fatal results
are inescapable:
Suppose conversely that the condition holds true for all m ?: n. This
unfolds into three cases:
°
propositions accepted by K are accepted. Moreover, if A is true and B is false,
then K*(A) = :<:; K*(B), and hence in the preference system provided by K* no
false proposition is ever preferred to a true one, as required by Lehrer in D8, p.
171. Finally, we do not worry about the reasoning system of1.c*, for the familiar
reason. All in all, the explication is remarkably smooth, and we may conclude
with
This covers also the case where no falsehood is accepted in K, i.e., where
K* =K.
On this basis, the rest of Lehrer's definitions is immediately translated:
4. WHAT TO CONCLUDE?
It is not clear what to think of these results. I find that the structure of what
I called justified acceptance in the strong sense is too weird to be worth
discussing. Hence, I proceed on the assumption that it is the weak sense that is
relevant. I have already indicated why I believe to agree with Lehrer on this
point. But Theorems 2 and 3 look troublesome as well, though it is not clear
where to locate the trouble. There are several ways, and more than one good
way, to respond.
(1) One may point to various flaws in my translation. For instance, I have
carelessly interchanged belief and acceptance, whereas Lehrer (pp. 12f.)
emphasizes their difference. Likewise, I have defined an acceptance system as
a set of accepted propositions, whereas for Lehrer it is a set of propositions ofthe
form "I accept that A". But these kinds of flaws are insignificant. Also, I have
neglected Lehrer's own foundations in terms of trustworthiness (pp.138ff.), but
taking them into account would have no effect on the present considerations.
However, my fixation of Lehrer's reasoning system by rightaway
assuming deductive closure is doubtlessly a major deviation from Lehrer. But,
again, I don't think it does any harm. The acceptance system to start with could
as well consist in an arbitrary, not deductively closed set of accepted statements,
as long as it is consistent. Working out what is justified on the basis of such an
WOLFGANG SPOHN 139
acceptance system means working out the preference system, which is a more
permanent disposition of the epistemic subject extending to all statements. And
it means, as we have observed above, working out the reasoning system, i.e.,
accepting the logical consequences of what one has justifiedly accepted. Hence,
we are back at deductive closure, at least as far as justified acceptance is
concerned. Theorem 2 then says that the deductive closure of an acceptance
system is justifiedly accepted relative to this system, and this is equally
troublesome.
No, Theorems I and 2 are generated by the characteristics of the
preference system as specified by a ranking function. The trouble lies there and
not in my light way of dealing with deductive relations.
(2) This suggests another conclusion. If one proceeds from preference
systems generated by ranking functions and arrives at such undesired results,
then the two theories do not fit together, and one has to look for other preference
systems. After all, nobody claims that ranking theory delivers the only legitimate
kind of preference systems. So, the conclusion would be that ranking theory may
have useful applications, but Lehrer's account ofjustification does not belong to
them.
(3) However, I tend to the reverse conclusion. The more offers for an
underpinning of Lehrer's account are rejected, the stronger the obligation to
come up with some sound theoretical foundation. As for my part, I doubt that
there is any better offer than the one made here. In any case, as long as the
foundation is missing, there is not really any theory of justification and
knowledge.
Perhaps this quest for a theoretical underpinning is too strong, though.
Concerning the acceptance system the picture rather seems to be that it collects
all the variegated items of information and inference, with the aim of truth, but
without critical standards. Any arbitrary set of statements may be formed in this
way. Then, it seems, there can be no theory about acceptance systems, and
asking for one is asking too much.
This attitude, however, does not carry over to the preference system. We
need not entertain the illusion that it is uniquely determined by rationality alone.
Carnap was under this illusion with inductive logic, but soon woke up. The
preference system and hence the standards ofjustification may well be subjective
to some extent. However, this is not to say that anything goes. This would mean
anarchy and throwing away the idea of rationality altogether. Some rationality
standards should be set up and defended. This is just what ranking theory
attempts to do, though perhaps in a debatable way. In any case, one cannot
simply be silent on the structure of preference systems. Hence, if one thinks that
the theorems above are undesirable, then, I think, Lehrer's account of
justification is really in trouble.
(4) Perhaps, though, one need not think that the theorems are undesirable.
When comparing ranking theory with Pollock's defeasible reasoning in Spohn
140 LEHRER MEETS RANKING THEORY
(2002) I concluded, among other things, that ranking functions are best
compared with what Pollock calls ideal warrant, which is, so to speak, the end
product of his defeasible reasoning machinery. IS Though Pollock's theory is
quite different from Lehrer's, this suggests that acceptance according to ranking
functions is already justified acceptance. This suggestion is supported by the fact
that ranking acceptance cannot yield any arbitrary acceptance system, but is
deductively closed, due to a correctly and maximally executed reasoning system.
In this perspective, then, Theorem 2 would not at all be surprising, it should be
expected. Accordingly, ranking theory and Lehrer's account ofjustification may
indeed be seen as mutually supporting each other, since Theorem 2 shows that
they reach the same result via entirely different considerations.
Too much harmony? Yes, I think so. First, the talk of ideal warrant is
Pollock's, and it was helpful in the above-mentioned comparison. But there is
nothing in ranking theory by itself forcing this comparison. Ranking theory is,
as I prefer to say more neutrally, about rational belief and its dynamics. So,
Theorem 2 rather shows either that ranking theory is about ideal warrant, despite
my disclaimer, or that ideal warrant reduces to rational belief. This would be my
preferred conclusion, but it reopens, it seems, a difference to Lehrer.
Secondly, even if Theorem 2 lends support to Lehrer's account of
justification, it does so in an unfriendly way, because it renders it vacuous at the
same time. The sophisticated considerations about answering and neutralizing
objections do not do any real work. This holds also when one starts, as Lehrer
does, from an arbitrary, unorderly acceptance system, because it is simply its
deductive closure that is justified relative to it. 16 This vacuity is certainly against
Lehrer's intentions.
Thirdly, there remains a problem with Theorem 3. If, as stated above,
ranking functions represent rational belief and nothing stronger, then the
reduction of knowledge to true belief as represented by ranking functions appears
doubtful, even though it has been defended by von Kutschera (1982, sect. 1.3),
and Sartwell (1991, 1992). In any case, it is unacceptable to Lehrer. Atthis point,
hence, harmony ceases at the latest.
So, as I said, it is not really clear what to conclude, though (3) would be
my preferred conclusion. The case shows once more how difficult it often is to
square different epistemological approaches.
However, there is, I think, a general lesson to learn. Justification is a
central notion in epistemology, and hence it is rightly scrutinized in many
discussions, from many perspectives, and with many examples and arguments.
In all that literature, though, I find very little rigorous theory. However, the
amenability to rigorous theorizing provides an important test. This test is usually
not even sought. I? But it is useful as a critical and as a constructive authority. At
least I hope to have shown this here with respect to the paradigm of Lehrer's
epistemology.
WOLFGANG SPOHN 141
ENDNOTES
I I am indebted to Gordian Haas for various valuable remarks and for discovering bad faults in a
first draft which required major corrections and to Erik Olsson for further helpful suggestions.
2 In the sequel, mere page numbers are always meant to refer to Lehrer (2000).
3 In D 1, p. 170, Lehrer mentions only the unconditional preference relation, but it is clear that he
requires the conditional one.
4 On p. 127, when introducing the reasoning system, Lehrer refers only to "cogent" reasoning, and
generally the notion of validity makes clear sense only relative to deductive inference.
5 See Al chourr6n et al. (1985), which is considered as the foundation of AGM belief revision
theory, or Gardenfors (1988, ch. 4).
6 In Spohn (1983, sect, 5.3) and (1988).
7 Cf., e.g., Nayak (1994).
8 The connection is not obvious and perhaps not cogent. In order to really understand it we would
have to go much more deeply into issues of belief revision than we can and need to do here.
9 A is the complement of A with respect to W. The thinking in negations will continue, though I
14 Lehrer (1971, p.221) took recourse to the same move when observing that the rule of induction
he proposed is not deductively closed.
15 Cf. Pollock (1995, sect. 3.10).
16 If the initial acceptance system should be inconsistent, the theoretical situation changes
drastically. Then paraconsistent logic may help, or the theory of consolidation (cf. Hansson 1994
and Olsson 1998b), or whatever. But there is no evidence in Lehrer's writings that this is his
problem.
17 For instance, it springs to one's eyes that at least three ofthe five conditions Bonjour (1985, pp.
95-99) offers for coherence are theoretically hardly explicable.
142 LEHRER MEETS RANKING THEORY
REFERENCES
Alchourr6n, C.E., P. Gardenfors, D. Makinson (1985), "On the Logic of Theory Change: Partial
Meet Functions for Contraction and Revision", Journal of Symbolic Logic 30, 510-530.
BonJour, L. (1985), The Structure of Empirical Knowledge, Cambridge, Mass.: Harvard
University Press.
Gardenfors, P. (1988), Knowledge in Flux, Cambridge, Mass.: MIT Press.
Hansson, S.O. (1994), "Taking Belief Bases Seriously", in: D. Prawitz, D. Westerstahl (eds.),
Logic and Philosophy of Science in Uppsala, Dordrecht, Kluwer, pp. 13-28.
Harper, W.L. (1976), "Rational Belief Change, Popper Functions, and Counterfactuals", in: W.L.
Harper, c.A. Hooker (eds.), Foundations ofProbability Theory, Statistical Inference, and
Statistical Theories of Science, vol. I, Dordrecht: Reidel, pp. 73-115.
Jeffrey, R.C. (1965), The Logic of Decision, Chicago: The University of Chicago Press, 2nd ed.
1983.
Jeffrey, R.C. (1992), Probability and the Art of Judgment, Cambridge: Cambridge University
Press.
Lehrer, K. (1971), "Induction and Conceptual Change", Synthese 23, 206-225.
Lehrer, K. (1974), "Truth, Evidence, and Inference", American Philosophical Quarterly 11,79-92.
Lehrer, K. (2000), Theory of Knowledge, Boulder, Colorado: Westview Press, 2nd edition.
Nayak, A.c. (1994), "Iterated Belief Change Based on Epistemic Entrenchment", Erkenntnis 41,
353-390.
Olsson, E. (1998a), "Competing for Acceptance: Lehrer's Rule and the Paradoxes of
Justification", Theoria 64, 34-54.
Olsson, E. (1998b), "Making Beliefs Coherent", Journal ofLogic, Language, and information 7,
143-163.
Pollock, J.L. (1995), Cognitive Carpentry, Cambridge, Mass.: MIT Press.
Sartwell, C. (1991), "Knowledge is Merely True Belief" American Philosophical Quarterly 28,
157-165.
Sartwell, C. (1992), "Why Knowledge is Merely True Belief", Journal ofPhilosophy 89, 167-180.
Spohn, W. (1983), Eine Theorie der Kausalitat, unpublished Habilitationsschrift, Munich.
Spohn, W. (1988), "Ordinal Conditional Functions. A Dynamic Theory of Epistemic States", in:
W.L. Harper, B. Skyrms (eds.), Causation in Decision, BeliefChange, and Statistics, vol.
II, Dordrecht: Kluwer, pp. 105-134.
Spohn, W. (2002) "A Brief Comparison of Pollock's Defeasible Reasoning and Ranking
Functions", Synthese 31, 39-56.
von Kutschera, F. (1982), Grundfragen der Erkenntnistheorie, Berlin: de Gruyter.
Chapter 9
Carl G. Wagner*
The University of Tennessee
1. INTRODUCTION
For the sake of simplicity it is assumed in what follows that your frame of dis-
cernment n regarding possible states of the world is finite. 2 It is also supposed
that there is an infinitely divisible unit of utility. Suppose that you are able to
assign to each event A c n a real number p(A) such that
1° You are willing to pay anything less thanp(A) units of utility in exchange
for receiving one such unit if A occurs, and nothing if A fails to occur;
and
2° In exchange for receiving anything more than p(A) units of utility, you
are willing to obligate yourself to pay one such unit if A occurs, and
nothing if A fails to occur.
whil£
cod I
blue
The set functions (3 and a are, respectively, the Strassenian lower and upper
probabilities 5 on events in 8 induced by p and the compatibility relation T.
The reasons for this terminology will soon be made clear, but let us first note
some basic properties of these set functions.
Theorem 3.1. The set functions (3 and a defined by (3.3) and (3.4) have the
following properties:
(i) 0 ~ (3(A) ~ a(A) ~ l,for all A c 8.
(ii) (3(0) = a(0) = 0 and (3(8) = a(8) = 1.
Then apply monotonicity of p and the principle of inclusion and exclusion for p.
The inequality (3.6) follows from (3.5) and (iv). Assertion (vi) follows from the
case r = 2 of(3.5) and (3.6). 0
CARL G. WAGNER 147
(iv) {3 = 0:.
Proof Straightforward. o
The above theorem simply confirms what one would expect, namely, that
when T is in effect a 8-valued random variable, then 0: and {3 coincide with the
usual probability measure induced on events in 8 by p and T.
We now demonstrate the appropriateness of calling 0: and (3 upper and
lower probabilities. In what follows, fA denotes the indicator of the event A,
i.e., fA(e) = 1 ife E A and fA(e) = 0 ife E A, and we routinely omit the
phrase "units of utility." The betting commitments described in 10 and 2 0 of
section 2 above will be tersely characterized, as a willingness to "buy fA for
p(A) - c, for all c > 0," and "sell fA for p(A) + c, for all c > 0."
Suppose that p is your subjective probability on events in D, with T, 0:,
and (3 as above. Except in the case described in Theorem 3.2, you have inad-
equate information to ground assessment of a subjective probability on events
in 8. There are, however, identifiable constraints on any such probability.
Theorem 3.3. Any coherent probability q that might be a candidate for repre-
senting your threshold prices for events in 8 must satisfy
As noted above, you are perfectly justified in the above situation in refus-
ing to announce threshold prices for events in 8. But there are some additional
bets that you ought to be willing to make:
Theorem 3.4. If f3 and a are defined by (3.3) and (3.4) you ought to be willing,
for each A c 8, and for all c > 0, to
1° buy lAfor f3(A) - c, and
Proof Suppose that A(A) < q(A) for some A. Then, by self-conjugacy of A,
A(A) = 1- A(A) > 1-q(A) = q(A), contradicting thefact that A is dominated
byq. 0
It follows from Theorems 4.1 and 4.2 that if A avoids a sure loss, and is
self-conjugate, then A is in fact a coherent probability. Buehler's argument that
A must be self-conjugate goes as follows: Clearly, A(A) + A(A) ~ 1; otherwise
you will suffer a sure loss. Suppose that A(A) + A(A) < l. Let a and b be such
that A(A) < a, A(A) < b, and a + b < l. Then you'll reject buying fA for a
and fA for b, even though by accepting both bets you would be guaranteed of
the net gain 1 - a - b > O. So you will miss out on a sure gain. Apart from the
fact that missing a sure gain is considerably less serious than suffering a sure
loss, this argument is further weakened by its dependence on your being offered
fA and fA one at a time, with no knowledge that both will be offered. If you
were offered these bets simultaneously, you would recognize immediately that
you were being offered a certain payoff of 1, and would clearly agree to pay any
price less than 1 in exchange.
Here is a simple way in which lower and upper probabilities satisfy-
ing (4.1) arise: Let P be a nonempty family of coherent probabilities on events
in D and let
5. CONCLUSION
ENDNOTES
positive gain, and the possibility of a loss (cf. Eannan 1992, p. 41). Our treatment also allows for
a natural segue to the account of upper and lower probabilities in section 4.
4 Devotees of the principle of insufficient reason would adopt the unifonn distribution here, thus
employing the same distribution in the case of complete ignorance that they would given reliable
infonnation that exactly half the balls in the urn are white.
5 In 1967 Dempster, unaware of Strassen's 1964 paper, published a similar analysis. Shafer
(1976) offered a sui generis account of set functions having the monotonicity properties of the
lower probability (3, regarding such set functions, which he called beliejjimctions, as directly
assessable measures of degrees of belief.
152 TWO DOGMAS OF PROBABILISM
REFERENCES
Buehler, R. 1976. "Coherent Preferences." Annals ojStatistics 4: 1051--1064.
Dempster, A. 1967. "Upper and Lower Probabilities Induced by a Multivalucd Mapping." Annals
of Mathematical Statistics 38:325-339.
Earman,1. 1992. Bayes or Bust? Cambridge: MIT Press.
Glymour, C. 1980. Theory and Evidence. Princeton: Princeton University Press.
Jeffrey, R. 1965. The Logic of Decision. New York: McGraw-Hill; 1983, 2nd ed., Chicago:
University of Chicago Press.
----. 1988. "Conditioning, Kinematics, and Exchangeability." In B. Skyrms and W. Harper
(eds.), Causation, Chance, and Credence, vol. I, Dordrecht: Kluwer,221-255.
- - - . 1991. "Postscript 1991: New Explanation Revisited." In Jeffrey 1992, 103-107.
- - - . 1992. Probability and the Art ofJudgment. Cambridge: Cambridge University Press.
'---. 1995. "Probability Reparation: The Problem of New Explanation." Philosophical Stud-
ies 77:97-102.
2001. "Petrus Hispanus Lectures, II: Radical Probabilism." Actas da Socieda-
de Portuguesa da Filosofia (forthcoming; and currently available in http://www .
princeton.edu/-bayesway/pu/Lisbon.pdf).
Kemeny, 1. 1955. "Fair Bets and Inductive Probabilities." Journal o/Symbolic Logic 20:263-273.
Lehman, R. 1955. "On Confinnation and Rational Betting." Journal ojSymbolic Logic 20:251-
262.
Lehrer, K. and Wagner, C. 1981. Rational Consensus in Science and Society. Dordrecht: Reidel.
Ramsey, F. 1990. "Truth and Probability." In Philosophical Papers ofF P Ramsey, D. H. Mellor,
ed., Cambridge: Cambridge University Press.
Roberts, F. 1976. Discrete Mathematical Models. Englewood Cliffs: Prentice-Hall.
Shafer, G. 1976. A Mathematical Theory oj Evidence. Princeton: Princeton University Press.
Strassen, V. 1964. "MeBfehler und Information." Zeitschri/i fur Wahrscheinlichkeitstheorie
2:273-305.
Wagner, C. 1989. "Consensus for Belief Functions and Related Uncertainty Measures." TheOlY
and Decision 26:295-304.
1992. "Generalized Probability Kinematics." Erkenntnis 36:245-257.
----. 1997. "Old Evidence and New Explanation." Philosophy ojScience 64:677-691.
- - - . 1999. "Old Evidence and New Explanation II." Philosophy oj Science 66:283-288.
Walley, P. 1981. "Coherent Lower (and Upper) Probabilities." Technical Report, Department of
Statistics, University of Warwick, Coventry, England.
- ---'-. 1991. Statistical Reasoning with Imprecise Probabilities. London: Chapman and Hall.
TRUSTWORTHINESS
Chapter 10
Lehrer calls the keystone of Reid's system, of crucial importance for the way in
which it supports the other elements and itself as well.
The first principles state that the convictions of our faculties are
true rather than evident, but the information that our convictions
are true is the evidence that grounds them. These first principles
are, therefore, principles of evidence as well as principles oftruth.3
I shall take it, then, that the central trait of first principles is their self-evidence.
A cautionary note: I do not necessarily equate Reid's epistemological
principles with first principles. As already noted, there are some first principles
(e.g., axioms of mathematics or metaphysics) that are not epistemological
principles. More significantly, I believe one should be open as well to the
converse possibility that Reid's epistemological principles are not first
principles themselves-that they specify first principles without being first
principles. So the questions I raise in this section and the next about Reid's
epistemological principles are not necessarily questions about principles he
regards as first principles.
With this in mind, let us tum to the question whether the epistemological
principles in Reid's list are principles of truth or principles of evidence. I think
the answer depends on a crucial but little noted scope ambiguity in the wording
of Principle 1 and several other of the principles as well. Here is how Reid
formulates Principle 1:
not be evident for the person because the person had no idea
whether the belief originated in a way that is trustworthy or
deceptive. In fact, the first principles of our nature not only yield
beliefs but also information about those beliefs, to wit, that they
are trustworthy and not fallacious in origin. 1o
If (1) and (2) are both true, skepticism is the inevitable consequence. Clearly,
if we cannot know either of two things without knowing the other first, we must
remain ignorant of both of them.
How would Reid, the great foe of the skeptic, respond to this threat? If
Reid is an externalist, as I have suggested he is at least some of the time, he
would deny KR and with it proposition (l). Consciousness, memory,
perception, and our other faculties give us knowledge of their deliverances even
if we are initially ignorant of their reliability. But Lehrer's Reid is not an
externalist, and it must be admitted that there are places where Reid seems to
accept a requirement like KR. Here is a notable passage:
7thly, Another first principle is, that the natural faculties, by which
we distinguish truth from error, are not fallacious. (EIP 6.5, p.
630)
I shall assume with Lehrer, then, that Principle 7 does indeed concern all
of our cognitive faculties-not just some specially delineated faculty or
faculties.
And yet the question remains: In what way does Principle 7 go beyond
the other principles on Reid's list? Why is it not merely redundant-perhaps
simply a device enabling Reid to make points concerning all the principles by
discussing a single one?
To this question, Lehrer's answer is that Principle 7 is a metaprinciple,
affirming the truth of the other principles, and a looping principle, affirming the
truth of itself as well. In this last way it goes beyond the other principles.
If we take Reid's formulations at face value, however, there does not
appear to be anything particularly "meta" about Principle 7. The sequence
Consciousness is reliable.
Memory is reliable.
Perception is reliable.
All our faculties are reliable.
The last item on each list does not seem to be much more than a summary of
what has gone before. If Principle 7 goes beyond the other principles, it is only
by virtue of implying either that I have no faculties not already mentioned or, if
I do, that they are reliable, toO.28
Nonetheless, the "meta" and "looping" character of Principle 7 may be
brought out more clearly if we rewrite Principle 7 in a way Lehrer has proposed.
Principle 7 may be recast, he suggests, as a principle affirming that all first
principles are true. 29 So construed, Principle 7 does indeed convey what the
other principles convey (that consciousness is reliable because Principle I is
true, memory reliable because Principle 3 is true, and so on). But Principle 7
also conveys more, because it is itself a first principle. It therefore implies its
own truth by way of self-subsumption:
2 + 2 = 4.
Aberdeen is northeast of Glasgow.
Water boils at 212 degrees F.
All the sentences in this list are true.
The concern is that the final sentence is semantically ungrounded, just like the
truth-teller sentence, 'this sentence is true'. If one of the sentences preceding it
were false, that would make the final sentence false. But if all others on the list
are true, whether the last is true comes down to whether it is true, and there
seems nothing to determine that. There is nothing to make it true or false, so
arguably it is neither. The situation seems similar with Principle 7. If the other
first principles are all true, then Principle 7 goes beyond them just to the extent
JAMES VAN CLEVE 167
that it ventures out over the void, with nothing to sustain a truth value for it. So
the worry is that in trying to come up with a keystone principle that explains the
others and itself in the bargain, we obtain a principle that explains nothing
because it lacks truth value.
I do not wish to suggest that there is anything automatically defective
about general principles that subsume themselves. All necessary propositions
are true is itself a necessary proposition, and therefore may be subsumed under
itself. But there is no question about its truth value; it is an accepted axiom of
modal logic. Similarly, everything that God believes is true would be both self-
subsuming and true (indeed, necessarily true) if there were an essentially
omniscient and infallible God who believed it. But it is significant that these
examples of self-subsuming propositions that are unproblematically true are
necessary truths. There is a necessary connection between the properties that
figure in antecedent and consequent, and that is what makes them true. Lehrer,
however, does not think that Principle 7 is a necessary truth. He thinks that
Reid's first principles of contingent truths are themselves contingent. 32 So he
does not have this way of alleviating the worry I have raised that under his
construal of it, Principle 7 is semantically ungrounded.
I shall return to this worry at the end of the next section. It is time to see
how two aspects of Lehrer's interpretation of Reid-the KR requirement on
knowledge and the endorsement of looping principles-are connected.
The problem we left hanging at the end of section 2 was this: if we cannot know
anything through a faculty without knowing that that faculty is reliable (as
required by KR), how can we know anything at all? The problem can be
formulated as a pair of premises that jointly entail skepticism:
When the problem is posed this way, it is clear that we can avoid
skepticism only by denying (1) or (2).
An externalist would deny (1), but Lehrer's Reid is no externalist. So let
us consider the option of denying (2)-of denying that knowledge of the
reliability of a source must be collected from various of the particular
deliverances of that source. If not derived from knowledge of its own particular
168 THE FIRST OF ALL PRINCIPLES
so evidence, which is the voucher for all truth, vouches for itself at
the same time. (EIP 6.5, p. 632) 37
The idea would be that as light discloses features both of visible objects and
itself, so our cognitive faculties disclose features both of their primary objects
and ofthemselves-in particular, their own reliability.
With this answer, we have drawn close to Lehrer's interpretation of
Principle 7. Recall that for Lehrer, Principle 7 undergirds all the others and
itself at the same time. It is a principle that affirms its own truth along with the
truth of the other principles. We are now suggesting that there are faculties that
apprehend their own reliability along with the reliability of other faculties. We
have thus arrived at a view structurally similar to Lehrer's, incorporating the
same looping strategy he favors. And we have been led to do so in the attempt
to show that knowledge is possible even under the tough demands imposed by
KR. It is no accident, then, that a philosopher who, like Lehrer's Reid, holds
that there is no knowledge through a faculty without knowledge of its reliability,
also holds that there are faculties that vouch for their own reliability.
It is now time to return to the worry left unresolved at the end of the
previous section-that self-subsuming principles of the sort Lehrer favors would
be semantically ungrounded. Is there not an analogous worry about self-
authenticating faculties? Suppose that intuition intuits its own reliability-that
is, that I have an intuiting whose content is that all intuitings are true. If all other
intuitings are true, what would make that one true? As before, I think it would
have to be some sort of necessary connection between the properties of being an
intuiting and being true. (If it were simply a matter of the individual truth values
of all intuitings, ungroundedness would threaten.) Is there plausibly such a
connection, and could Lehrer accept it? I noted above that he does not believe
that Reid's principles are metaphysically necessary truths; he would deny that
there is any metaphysically necessary connection between being an intuiting and
being true. But I expect he might allow that there is a nomologically necessary
connection between being an intuiting and being true, and perhaps such a
connection would suffice as a truth-maker for the intuiting that all intuitings are
true. I leave further exploration ofthat possibility for another occasion.
There is a side of Lehrer's Reid I have so far left out of account. As we have
seen, Lehrer's Reid holds that general principles affirming the reliability of our
faculties are evident in themselves and known immediately. But Lehrer also
says that particular deliverances of our faculties, such as the beliefthat there is
a hard object in my hand, are evident in themselves and known immediately.
170 THE FIRST OF ALL PRINCIPLES
"Our knowledge of both the first principles and the particular beliefs to which
they give rise are both immediate .... The [general] principles and [particular]
beliefs of common sense fit together like links in a chain, and he that is not fit
to pick up the whole should not attempt to lift up any of the parts.,,38
Here Lehrer seems to be telling us that general principles and particular
beliefs are both self-evident, but at the same time, that neither generals nor
particulars would be evident unless both were part of our system of belief. Is
that a consistent combination?
Lehrer's Reid is beginning to sound like a coherentist, for whom
particular beliefs and general beliefs become evident together once a wide
enough body of mutually supporting beliefs is in place. 39 Yet coherence theories
are normally taken to repudiate the category of the self-evident, while Lehrer's
Reid asserts that plenty of things are self-evident, general principles and
particular beliefs alike. What are we to make ofthis?
Suppose we say that one belief is epistemically prior to another (or that
beliefs in one class are epistemically prior to those in another) iff the latter
belief(s) derive their evidence from the evidence of the former, in such fashion
that the latter beliefs could not be evident unless the former beliefs were already
evident. Suppose we then say that a belief is self-evident iff it is evident, but
there is nothing epistemically prior to it. We could then say that a coherence
theory that rejects the very idea of epistemic priority is a theory that, rather than
repudiating the category of the self-evident, makes every evident belief self-
evident. Reid, while not going this far, could hold that particular beliefs and
general beliefs depend on each other for their evidence without beliefs of either
sort being prior to beliefs of the other sort; thus particulars and generals would
both be self-evident. This could be one way in which Reid, as Lehrer says,
transcends the dichotomy between foundational ism and coherentism. 40
ENDNOTES
1 Keith Lehrer, Thomas Reid (London: Routledge, 1989), p. 197. This book is cited hereafter as
Reid.
2 Reid, p. 198.
) Reid, p. 157.
4 This abbreviates Essay 6, Chapter 7, of Reid's Essays on the Intellectual Powers of Man. My
page references are to the edition edited by Baruch Brody (Cambridge, Mass.: MIT Press, 1969).
5 Another assumption that would convert lA and the other A-style principles to principles of
evidence is a reliability theory of justification, but Lehrer and Lehrer's Reid both repudiate
reliability theories.
(, Reid, p. 157.
7 James Van Cleve, "Reid on the First Principles of Contingent Truths," Reid Studies, 3 (1999),
3-30.
JAMES V AN CLEVE 171
8 This abbreviates Chapter 6, Section 12, of Reid's Inquiry into the Human Mind. My page
references are to the volume edited by Derek R. Brookes (Edinburgh: Edinburgh University Press,
1997).
9 Reid, p. 198.
IOReid,p.187.
11 Lecture at the NEH Seminar on Thomas Reid, Brown University, August, 2000.
12 Keith Lehrer, "Chisholm, Reid, and the Problem ofthe Epistemic Surd," Philosophical Studies,
Ga'; a confirmation instance of it would be the conjunction 'Fa & Ga' (or perhaps an object that
is both F and G). What Reid and Lehrer mean by an instance is neither ofthese things, but simply
'Ga'.
14 Ernest Sosa offers an analysis of "implicit commitments" along these lines in Ernest Sosa and
James Van Cleve, "Thomas Reid," in The Blackwell Guide to the Modern Philosophers from
Descartes to Nietzsche, edited by Steven M. Emmanuel (Oxford: Blackwell, 2001), pp. 179-200,
beginning on p. 190. For the more radical view that general belief is never anything over and
above such BFx -> BGx dispositions, see David Armstrong, Belief, Truth, and Knowledge
(Cambridge: Cambridge University Press, 1973), ch. 6.
15 Reid may be committed to rejecting this possibility. In his account ofthe system ofLeibniz, he
explicitly disapproves of Leibniz's distinction between perception and apperception. "As far as
we can discover, every operation of our mind is attended with consciousness, and particularly that
which we call the perception of external objects .... No man can perceive an object, without being
conscious that he perceives it" (EIP 2.15, pp. 236-37 in Brody). In another place (EIP 3.1, p.
325), he says that consciousness is always attended with belief of that whereof we are conscious.
Putting these together, it follows that we never perceive without believing that we perceive.
16 Stewart Cohen, "Basic Knowledge and the Problem of Easy Knowledge," forthcoming in
22 "Surd," p. 42.
23 Keith Lehrer, "Reid, Hume, and Common Sense," Reid Studies, 2 (1998),15-25, on p. 15 and
elsewhere.
24 Philip de Bary, "Thomas Reid's Metaprinciple," American Catholic Philosophical Quarterly,
74 (2000), 373-83. He reports in note 22 on p. 380 that in Reid's manuscript of the Intellectual
Powers at Aberdeen University Library, there are no commas surrounding "by which we
distinguish truth from error."
25 I thank Gideon Yaffe and Sue Cox for this suggestion.
26 I thank Alan Hazlett, Nick Treanor, and Ali Eslami for this suggestion.
27 I have in mind the twelfth paragraph of commentary on Principle 7. This paragraph may not be
decisive, however, since Reid says he is noting a property that Principle 7 has in common with
other principles, and it is just possible that the trust we repose in our senses, our memory, and our
reason may illustrate the implicit belief we have in principles other than Principle 7.
28 So that Principle 7 not convey less information than the preceding principles, I am assuming
that a reference to the faculties already mentioned is implicit-'the above-mentioned faculties, as
well as any others that I possess, are reliable'.
172 THE FIRST OF ALL PRINCIPLES
29 Lecture at the NEH Seminar on Thomas Reid, Brown University, August, 2000. Lehrer's actual
formulation was 'all first principles are trustworthy,' where being trustworthy implies both being
evident and being true. The fact that trustworthiness implies truth generates the worry about
ungroundedness that I raise in the text.
30 Ifwe render 'all our faculties are reliable' as 'all deliverances of our faculties are true', however,
we would obtain a principle that is self-subsuming provided it is itself a deliverance of our
faculties.
31 Keith Lehrer, Self-Trust (Oxford: Clarendon, 1997), pp. 22-23.
32 The "first principles of contingent truths" are so called because they make knowledge of
contingent truths possible; this leaves it open whether they are themselves necessary or contingent.
It is clear from Reid, p. 157, however, that Lehrer takes them to be contingent. Moreover, to the
extent that Lehrer construes the principles as attributing trustworthiness and not just truth, he has
a principled reason for regarding them as contingent: he believes that epistemic properties never
supervene with metaphysical necessity on nonepistemic properties. On this point, see Chapter 3
of Self-Trust.
33 "Surd," p. 40, and "Reid, Hume, and Common Sense," pp. 22-23.
34 Here I may disagree with Lehrer, who says at p. 22 of "Reid, Hume, and Common Sense" that
the first principle about perception comes from the faculty of perception itself.
35 Perhaps a good Reidian name for it would be 'common sense', for Reid tells us that "the sole
province of common sense" is "to judge of things self-evident" (EIP 6.2, 567).
36 We see here, by the way, that the Reid who gets around the skeptical impasse by denying (2)
must also deny (1). Though he accepts KR, he denies that to have knowledge through K you must
have knowledge of K's reliability first. Otherwise, there could be no such thing as basic
knowledge of the reliability of a source. Rather, there are cases in which knowledge through K
and knowledge ofK's reliability arise simultaneously.
37 Lehrer has drawn on this passage for a somewhat different purpose. He poses a question about
Reid from Chisholm: "How can we tell that a belief is evident if not by appeal to a general
principle?" He cites the paragraph about light as Reid's answer, glossing it as "the evidence of
some beliefs is itself evident.... Ifthere are some beliefs whose evidence is evident to us, we have
no need for a criterion to pick them out as evident" ("Surd," p. 41).
38 "Surd," p. 40; see also Reid, pp. 199-200.
39 In the article cited in footnote 16, a coherence theory is Cohen's way around the impasse created
by (1) and (2). In effect, he denies both of them, holding that knowledge of the reliability of a
source and knowledge of the truth of its particular deliverances depend on each other without
either being prior to the other.
40 "Reid, Hume, and Common Sense," p. 16.
Chapter 11
G. J. Mattey
University a/California, Davis
2. REASONABLENESS
What does it mean to say that it is reasonable, to some degree, for a person
to accept the information that p to fulfill the epistemic purpose of obtaining truth
and avoiding error? Lehrer treats reasonableness as a primitive notion, though
he does note a relation between reasonableness and the epistemic purpose. For
the information thatp to be reasonable (to some degree) to be accepted, it must
be subjectively probable to a certain degree, which promotes the goal of avoiding
error. Conversely, accepting information that p is made more reasonable as it is
more informative, which promotes the goal of obtaining truth ([6], p. 144-145).
Typical of somewhat risky, but highly informative, information are "major
scientific claims, those concerning galaxies, genes, and electrons" ([6], p. 145).
How reasonable it is to accept epistemically a piece of information would
seem on the face of it to be a complex matter, which is perhaps not easily
determined. Lehrer sidesteps this issue by simply assuming that "we are able to
tell, at least intuitively, when it is more reasonable to accept one thing than the
other" ([6], p. 128).11 This allows him to make reasonableness the determining
factor in any evaluation that results in the acceptance of the information that p
to fulfill the epistemic purpose. "I confront the question of whether or not to
accept some information that I receive," and I answer the question on the basis
of "how reasonable it is to accept the information in comparison to other
competing considerations" ([6], p. 126).
G. J. MATTEY 177
3. TRUSTWORTHINESS
not be very reasonable to accept anything unless one accepted that one is
trustworthy in what one accepts in general.
On the interpretation of trustworthiness as competence, the principle
means that in accepting what I do in general, I exercise competence in fulfilling
the epistemic purpose of acceptance. In that case, the reasonableness of principle
(T) would be supported by acceptances about one's overall record of one's
success in everything one accepts. It is reasonable to accept that I am generally
worthy of my own trust to the extent that I accept that I have earned that trust,
so to speak.
Lehrer has a second way of understanding trustworthiness: as a
deontological notion, "an irreducible element of epistemic value" ([5], p. 138).
He describes it as "a notion of what is worth accepting and what methods are
worth using" ([5], p. 138). In his account of the normative dimension of
trustworthiness, he divorces it entirely from considerations of actual competence
and success. His purpose in so doing is to accommodate the intuition that it is
reasonable to accept what one does even if one is the victim of massive
deception. Though Lehrer does not make this point, it is clear that such an agent
would then be completely incompetent, not merely unsuccessful, in fulfilling the
epistemic purpose. Since, reasonableness requires acceptance of trustworthiness,
Lehrer wants to say that a victim of deception may nonetheless be trustworthy.
"I am worthy of my trust in what I accept though I am deceived. I am as
trustworthy as the circumstances allow" ([6], p. 140).
If worthiness of one's trust in acceptance does not require actual
competence in fulfilling the epistemic purpose, what does it require? Lehrer
casts himself in the role of a hypothetical demon-victim and describes himself
as being deceived "through no fault of my own" ([6], p. 139). Being worthy of
one's own trust, on this deontological construal, is a matter of having followed
certain standards in searching for the truth. As Lehrer puts it regarding the
demon case, "I seek to obtain truth and avoid error with the greatest intellectual
integrity" ([6], p. 140). Similarly, one is trustworthy when one is "circumspect
and seeks to detect every error" ([6], p. 192).
Trustworthiness, viewed deontologically, is the result of the use of a
general method of approaching acceptance, in the exercise of which one takes on
objections forthrightly, meeting them when one can and changing one's view
when one must. It also requires the willingness to change one's methods of
getting at the truth if need be. In general, Lehrer says that his trustworthiness
"rests on a dynamic process of evaluation and amalgamation of information I
receive from others and from my own experience" ([6], p. 140).
Lehrer's example of trustworthy but unreliable acceptance is described in
a way that suggests that the agent has, in general, done her best to fulfill the
epistemic purpose. But it is psychologically unrealistic to assume that there is
a constant level of circumspection applied in every act of acceptance, so in
general, one's acceptances will fall somewhat short of this standard. If
G. J. MATTEY 181
Since there is no restriction on the value ofp, the conclusion must be taken to be
generalizable to all acceptances.
Lehrer notes that the first inference is meant not to be deductive, but rather
inductive. That is, the first premise is not intended to be a universal
generalization "to the effect that I am always trustworthy in what I accept" ([6],
p. 139). Instead, it is supposed to be taken as a claim to the effect that I am
generally trustworthy in what I accept.
The most plausible construal of the description of the other person's acceptance
of my trustworthiness is that she uses it, along with the fact that I accept some
information, as a factor in evaluating that information and making a decision to
accept it. But in that case, the analogy breaks down, since I cannot make what
I already accept a factor in my deciding to accept it. My acceptance of my own
trustworthiness can playa role in my deciding what to accept, in that without it,
I might be disposed to withhold judgment rather than accept any information at
all. It will be argued below that this generic way in which trustworthiness makes
accepting reasonable is also the only way in which it makes what is already
accepted reasonable.
Given the interpretations ofthe two premises, the first conclusion must be
read in this way: I have accepted that p on the basis of good-enough methods for
obtaining truth and avoiding error in the acceptance of p. In the context of the
trustworthiness argument, what those methods are is immaterial. Since I am
trustworthy, the fact that I accept that p means that I have (most likely) relied on
those methods I deem fit to make my acceptance a correct one.
Now we can see how the first conclusion supports the second one: why
trustworthiness entails reasonableness. It has been noted that Lehrer assumes
that one can tell how reasonable it is to accept a given piece of information,
relative to one's evaluation system. Presumably this means that one can
determine how suitable its acceptance is to advance the epistemic purpose. Then
the idea would be that if I use the methods I deem to be good enough for
fulfilling the epistemic purpose in accepting that p, then I should regard the
purpose as being fulfilled. Lehrer states the relation this way: "My
trustworthiness serves the objectives of reason, and if I am trustworthy in the
way I serve the objectives of reason in what I accept, then I am reasonable to
accept what I do" ([5], p. 136).
So the thrust of the whole argument is this. IfI am disposed generally to
use good-enough methods in accepting what I do, and I accept some piece of
information p, then I can conclude inductively that I have used good-enough
methods in accepting that p. If I have used good-enough methods for accepting
that p, then my acceptance that p is reasonable, to some degree.
Ordinarily, when we evaluate the reasonableness of acceptance, we take
into account the specific methods which are applicable to the specific
G. J. MATTEY 185
information in question. The statement "I am reasonable in accepting that p" can
be understood in two very different ways. It can be, and ordinarily is, read as a
statement about the reasonableness of accepting the specific information p. Or
it can be read as a statement about accepting any information at all, regardless
of its specific content. It is only the latter sense that could possibly be
established by the "trustworthiness argument." It is only in this sense that Lehrer
could be entitled to assert that, "A consequence of adding principle (T) to my
evaluation system is that I may reason from it and the acceptance of some target
acceptance that p to the conclusion that the target acceptance is reasonable" ([6],
p. 139).
A more specific counterpart to the generic "trustworthiness argument"
might be one to the effect that one is disposed to be circumspect one's
investigations and that those methods of investigation sanction the acceptance
of the specific information p, so that it is reasonable to make a commitment to
the truth of p. One would expect that the first premise would be established
inductively. The second premise would be established by appeal to the specific
evidence in favor of accepting that p. The original "trustworthiness argument,"
on the other hand, says nothing about what makes it reasonable specifically to
accept that p rather than some other information. So whatever degree of
reasonableness it establishes is minimal compared to that established by the
counterpart argument.
Suppose an ordinary person were to ask me why it is reasonable for me to
accept that the water in my glass is safe to drink. If I were to respond, "Because
it is something that I accept, and I use good-enough methods to accept what I
do," my response would most likely be met with bewilderment. On the other
hand, if a foundationalist like Chisholm were to ask this question in the context
of his epistemological investigations, the answer would make sense, since he
then would be concerned with the source of reasonableness as such.
The "trustworthiness argument" is really appropriate only in the context
in which Lehrer raises it, i.e. as a response to a foundationalist objection.
Justification depends on the reasonableness of what one accepts, and
reasonableness depends on the acceptance of trustworthiness. "The
foundationalist will surely note that everything now depends on the claim that
my acceptance is a trustworthy guide to truth and that I am trustworthy, as I aver.
She will inquire how that claim is itself justified" ([6], p. 138).
The foundationalist inquiry can be extended to the issue of reasonableness:
what makes it reasonable for me to accept that I am trustworthy? What gives that
acceptance what Chisholm called "substance and rigidity?" Though appeal may
be made to competence and success, the following response has to be given:
"when I accept something, that is a good enough reason for thinking it to be true,
so that it is reasonable for me to accept it" ([6], p. 138). Again, "If! accept that
I am trustworthy in this way, then my accepting something will be a reason for
me to accept it" ([6], p. 138).
186 THE REASONABLENESS OF ACCEPTANCE
Lehrer offers two accounts of what makes principle (1) reasonable, both
of which require that it contribute in some way to its own reasonableness. On one
account, the contribution is indirect; on the other it is direct. The first account
will not be plausible to someone who rejects all circularity in the relation of
evidential support. The second account is burdened with its own variety of
circularity and has additional problems of its own.
In the first account, the claim that one is generally disposed to be
trustworthy in acceptance is supported by an inductive generalization. The
starting-point is the trustworthiness of most of one's specific acceptances and the
conclusion is that one is generally disposed to be trustworthy in acceptance.
He opts for this small circle over the larger circle because it allows him to
avoid a regress in explanation. "I could argue for my trustworthiness by
consideration of other things I accept and my success in attaining truth, but that
way a regress threatens, whatever the merits of such arguments in supporting the
principle" ([5], p. 136). Butthere is no regress when the other things one accepts
are made reasonable in part by the acceptance of one's trustworthiness, so we are
not forced to apply the principle to itself to account for the reasonableness of
acceptance. Whether the small circle actually explains anything remains to be
seen.
In the second edition of Theory ofKnowledge, published in 2000, Lehrer
does not mention the regress argument, and as seen above, he explains the
reasonableness of accepting (T) by appeal to its concurrence with the whole
acceptance system. Still, though Lehrer does not defend the use of principle (T)
to explain its own reasonableness without appeal to any other information, he
does endorse the direct application of (T) to itself. What reason is there for doing
this, except as a formal exercise? Lehrer notes that the argument is "more direct"
than one using an inductive argument from individual cases of trustworthy
acceptance ([6], p. 142). This directness has the advantage of economy, but it
gains this advantage at the expense of content.
It is useful to note that Lehrer recognizes that he is not making the
stronger claim that principle (T) is completely self-justifying, though "the
principle of our own trustworthiness contributes to its own personaljustification"
([6], p. 202). It does not justify itself fully because it "must cohere with what we
G. J. MATTEY 189
accept about our successes and failures in past epistemic employment" ([6], p.
202). In that case, one must ask why Lehrer restricts this requirement to
justification. Can we plausibly say that it is reasonable to accept a piece of
information without regard to whether it coheres (or "concurs") with information
we have about our past record of success, among other things? The fact that
reasonableness (to some unspecified degree) need not meet the standard of
justification does not exempt it from the need for a comprehensive base of
support.
This consideration raises the more general question of what kind of
explanation could be provided by the direct self-application of principle (1). I
want to know why it is reasonable for me to accept that I am trustworthy in what
I accept. In terms of the interpretation of trustworthiness developed thus far, the
question is why it is reasonable for me to accept that I have a disposition to use
good-enough methods in accepting what I accept. The obvious indirect
explanation for this is on the basis of what I accept about how I have used good-
enough methods in the past. The indirect loop is generated by adding that part
of what makes those acceptances reasonable is the acceptance of my own
trustworthiness.
The direct explanation is simply a vastly diminished version of the indirect
explanation, one which omits all the details that enlighten me as to why it is
reasonable to accept that I am trustworthy. The result is hardly edifYing.
Suppose I were testifYing to a jury and averred that I am trustworthy in my
testimony. When asked to explain why it is reasonable to accept that I am
trustworthy, I answer that my original statement that I am a trustworthy witness
is what explains why it is reasonable for me to accept that I am. Would I have
explained, to anyone's satisfaction, anything at all?
Another way to put the point is by noting what kind of reasonableness is
supposed to be explained. As was noted above, we can ask, for any given piece
of information (P), whether it is reasonable to accept that (P) as having a given
content, or whether it is reasonable to accept that (P) in the sense that it is
reasonable to accept what we accept in general. And as has been argued earlier,
the kind of reasonableness established by the "trustworthiness argument" is
generic. So when principle (1) is the target acceptance, the most the application
of the trustworthiness argument to (1) can explain is that it is reasonable for me
to accept that (1) insofar as it is reasonable for me to accept anything at all. So
the direct self-application of principle (1) does nothing to explain why it is
reasonable to accept information with the specific content that I am trustworthy
in what I accept.
Lehrer might respond to this description ofthe thinness of the explanation
by claiming that any explanation is better than none. The principle itself is part
of the account of the reasonableness of any acceptance, and so if it were not
reasonable to accept principle (1), there would be no explanation at all. In that
case, principle (1) "should be a kind of unexplained explainer that explains why
190 THE REASONABLENESS OF ACCEPTANCE
Lehrer might also have noted that the keystone would fall to the ground if the
other stones were removed. The keystone supports itself only through its support
of the other stones. A keystone is not a foundation stone. So this analogy, if it
has any value at all, favors an account of the reasonableness of accepting the
principle of trustworthiness in which the principle supports itself indirectly. In
general, it favors the "broad concurrence" account of the reasonableness of
acceptance.
In summary, there seems to be nothing favoring the direct application of
(T) to itself other than the fact that it can be made and that it simplifies the
response to a skeptical objection to the reasonableness of what one accepts. But
in fact it is no response to a skeptic, and if it explains anything at all, it explains
only how it is reasonable to accept to some extent, merely as something that is
accepted in general. Even this explanation is largely incomplete and can only be
completed by appeal to a large number of other acceptances. Finally, there is no
explanation of why the specific content of (7) is reasonable to accept.
All of these deficiencies disappear when "broad concurrence" is invoked
to explain what makes principle (T) reasonable to accept epistemically.24 So the
direct self-application of (T) appears to be a useless exercise. It might even do
some harm by engendering the illusion that the principle of trustworthiness is
foundational rather than a "first among equals." Given the argument of this
paper, Lehrer has no reason to make his "ecumenical" concession to
foundationalism ([6], pp. 201-3).
6. CONCLUSION
ENDNOTES
1 This seems to be what Lehrer calls mere belief. which "arise[s] in us naturally without our
bidding and often against our will" ([6], p. 40).
2 Lehrer acknowledges in his "bear print" example that sometimes circumstances dictate a greater
degree of scrutiny than in normal circumstances ([6], p. 73). Complex or highly general
information would also call for reflection.
] Ifreflection is not involved, there is no "decision" strictly speaking, but a commitment must be
made in a manner analogous to the making of a decision.
4 This kind of comparison was made by Chisholm in [1].
5 It would be fortuitous from the point of view ofthe agent. There might be some sort of external
factor that makes the correctness of the decision non-accidental, as in the case of a device
implanted in the brain that brings about correct acceptances. See [6], p. 186-8.
6 See [2], pp. 192-197.
9 The "justification game" illustrates the way in which critical objections might be handled ifthey
were to arise. See [6], pp. 132-128.
]0 This feature of acceptance is highlighted by theories of prima facie justification.
11 Presumably, one would be able to tell as well whether it is more reasonable to accept than to
withhold acceptance.
12 This is less evident with respect to informativeness than with respect to probability, but very
general information can be completely uninformative to a person who does not have the
conceptual resources to integrate it into his view of the world.
13 The other components are the "preference system" and the "reasoning system." See [6], pp.
126-127. In [5], Lehrer notes that only the acceptance system is relevant to the issues that he raises
there, and these are the issues discussed in the present paper. See p. 138, note 2.
14 It appears to be psychologically unrealistic to assume that an epistemic agent can properly
distinguish between what he already believes and what he has already accepted. One reason is that
we often forget how we came to take a given piece of information to be true.
15 The information contained in the acceptance system is also the basis for the determination of
justification.
16 This term is taken from Chisholm, who attributes the concept to the ancient academic skeptic
Carneades ([1], p. 43).
17 It may be that in many, most, or all cases, concurrence involves dealing with objections, in
which case it is closely related to justification.
18 See [1], Chapter 3.
19 Since following good-enough methods requires evaluation of our own competence, we will
21 Lehrer allows that it is plausible that "I believe something" is a self-justified belief. See [6], p.
54. See also p. 67-8, where he writes that "fallibility infects almost all our beliefs" (emphasis
added).
22 In the first edition of Theory of Knowledge, Lehrer had called the self-application of (T) "more
natural" than trying to avoid the self-application for fear of self-referential paradox (p. 123).
23 See [6], Chapter 3.
24 This is not to say that these advantages make "broad concurrence" a convincing alternative to
foundationalism.
194 THE REASONABLENESS OF ACCEPTANCE
REFERENCES
Roderick Chisholm. Theory o/Knowledge. Prentice Hall, 1966.
Keith Lehrer. Knowledge. Oxford University Press, 1974.
Keith Lehrer. A Theory 0/ Knowledge. Westview Press, first edition, 1990.
Keith Lehrer. Self Trust: A Study o/Reason, Knowledge and Autonomy. Oxford University Press.
Keith Lehrer. Knowledge, scepticism, and coherence. Philosophical Perspectives, 13: 131-139,
1999.
Keith Lehrer. Theory o/Knowledge. Westview Press, 2000.
Chapter 12
Richard N. Manning
Carleton University
not require justificatory support from other claims. That is, according to
foundationalism, there are basic claims which are non-inferentially justified.
To suppose that there are such non-inferentially justified claims is not, of
course, to suppose that agents do or even could have some good reasons for
believing of any particular claim, or any class of claims, that it is so justified.
Indeed, the foundational status of a belief means that no such belief could owe
its justificatory status to anything the agent possesses as a reason. To suppose
otherwise would reinstate the impotent regress of justification that
foundationalism is in part designed to avoid. On the foundationalist picture,
then, not every belief that isjustified need be justified by an agent's reasons. As
William Alston notes, all that is needed to stop the regress of justification is
belief which is immediately justified, whether or not the agent has reason to
believe that it is so justified. l Faced with the natural objection that this would
mean that ajustificatory chain could terminate in a claim which the agent has no
reason to hold true, that from the agent's point of view this claim would amount
to a brute assumption, Alston argues that there is no bar to adding the additional
requirement that the agent have a justification for supposing the foundational
claim immediately justified, but that this justification need not (and cannot)
itself be immediate, but must be inferential. As it is a reason for holding the
putative basic claim justified, it is moreover an epistemic reason in the sense of
being a second order claim about the justificatory status of a first order claim.
(More generally, a claim of order n about the justificatory status of some claim
or claims oflevels ml, m2, ... mi<n.) But the inferential grounds an agent has or
needs for supposing a claim immediately justified-his epistemic reasons-
cannot form an essential part of the justification of the claim, lest they
compromise its foundational status. Thus the epistemic reasons a person might
have for holding a belief are on this view distinct from the justification of that
belief, which is non-inferential.
This divorce of justification and reason is precisely what the
epistemological coherentist rejects. For the coherentist, an agent's reasons for
supposing a claim to be justified just are that agent's justification for holding the
claim. As these reasons must be inferential, the claim's status is at best one of
being inferentially, mediately supported. There is no room in the coherentist
picture, then, for claims which are justified for an agent by anything other than
the inferential reasons the agent has for holding the claims to be justified. To
accept this is to accept the core of coherentism. It is also to reject the possibility
of exploiting the foundational structure that at least made the truth-
conduciveness problem seem solvable.
With this in mind we can characterize the central tenets of epistemic
coherentism as follows. First, all justification is inferential. Second, nothing
can confer justification without itselfbeingjustified. Third,justification cannot
be achieved by an indefinitely long inferential regress. Fourth, the only things
RICHARD N. MANNING 197
that can justify a belief are further beliefs? As we have seen, coherence accounts
of justification do not number among their virtues a logical structure obviously
suited to the solution of the epistemic regress, and hence the truth conduciveness
problem. Moreover, given the fourth condition, which holds that belief-belief
relations are exhaustive of justificatory relations, the truth conduciveness
problem has an immediate intuitive grip. Truth, after all, is a relation (or so we
tend to think) between beliefs, among other things, and the world, not between
beliefs and other beliefs. And why should we suppose that the first kind of
relation is guaranteed on the basis of the second, on any particular account ofthe
belief-belief relations that comprise coherence? Absent a good answer to this
question, it is hard to see how belief-belief relations, so conceived, can be
epistemic at all. So, despite the considerable intuitive appeal of the notions that
it is the business of rationality to seek reasons for all claims, and to seek the kind
of systematicity, unity, comprehensiveness, and explanatory power implied in
the notion of coherence, coherence accounts of justification have continually
foundered against the truth conduciveness problem. We can see this from the
fact that the other main objections usually raised against coherence accounts are
such that a solution to the truth conduciveness problem would go a long way
toward mooting their force. Consider, for example, the objection that a fully
coherent system of claims (including what seem to be empirical claims) might
be utterly cut off from the reality it purports to capture; it is hard to see how this
so called 'isolation objection' could get a grip, if it were satisfactorily
demonstrated that the contents of such a system are likely to be true. Surely the
demand for a connection between world and system is motivated by the sense
that, absent such a connection, the system's contents could not be true of the
world. So a demonstration of truth conduciveness, whether it invokes such a
connection or not, disarms the threat that motivates the objection. Consider also
the objection that there might be many possible coherent systems, but only one
world for them to be about, or only one true such system. An adequate solution
to the truth conduciveness problem would moot this objection too, either by
showing that only one such system is possible, if truth is univocal, or by showing
that truth is not univocal, if many equally coherent systems can be constructed
by truth conducive means? The task for any coherentism, then, is to show that
beliefs which cohere with one another in some specified sense are likely to be
true, in a way that makes one whose beliefs are coherent in that sense
epistemically justified in her beliefs.
To understand the approach Lehrer takes to this task, it will help to
contrast it with alternative coherentist approaches to the same difficulty.
Coherence would obviously conduce to truth if truth were identified with
coherence. But, as Laurence Bonjour points out, the truth conduciveness of
justification on a particular conception cannot count in its favor as compared
with other such conceptions, if our view of truth has been tailored, ad hoc, to
198 THE DIALECTICAL ILLUSION
guarantee that result. The same move would be open to proponents of any view
of the nature ofjustification, and cannot help us to decide between them. Given
this constraint, Bonjour despairs of the kind of demonstration of truth
conduciveness I discuss here, one made from entirely from within the coherent
system. His own approach to the problem is to attempt an a priori demonstration
that coherent systems of belief meeting specific conditions are more probably
true than false. This approach cannot work in the context of a pure coherentism,
since it depends upon recourse to a priori judgments of comparative probability
which are not themselves justified on coherence grounds. Bonjour recognizes
this and consequently presents a coherence theory of empirical knowledge only.4
But perhaps it is possible, after all, to find an independent, non-trivial
basis for identifying truth with coherence. This was the route taken by the
classical coherentists, on whose view reality is a unified whole and claims-
beliefs contents or propositions-are abstracted parts of that whole which
necessarily capture it in a merely partial way. The more comprehensive and
coherent a set of such abstacta is, on such a view, the more closely it captures
concrete reality itself, since the unification of comprehensive abstracta through
coherence is just the reverse of the process of abstraction itself. Such views are
out of favor in this predominantly atomistic age. Be that as it may, the general
program of attempting to provide non-trivial grounds for identifying truth and
coherence is still live. In his A System of Pragmatic Idealism S, Nicholas
Rescher offers such an argument. Truth conceived as adequation to fact, is,
according to Rescher, identifiable with the notion of optimal coherence with a
perfected data base, where a data base is conceived as a set of candidates for
truth. Whatever else may be said about this argument 6 , the highly idealized
character of the conditions under which it concludes that coherence yields truth
stand out. There is no pretending that we now have belief systems which are
optimally coherent in Rescher's sense, and we are of course never presented
with a perfected data base. Ironically, Rescher's approach is on this score
similar to Bonjour's own effort to show that the stable coherence of empirical
belief systems meeting certain requirements is in the very long run destined to
yield truth. Whether or not idealization of this kind is appropriate in this context,
Lehrer's approach makes no such appeals to the ideal, intending instead to show
that concrete agents have a reason to think their current, certainly less than
comprehensive, beliefs nonetheless I ikely to be true in virtue oftheir coherence,
again less than ideal, with one another.
One way to attempt to show that an agent's coherent beliefs are likely to
be true under the normal, less than ideal conditions with which we situated
humans are faced is to approach the matter through an analysis of meaning. On
such a view, the truth of belief is built in through the relations with the world in
virtue of which the very content of belief is fixed. The approach at which I am
hinting is Davidson's. It invokes its own idealizations, thought in an entirely
RICHARD N. MANNING 199
different spirit, and is not without its difficulties, which I will not pursue here. 7
Lehrer's approach, however, declines to see epistemology in the mirror of
meaning, taking the contents of belief to be fixed, apparently, independently of
any relations to the world which would ensure their likely truth. The contents
of belief are what they are; the truth of belief is another, independent matter.
If strictly all of the beliefs in a given system must be justified on the basis
of coherence, and if the external connections that fix the contents of belief are
put out of play in addressing questions of epistemic warrant, then there is
nothing left to do but to seek the reasons for supposing coherence truth
conducive in the contents of the members of the belief system themselves. We
are left with the hope that beliefs about the probable truth of our beliefs can
themselves be justified by their coherence with the very beliefs whose probable
truth they assert. As our reasons will have to be reasons for holding a belief
justified, they will be epistemic in the sense mentioned above, higher-order
beliefs about the justificatory status of lower-order beliefs. We are to pick
ourselves up to truth by our own epistemic bootstraps. Such is Lehrer's
approach.
For Lehrer, knowledge is undefeated justification, where justification is
defined in terms of coherence with a background system. Lehrer offers the
following definition:
Acceptance, for Lehrer "is a sort of positive attitude toward a content, resulting
in employment of the content as background information in thought and
inference."9 Acceptance is thus epistemically oriented belief, individuated
functionally.
Condition (iii) is concerned with the general structure ofjustification and
the conditions under which a person is personally-subjectively- justified in
accepting some content. For Lehrer, justification is coherence with a
background system of acceptances. A subject S's acceptance system is the set
of all statements of the form "S accepts that ... ," attributing to S just those
contents S accepts in Lehrer's sense. S's acceptance that p is justified if and
only if p coheres with her acceptance system.
Lehrer takes the notion of comparative reasonableness as a fundamental
primitive. A content coheres with an acceptance system if it is comparatively
more reasonable for a subject to accept that content than to accept any
competing claims, relative to that acceptance system. A claim c competes with
a content p just in case it is more reasonable for S to accept that p on the
200 THE DIALECTICAL ILLUSION
assumption that c is false than on the assumption that c is true. That content p
coheres with S's acceptance system if it beats or neutralizes all competitors. p
beats c if it is more reasonable for S to accept p than c on the basis of S's
acceptance system. c is neutralized for S if there is some n such that the
conjunction (c&n) is as reasonable to accept on the basis of S's system as c
itself, and (c&n) does not compete with p. Consider the following abstract
illustration.
Suppose a subject S avers some content q. An interlocutor, call him D,
contests this claim, saying "not-q". Obviously it is more reasonable for S to
accept that q on the assumption that this counterclaim is false than on the
assumption that it is true. Thus, D's claim competes with S's. Is S justified in
his averral? This depends on whether the counterclaim is beaten or neutralized
on S's acceptance system. Suppose S's acceptance system also contains the
following: "S accepts that r", and: "S accepts that r implies q". It is tempting to
conclude that since D's claim that not-q is logically inconsistent with the
entailments of these contents of S's acceptances, S's claim beats D's and is
justified. But this would be premature. Nothing present in S's acceptance
system bears on the reasonableness of accepting q over its competitor not-q. To
see that this is so, simply note that the members of S' s acceptance system are not
the contents accepted by S but rather statements of the form "S accepts that ..
" And it is relative to members of S' s acceptance system rather than to
contents accepted by S that competitors must be beaten or neutralized. Indeed,
this point is important for Lehrer. If competitors had to be beaten relative to the
contents accepted by S, rather than to the members of the acceptance system of
S, then accepted contents would trivially beat competitors. For it is always more
reasonable to accept some p on the basis of p than it is to accept some
competitor to p (e.g., not-p) on the basis of p. Moreover, it is precisely the
presence of the acceptance operator that distinguishes the relata of the coherence
relation as doxastic contents, as opposed to, say, facts or propositional contents
not believed; it is what makes Lehrer's coherentism doxastic.'o Now the
members of S's acceptance system so described, "S accepts that r" and "S
accepts that r implies q", bear no inferential or other rational relation to either
q or not-q. How, then, is D's competitor to be beaten relative to S's acceptance
system? Acceptance has the goal of obtaining truth and avoiding error. Hence
the presence of"S accepts that q" in S's system makes it less reasonable for S
to accept that not-q only if S' s acceptance of those contents is a reason for him
to suppose them more likely true than not. That S accepts some contents,
without more, is not a reason for him to suppose them true. Thus subjective
justification is, to this point, unable to get off the ground.
Before completing Lehrer's account of how justification might get
airborne, let us first briefly turn to his fourth condition. This condition, which
requires that a person's justification be "undefeated", is designed to help avoid
RICHARDN. MANNING 201
acceptance ofT is supposed, then, to give one a reason for supposing that one's
warrant is veridical as well as personal, and thus personal warrant is to be made
epistemic. So if Lehrer's principle T works at all, it bootstraps acceptance to
likely truth, turning coherence with what is accepted into epistemic warrant.
But does it work? Recall that an acceptance system is comprised of
statements of the form "S accepts that p." Lehrer calls Principle T a principle
of detachment: he holds that one who accepts T is thereby entitled to detach the
contents of her acceptances from the fact that she accepts them. This in turn
permits those contents themselves to justify one another and to beat competitors
in appropriate cases. But this detachment process fails, as the following
argument shows.
Let x, y, ... be variables ranging over propositional contents P, Q, R, ..
. Introduce operators "As", such that "AsP" means "S accepts that P," and
"reass", such that "reassP" means "S's acceptance that P is reasonable." Let P
be some arbitrary content accepted by S.
(1) AsP
(2) AsT
T: (X)(Asx -7 reassx).
Assuming that the acceptance operator does not create an opaque context 15, then
(7) says that S accepts that his acceptance ofP is reliable. But this is not
what S needs. That S accepts that P is reliable is not sufficient to show that his
acceptance of P is reliable. Even with the strong logical apparatus we have
supplied, acceptance of Principle T has failed to help detach P from its
acceptance operator. Thus P is no more reasonable to accept than its denial
relative to S' s acceptance system. As Davis and Bender put it, information
about what S accepts is not "adequate to drive the engine of justification." 17
What is required is access to the content p itself. Clearly, if T itself
appeared in the acceptance system of S, then the necessary detachment would
be possible. But T does not appear; the acceptance of T appears, and this
acceptance is as impotent to detach itself as it is to detach other contents:
That is, S' s acceptance ofT is reasonable. But a glance at (12) shows that
it is not of the form "S accepts that P" and thus cannot be a member of S' s
acceptance system. The engine has no spark.
The foregoing objection may appear overly technical, relying as it does
on the precise definition of the acceptance system. This reply is inapt, for the
exclusive use of doxastic states qua doxastic states in justification is of the
essence for coherentism, and on Lehrer's view, it is the presence of the
acceptance operator which marks a content as a member of an agent's
acceptance system. Moreover, as we saw, justification for accepted contents
would be trivial matter but for the presence ofthat operator. Be that as it may,
it is clear that if justification is ever to get started, Lehrer needs some ground for
including (12) or Principle T itself in S's acceptance system, free of any
acceptance operator.
Though Lehrer does not recognize that T is impotent as a principle of
detachment, and thus does not directly address whether it could appear in an
acceptance system without the acceptance operator, he does emphasize that T
is a "special," "make or break" principle. Perhaps what he says in this context
may show that T merits this special status.
accept that such circular justifications do not increase the antecedent likelihood
or reasonableness of the justificandum. This conflict might be resolved by
recourse to the idea that this specific acceptance is a competitor to the claim that
T justifies itself, and that this competitor can be neutralized by being conjoined
with the claim that this case of circular justification is benign. That is, while
competes with
does not. But in order for c&n to neutralize c it must be as reasonable to accept
c&n as it is to accept c itself. But an acceptance can only be reasonable if the
content is more probable than its denial, relative to the acceptance system. This
is precisely what circular justification, generally and in the case of Principle T,
cannot show. Since the general prohibition on viciously circular justification
clearly serves the interest in obtaining truth and avoiding error, c&n can work
to neutralize conly ifTR is treated as an exception to the general prohibition on
circular reasoning. But to grant an acceptance such an exemption seems utterly
ad hoc. Now it is clear that, as a functional matter, we do trust ourselves as
evaluators oftruth. But that is not to say that we in fact exempt the acceptance
of our reliability from the standard proscription against circular justification. It
would only follow that we exempt self-trust from this proscription if we
appealed to self-trust as another acceptance exploited in an inference or
argument for its own validity; and this is exactly what this neutralization
involves. If that is how T achieved its special status, it would be at the cost of
making us ad hoc, bad reasoners. But, as we shall see, general self-trust need
not be conceived as an acceptance exploited in justificatory inference.
Since Principle T is not justified as a foundation, and the regressive and
circular justifications are unacceptably vicious, there is no reasonable ground for
permitting Principle T to appear in one's acceptance system free of the
acceptance operator. Consequently, the principle cannot operate as it must on
Lehrer's account if we are ever to be justified in our acceptances. That we think
we are trustworthy does not make what we think worthy of trust. Consequently,
Lehrer's coherentist solution to truth conduciveness problem fails. Including an
epistemic bootstrap among the relata of the coherence relation does not render
that relation epistemic.
In Self Trust, a comparatively recent work, Lehrer seems more aware of
the futility of a bootstrapping demonstration of the reasonableness of our
208 THE DIALECTICAL ILLUSION
benefit of Archimedes' one fixed point. In the economic sphere, however, some
people are credited with pulling themselves up from destitution "by their own
bootstraps", and we admire them for turning the trick. But they do not really do
so. They use their labor, skills, wits and guts, each of which can help them rise
only if the surrounding world, social and other, complies by valuing and
rewarding them. I think a lesson can be drawn from each of these images
associated with the notion of bootstraps. First, the perceived need for such
desperate epistemic efforts only arises when we think of ourselves as utterly
unmoored, unanchored, unconnected to what our beliefs are about. But this
supposition is false, as the phenomenon of empirical belief itself shows. Second,
we can expect that our own efforts in the epistemic realm, however sincere,
coherent, and strenuous they might be, will not achieve their valuable ends
without the help of what is outside us: the world and other inquirers. As the
related phenomena oflanguage and thought show, we are not without such help.
ENDNOTES
* Thanks to Arthur Fine and Meredith Williams for especially valuable comments on earlier
versions of this paper.
1 William P. Alston, "Two Types of Foundationalism", pp. 76-94, in Empirical Knowledge, ed.
view. So Donald Davidson, a coherentist, more or less identifies coherentism with the fourth
thesis, and John Pollock rejects coherentism precisely because, in his view, the fourth thesis must
be false. But it is worth asking whether the only possible coherentism is doxastic coherentism.
It is perhaps possible that some factive states with empirical content, such as takings-in of
perceptual, memory, and perhaps even testimonial contents (i.e., seeing, remembering, and being
told that p), are involved in inferential justification. The obvious problem with this line is that, by
the second thesis, in order for such factive states to provide justification, they must themselves be
justified inferentially, and it is hard to see how factive states, qua factive, could need or bejustitied
by inference. So what is clear is that the contents of such states could not appear in justifications
qua [active, but only qua accepted contents. Indeed, in this sense, the way factive states might play
a role in justification fully respects the rationale behind the fourth thesis.
3 It is a further objection to coherence accounts that the notion of coherence has been less than
fully worked out. But in this regard, coherence is innocent by association; whether or not there
are any non-inferentially justified claims to serve as foundations, still it is pretty well agreed upon
that some justification is inferential, and that strictly deductive inference is insufficient to support
anything like what we regard as justified in our accounts ofthe world. Thus the sorts of relations
which figure heavily in coherence accounts, explanatory, unifying and inductive, figure in
competing accounts as well.
4 See Bonjour's The Structure of Empirical Knowledge, (Harvard 1985), especially ch. 8.
214 THE DIALECTICAL ILLUSION
5 Princeton University Press, 1992. See Part Three, and especially the appendix to Chapter 12, at
pp. 216-222.
6 One problem for Rescher is his notion of a datum. A datum is not simply a possible truth; it has
a stronger claim on veridicality than that, such that it is "to be classed as true, provided that doing
so creates no anomalies." Gp. Cit., at p. 166. One may well ask in virtue of what an otherwise
mere claim is entitled to this more presumptively alethic status. Rescher has much to say about
this concedely crucial issue which I must pass over here. I can say, however, that a metaphysical
view like that of the idealists, for whom all claims are connected to the real as abstracta, could be
pressed into service here. Likewise the pragmatics of praxis might be invoked on the behalf of the
treatment of claims as data. Neither ofthese moves is obviously out of bounds in Rescher's system
which is, after all, a pragmatic idealism.
7 In my "Interpreting Davidson's Omniscient Interpreter", Canadian Journal of Philosophy,
all competitors, and that the agent's acceptance system does not plausibly reflect all the these
competitors, it is clear that even personaljustification on Lehrer's account fails to be doxastic. See
my "Justified Acceptance, Information and Knowledge" Philosophical Forum, XXV, 1994:212-
230, for the details of this argument.
12 CTC, p. 253.
RICHARD N. MANNING 215
13 TK, p. 122.
14 Lehrer is happy to include in an agent's acceptance system specific acceptances whose contents
are denials of specific defeaters, where that is required to preserve intuitions about knowledge in
particular contexts, including those not involving essential reliance on falsehoods. He can do this
because of the functional character of acceptance. In accepting the things I do, I evidently also
accept, with respect to specific possible defeaters, that they do not obtain. Ifmy being cut off from
the world, for example, would defeat my claim to true belief about it, then, in accepting that I have
such true belief, I evidently accept that I am not cut off from the world. In my "Justified
Acceptance, Information, and Knowledge", I urge that this has the consequence of obliterating the
distinction between justified and unjustifed beliefs.
15 It is an interesting question whether opacity problems could arise in the context of acceptance
for Lehrer. Since acceptance is a functional notion, acceptance being a matter of how contents are
actually used in inference, it might be though purely extensional. On the other hand, to the extent
that one's behavior in assenting to and dissenting from statements is functional evidence of
acceptance, opacity problems might arise. I can assent to "Venus is the morning star" while
dissenting from "Venus is the evening star" and can treat these claims differently from a functional
standpoint notwithstanding that they are coextensive.
16 Given the functional notion of acceptance, the general iterative principle seems reasonable, and
the conjunctive principle is not counter-intuitive. The deductive closure principle is certainly false
if taken in its generality. This is why I have used the permissive "may employ" in characterizing
the operation of the rules. Thanks, by the way, to Myles Brand, for bringing to my attention a
premise missing from a previous formulation of my argument here.
17 Wayne A. Davis and John W. Bender, "Technical Flaws in the Coherence Theory", Synthese
79.2 (1989): 271.
18 KR, p. 143.
22 TK, p. 124.
23 K. Lehrer, Self-Trust: A Study o/Reason, Knowledge, and Automony (Oxford 1997), p. 22.
24 Ibid.
25 Ibid.
26 Ibid.
27 This anti-naturalism is made manifest in many places. A particularly clear and worked out
argument for the claim that epistemic properties are not natural can be found in chapter 3 of Self-
Trust. There Lehrer argues that knowledge does not supervene on any natural properties, since my
actual trustworthiness as an acceptor is crucial to knowledge, and "nature is silent about what has
worth, what is worthy of our trust, and what is justified" (p. 72).
28 The situation is indeed quite like that which faced Descartes as he tried to ground our
knowledge. He recognized that what was needed was a way of defusing the worry that our beliefs
might be mistaken, despite our most careful use of our reasoning powers. His method was to
attempt to instill this self-confidence by causing us to reflect, first, on our inability to doubt our
own existence, and second, on the general features - the clarity and distinctness- of the "I exist"
216 THE DIALECTICAL ILLUSION
Hans Rott
University ofRegensburg
1. INTRODUCTION
Arguably, the representation of pieces of knowledge should not differ from the
representation of beliefs. From a first-person perspective there is little if any-
thing that allows one to distinguish between mere belief and real knowledge,
and it is doubtful whether we should expect the representation of belief and
knowledge to represent more than what is accessible to the reasoner or reason-
ing system itself.
Now let us suppose, for the sake of argument, that questions concerning
the representation of knowledge and belief have been answered to everybody's
satisfaction. Many interesting questions are still left open. Solutions regarding
the concept and the representation of knowledge do not automatically give an-
swers to questions concerning the dynamics of belief and knowledge. How is
'knowledge', or better: alleged knowledge, revised in the light of new evidence?
How should it be revised? Agents are fallible, and what they consider to be
knowledge quite often turns out to be false-and hence not to be proper knowl-
edge at all. At this stage the focus of attention gets shifted from knowledge
to belief, and normative questions are seen to become increasingly important.
Epistemology, the philosophy of mind and psychology analyse and describe
what knowledge is and how it is obtained and represented in human beings.
Knowledge representation has to do with normative issues in so far as there are
HANS ROTT 221
(ii) the analysis of belief revision should not be conducted without a proper
understanding of the concepts and categories that have been used in study
of knowledge.
We must actually restrict claim (i) considerably because the following consid-
erations will be based on the particular epistemological theory of Keith Lehrer.
I take Lehrer to be following in great detail a trail that was started by Plato, in a
beautiful passage in one of his earlier dialogues:
True opinions too are a fine thing and altogether good in their effects so
long as they stay with one, but they won't willingly stay long and instead
run away from a person's soul, so they're not worth much until one ties
them down by reasoning out the explanation .... And when they've been
222 LEHRER'S DYNAMIC THEORY OF KNOWLEDGE
tied down, then for one thing they become items of knowledge, and for
another, pennanent. And that's what makes knowledge more valuable
than right opinion, and the way knowledge differs from right opinion is
by being tied down. (Meno 97e-98a, Plato 1994, p. 69)
Most of the following considerations will not depend on the details of Lehrer's
theory; in fact I shall offer a few non-trivial improvements on some of his defi-
nitions. But I want to base my arguments on the overall architecture of Lehrer's
undertaking. If Lehrer were completely misguided, then what I say about the
relation between epistemology and the theory of belief revision might equally
well be mistaken.
Claim (ii) needs to be qualified as well. The analysis of belief revision is
not dependent on features that distinguish genuine knowledge from mere belief.
It is rather dependent on the structure and the fonnation of beliefs as they are
relevant in the theory of knowledge. What I have in mind above all is the funda-
mental distinction between foundations and coherence theories of knowledge.
This distinction happens to have come to the fore in the theory of knowledge,
but it may just as well be placed in a theory ofbelief. 4 It is primarily concerned
with the inferential relations between various beliefs, that is to say, with the
internal structure of our belief systems. The contrast lies in the answer to the
question whether there is such a thing as a belief base, that is, a distinguished
set of beliefs that are not in need of an inferential justification by other beliefs
and that taken together inferentially justify all the remaining ('derived') beliefs.
All of this can be dealt with in a general theory of belief; nothing requires to
refer this topic to the theory of knowledge. Claim (ii) can still be upheld, given
the fact that many relevant aspects of the structure of beliefs have as a matter of
fact come out most clearly in epistemological discussions.
In this section we unroll the central parts of Lehrer's theory of knowledge and
show how they are rooted in problems and questions which belong to the theory
of belief change. The following presentation is based on the summary in Lehrer
1990, pp. 147-149. The time parameter t which does not play any interesting
role in Lehrer's theory will be removed.
For a long time it has been thought in philosophy that knowledge is jus-
tified true belief. The short and famous article by Gettier published in 1963
has made it clear that this analysis is inadequate. Let us have a look at one of
Gettier's counterexamples.
Suppose that Smith has very strong evidence for
HANS ROTT 223
relies on a false statement. The above phrasing of clause (iv) expresses this
more accurately than Lehrer's official formulation. 7
Lehrer further characterizes the concept of knowledge in a two-layered
strategy.8 First he develops the notion of personal justification which is based on
an agent's subjective acceptance system. In a second step the purely subjective
standpoint gets transcended by several operations on the agent's current belief
set that could help him to approximate the whole truth. 9
where p( cp) and p( ---,cp) = 1 - p( cp) are the probabilities of cp being true or false
respectively, and Ut( cp) and Ufi cp) are the positive or negative utilities of ac-
cepting the hypothesis cp, when cp is true or false, respectively. The utility Ut( cp)
is supposed to reflect the informativeness of cp, and possibly other virtues such
as cp's explanatory power, simplicity, or pragmatic value, and the advantage of
conserving existing beliefs (Lehrer 1990, p. 131).
It should be noted that this absolute degree of reasonableness of accept-
ing is not quite sufficient for what we need in order to understand the fore-
going definitions. In the definition of competition, Lehrer appeals to the rea-
sonableness of accepting cp on the assumption that?j; is true or false. What we
need, then, is something like expected conditional epistemic utilities r( cp!?j;) and
r(cp!---,?j;), and it is left unspecified how we can get them. There is no problem
with the well-known concept of conditional probabilities,15 but it is not quite
clear whether the utilities of accepting a hypothesis cp conditional on accepting
?j; or ---,?j; should be thought of as different from the plain, unconditional utility of
accepting cpo But if the utility of accepting a sentence is dependent on accepting
some other sentence, should not the utility of accepting a sentence be sensitive
to the context of acceptance, that is, to the acceptance system X as a whole?
This question leads us to a problem that is both more general and more
important. As indicated in definitions D4-D6, the reasonableness of accepting a
statement may be-and probably should be-relative to the acceptance system
of the agent. But the above definition of r( cp) does not reflect this. It is rather
an absolute measure of reasonableness. This is quite contrary to the coher-
entist's aim of evaluating systems of hypotheses rather than single hypotheses
taken in isolation. Lehrer can counter this objection by saying that the utility
functions Ut and Uf depend on the current acceptance system X. But then one
may ask whether it is illuminating to base an analysis of 'coherence of cp with
a system X' on an unexplained notion of 'utility of accepting cp on the basis of
systemX,.16
Appealing to cognitive decision theory suggests that the acceptance of a
proposition is a matter of decision. Such an assumption is at least controversial.
But in this respect Lehrer's replacement of 'belief' by 'acceptance' is a prudent
move. It is certainly much more plausible to say that an agent decides to accept
something than that he decides to believe something. Another potential point
of criticism is that it need not be (objective? subjective?) probability that is
taken into account when the reasonableness of accepting certain hypotheses gets
assessed. Perhaps plausibility, a notion with different formal characteristics,
is an equally suitable candidate. This objection, too, could be countered, by
pointing out that decision theory is simply a theory based on probabilities. 17
HANS ROTT 227
belief. Let us suppose that Jones told Smith that he has a Ford, showed him
papers stating that he, Jones, owns a Ford, and always drives a Ford on his
way from his home to his office. All this is believed and known (in some pre-
theoretical sense) by Smith, and justifies his belief that Smith does in fact own
a Ford. So the reason for Smith's not knowing that sentence X above is true
is not that he accepts false sentences, but rather that he is not aware of all true
sentences that are relevant to the case. Notice that it is not enough to know some
relevant facts since a biased selection of true facts may be utterly misleading
and tum the agent away from some other truths. Lehrer suggests to solve this
difficulty be looking at what he calls the 'ultrasystem' of an agent. Here are his
definitions.
D 11 S is justified in accepting that cp in a way that is undefeated if and only
if S is justified in accepting cp on the basis of every system that is a
member of the ultrasystem of S.
D12 A system M is a member of the ultrasystem of S if and only if either
M is the acceptance system of S or results from
- eliminating one or more statements of the form, 'S accepts that 1/)' ,
when 'ljJ is false,
- replacing one or more statements of the form, 'S accepts that 1jJ' ,
with a statement of the form'S accepts that not 'ljJ', when .1jJ is
false,
- or any combination of such eliminations and replacements in the
acceptance system of S,
with the constraint that if'ljJ logically entails X which is false and also
accepted, then'S accepts that X' must also be eliminated or replaced
just as'S accepts that 'ljJ' was.
Before moving on to the criticism of Lehrer's, we should mention his key
result: 'Knowledge reduces to undefeated justification, a just reward for our ar-
duous analytical efforts.' (Lehrer 1990, p. 149) Clearly, undefeated justification
implies personal justification: the actual acceptance system of S is a member
of the ultrasystem; it also implies verific justification: the sceptic can make S
eliminate all his false beliefs, thus effecting a transition to the agent's verific
system;21 finally, it also implies truth: if cp were false, the sceptic could make S
eliminate cpo
Lehrer's theory of knowledge can be called a dynamic one because we
can think of the repeated operations mentioned in D 12 as (potential) steps in a
journey through the space of belief states.
230 LEHRER'S DYNAMIC THEORY OF KNOWLEDGE
One may wonder why Lehrer requires for the undefeated justification
of ip that every member of the ultrasystem must support ip. It would seem suffi-
cient that S is ultimately justified. By this we mean that for every member M
of the ultrasystem of S there is another member M' of the ultrasystem of S
which improves on M and on the basis of which S is justified in accepting ip.
That NI' improves on M means, of course, that M' can be reached from M
by some combination of 'truth-conducive' eliminations and replacements of the
kind specified in definition D12. This concept seems more adequate since even
if S knows that ip, pre-theoretically understood, a mischievous sceptic may well
advance an impressive battery of true facts speaking against the truth of ip, so
that S looses his confidence that ip is true. Only later in his conversation with
the omniscient sceptic, when S comes to know more about the truth, will he
regain his old true and justified belief. Although the correct belief would be
dropped on the receipt of true but misleading information, we may consider it
to constitute knowledge, since one can later learn that this information has in
fact been misleading. Temporary doubts about ip should not count, so it seems,
as long as all potential paths of the ultra justification game finally lead to ip'S ac-
ceptance. 22 Lehrer himself seems to agree with that when discussing his Grabit
example: 23
Suppose I see a man, Tom Grabit, with whom I am acquainted and have
seen often before, standing a few yards from me in the library. I observe
him take a book off the shelf and leave the library. I am justified in
accepting that Tom Grabit took a book, and, assuming he did take it, I
know that he did. Imagine, however, that Tom Grabit's father has, quite
unknown to me, told someone that Tom was not in town today, but his
identical twin brother, John, who he himself often confuses with Tom, is
in town at the library getting a book. Had I known that Tom's father said
this, I would not have been justified in accepting that I saw Tom Grabit
take the book, for if Mr. Grabit confuses Tom for John, as he says, then I
might surely have done so, too. (Lehrer 1990, p. 139)
personally and ultimately justified in believing that either Jones owns a Ford or
Brown is in Barcelona (0. What prevents him from knowing this is that his
belief that Jones owns a Ford is false, so he is not verifically justified to believe
that ~.24 Ultimate justification is not sufficient for knowledge.
Since every theory can be improved by a transition to the one true and
complete theory about the world,25 ultimate justification reduces to justification
on the basis of that theory. On the one hand, it does not seem objectionable to
call for that theory as the final arbiter of knowledge. On the other hand, I cannot
see why the whole truth must be coherent, in Lehrer's or in any other but the
purely logical sense. Shouldn't we try to avoid stipulating that the one true and
complete theory is coherent, because that would mean basing epistemology on
a questionable metaphysics? Similarly, I do not see any intuitive reason why
every truth should be justified, on the basis of the true and complete theory. If
this is right, then it is impossible, by Lehrer's own definition of knowledge as
well as by the definition using ultimate justification, that an agent will know
the whole truth. Shouldn't we try to avoid this conclusion? This suggests that
undefeated justification may not be necessary for knowledge, and even ultimate
justification may not be.
But let us stop with these cosmic speculations now and return to more
definite matters again. In Lehrer 2000, especially pp. 153-154, 168-169, there
are no replacements any more, and there is no talk of strong corrections. In this
new account, the ultrasystem is closer in essence to what was called the verific
system earlier,26 and undefeated justification is similar to verific justification,
i.e., justification on what remains when everything false is eliminated from the
person's acceptance system.
Unfortunately, Lehrer does not tell the reader why he has changed his
earlier definitions and given up on the idea that not only eliminations, but re-
placements, too, may be prompted by the sceptic. It is not clear whether he
just considers it as a simplification of his former account or whether he thinks
that the new edition of his book actually corrects an inadequacy of his former
account. I presume that the reason lies in the problem of misleading informa-
tion. In cases like the Grabit example, receipt of information about what Tom
Grabit's father said (without information about the father's mental state) would
probably have done away my acceptance that Tom Grabit stole the book, even
though this seems to be a bit of genuine knowledge. So it appears that accord-
ing to Lehrer, we should not admit replacements or additions to our stock of
accepted propositions when testing for knowledge. This is an interesting ar-
gument, but its validity may well be doubted. Misleading effects cannot only
be achieved by adding truths but also by removing falsehoods. There are other
cases of a similar structure where in fact no genuine knowledge seems to be in-
232 LEHRER'S DYNAMIC THEORY OF KNOWLEDGE
These are the only constraints Lehrer enters into his official definitions,
but in the running text he acknowledges more constraints of a similar kind.
They are in a sense complementary to the ones we just mentioned. Let 7/J again
be the false sentence to be eliminated or replaced by its negation -,7/J. While the
first group of constraints concerns (false) sentences implied by 7/J, the second
group deals with (necessarily false) sentences implying 7/J. The first group of
constraints is forward-looking, the second group is backward-looking. Here is
the quotation from Lehrer 1990, p. 141:
The sceptic ... may require the claimant to eliminate anything the claim-
ant accepts that is false, and the claimant must eliminate the specified
item from his acceptance system and at the same time eliminate anything
he accepts that 10gicaIIy implies the eliminated item. Or the sceptic may
require the claimant to replace anything the claimant accepts that is false
with the acceptance of its denial and at the same time replace anything
that 10gicaIIy implies the replaced item with acceptance of its denial. [My
italics]
effect consider the accepted items in isolation rather than as items in an accep-
tance system. The constraints do not really address the logical coherence of a
belief with all its surrounding beliefs. It is important to take into account the
context of the remaining accepted items when formulating logical constraints
for eliminations and replacements. Without any claim that these are 'the right'
constraints, the following ones are certainly more adequate in that they show
some sensitivity to the context in which beliefs are situated. We keep on using
the variable '1jJ' for the false sentence that is to be eliminated or replaced by its
negation, and give both a formulation that is close to Lehrer's own statements
and a slightly more formalized version.
In this paper, I have used concepts and ideas borrowed from belief revision the-
ory to elaborate on an important contemporary account in the theory of knowl-
edge. But the symbiosis between the two research areas may equally well be
viewed from the opposite perspective. In this concluding section, I want to give
an indication of how concepts and ideas developed in epistemology can help to
structure and interpret much of the work that has been done in belief revision
theory.
One of the most relevant distinctions for belief revision is that between
foundationalist and coherentist approaches in epistemology. Lehrer (1990,
p. 13) characterizes the fundamental difference between foundationalist and co-
herentist views of knowledge as follows.
According to foundationalists, knowledge and justification are based on
some sort of foundation, the first premises of justification. These prem-
ises provide us with basic beliefs that are justified in themselves, or self-
justified beliefs, upon which the justification for all other beliefs rests.
Coherentists argue that justification must be distinguished from argu-
mentation and reasoning. For them, there need not be any basic beliefs
because all beliefs may be justified by their relation to others by mutual
support. [My italics, HR]
Ernest Sosa (1980, pp. 23-24) makes essentially the same point in more meta-
phorical terms:
HANS ROTT 237
(1) Can a distinction between basic beliefs and derived beliefs be validly
drawn?
(2) And if so, are changes of beliefs made primarily on the base level or on
the level of 'coherent theories'?
It is clear what the foundationalist's and the coherentist's answers will look like.
The former af'finns while the latter denies the first question. In response to the
second question, the former would say' on the base level', while the latter, lack-
ing a distinguished base level, must opt for the level of coherent theories. It is
necessary to work out in greater detail the concepts and distinctions on which
these answers are based. In Rott 2001 I have tried to provide a framework that
helps us to understand the issues involved and to characterize two fundamen-
tally different perspectives on the process of belief revision. These perspectives
turned out to be related to, but not identical with, the dichotomy between foun-
dationalism and coherentism. I did not go as far, however, as Hansson and Ols-
son (1999) who argue that the coherence theory in the epistemologist's sense is
trivialized in the context of the coherence perspective on belief revision.
When transferring the idea of 'foundations' of knowledge to the area of
belief change, it is not particularly important whether the basic beliefs are true,
let alone infallibly true. Neither is it important that they are justified to such
a high degree that they may be regarded as certain. In the current theories of
belief change, belief bases are not supposed to carry any of these connotations.
Basic beliefs are distinguished from derived beliefs only by the fact that they
are somehow 'given', either explicitly or as things that are taken for granted.
Givenness is not at all supposed to imply indefeasibility here. Still, as I tried to
238 LEHRER'S DYNAMIC THEORY OF KNOWLEDGE
show in Rott 2001, these categories coming from epistemology can be exploited
for an illuminating analysis of the dynamics of belief.
ENDNOTES
15 Let us assume that p( 1j;) and p( ~1j;) are positive, so that conditionalizing by either 1j; or ~1j; is
not beset with the problem of an ill-defined division by zero.
16 In his brief discussion of expected epistemic utilities, Lehrer shows little awareness of the fact
that Ut and Uf mayor should depend on X, if reasonableness of acceptance is to be relative to the
current acceptance system. Lehrer contrasts Ut( 'P) with 'P's (objective or subjective) probability
and links it to 'P's truth, but not to the acceptance of other sentences. He does not say anything
about Ul
17 Theories of plausibility of the kind I have in mind are offered, amongst others, by Grove
(1988), Rescher (1976), Shackle (1961), and Spohn (1988). They would have to be supplemented
by a qualitative decision theory.
18 The distinction between relational coherence (coherence as a relation) and systemic coherence
(coherence as a property of a system) is discussed in connection with Lehrer's theory by Olsson
(1999).
19 This problem for Lehrer's theory has also been treated by Olsson (1998).
20 This precept is plausible only if one assumes, as Lehrer apparently does, that the expected
utility of rejecting a hypothesis is zero. An alternative idea would be to accept just those proposi-
tions 'P which have a degree of reasonableness that exceeds a contextually fixed threshold value.
21 We neglect the possibility that the removal of a false belief may tear along some true beliefs.
22 Could there be an eternal wiggling of the acceptance value of'P in response to the sceptic's
challenges? Not if we neglect the possibility that a truth-conducive change can make 5 drop
truths (as we decided to do in footnote 21) and if we set aside questions of infinity. In such a
context, the sceptic has the means to make 5 accept the true and complete theory about the world
which, of course, cannot be further improved.
23 A similar example about barns and papier-mache facsimiles originally due to Carl Ginet is
discussed in Nozick 1981, pp. 174-175, and Bach 1984, pp. 40--41.
24 This argument depends on the assumption that if ~ is not part of an acceptance system, then
it cannot be justified on the basis of that system. Strictly speaking, Lehrer does not make that
assumption; compare Lehrer's definitions D2-D6 and especially my comment on D3 above.
25 Saying this actually steps beyond Lehrer's account, which does not provide for the possibil-
ity of knowledge expansion through the sceptic-which marks an important difference between
Lehrer's concept of replacements and the usual understanding of the concept of revision. The
uniqueness involved in talking about 'the one true and complete theory about the world' is of
course relative to the language used, which I assume as given.
26 I neglect here, perhaps uncharitably, the fact that Lehrer's second edition uses richer 'evalua-
tion systems' instead of the 'acceptance systems' of the first edition. The presentation of Lehrer
2000 follows that of Lehrer 1990 more closely than is warranted by its contents. While the first
edition has the ultrasystem as a genuinely new and complex system, in the second edition the
ultra system is nothing more than the pair consisting of the original system and the verific system.
Introducing the term 'ultrasystem' for this entity seems a bit pompuous, but it is just a reflection
of how the book came into being.
27 This is reminiscent of the idea that the justification for 'P does not only consist in the presence
of reasons for 'P, but also in the absence of reasons against 'P. The point is given pride of place in
nonmonotonic reasoning in the tradition of Doyle (1979).
28 I have mentioned an important difference between 'replacements' and 'revisions' in foot-
note 25: replacements substitute ~'P for some previous belief 'P, while revisions include expan-
sions of belief sets by sentences about which there had not been any opinion before.
29 If this were true, then Lehrer would tum out to be a foundationalist (at least, a foundationalist
in the sense of Chapter 3 of Rott 200 I).
30 My italics. Compare footnote lion p. 194 of Lehrer 1990: ' ... with the constraint that
240 LEHRER'S DYNAMIC THEORY OF KNOWLEDGE
if'lj; logically entails X, which is false and also accepted, then "3 accepts that X" must also
be eliminated or replaced in the same way as "3 accepts that 'Ij;" was.' (Again, my italics) I
am reading the phrases "in the same way as" (in Lehrer's footnote I I) and 'just as" (in D12)
as indicating that a replacement of 'Ij; by its negation should enforce a replacement of X by its
negation.
31 Another fundamental idea which is not mentioned in Lehrer's account is that doxastic changes
should be conservative, i.e., that they should incur only minimal changes to the previous accep-
tance system. On this idea, and its role in belief change theories, compare Rott 2000.
32 As Lehrer realizes very clearly, this objective question has to be linked to the subjective ques-
tion of coherence in a separate step.
33 There has been an ongoing controversy over the coherentism-vs.-foundationalism issue in the
belief revision literature for more than a decade, see Nebel 1989, Giirdenfors 1990, Doyle 1992,
Nayak 1994, del Val 1994, del Val 1997, Hansson and Olsson 1999, Bochman 2000, Bochman
2001 and Rott 2001. Particularly influential as a mediator between epistemology and belief
change was Harman (I 986).
HANS ROTT 241
REFERENCES
Bach, Kent. 1984. "Default Reasoning: Jumping to Conclusions and Knowing When to Think
Twice." Pacific Philosophical Quarterly 65:37-58.
Bochman, Alexander. 2000. "A Foundationalist View of the AGM Theory of Belief Change."
Artificial Intelligence 116:237-263.
---.2001. A Logical Theory ofNon monotonic Inference and BeliefChange. Springer Verlag.
Bonjour, Laurence. 1985. The Structure of Empirical Knowledge. Cambridge, Mass.: Harvard
University Press.
del Val, Alvaro. 1994. "On the Relation Between the Coherence and Foundations Theories of
Belief Revision." AAAI'94 - Proceedings of the National Conference of the American
Association on Artificial Intelligence, pp. 909-914.
- - - . 1997. "Non-Monotonic Reasoning and Belief Revision: Syntactic, Semantic, Founda-
tional, and Coherence Approaches." Journal ofApplied Non-Classical Logics 7:213-240.
Doyle, Jon. 1979. "A Truth Maintenance System." Artificial Intelligence 12:231-272.
- - - . 1992. "Reason Maintenance and Belief Revision: Foundations vs. Coherence Theo-
ries." In Peter Gardenfors (ed.), BeliefRevision, Cambridge University Press, Cambridge,
pp.29-51.
Gardenfors, Peter. 1990. "The Dynamics of Belief Systems: Foundations vs. Coherence Theo-
ries." Revue Internationale de Philosophie 44:24-46.
Grove, Adam. 1988. "Two Modellings for Theory Change." Journal of Philosophical Logic
17: 157-170.
Hansson, Sven 0., and Erik 1. Olsson. 1999. "Providing Foundations for Coherentism." Erkennt-
nis 51:243-265.
Harman, Gilbert. 1973. Thought. Princeton: Princeton University Press.
- - - . 1986. Change in View. Cambridge, Mass.: Bradford Books, MIT Press.
Lehrer, Keith. 1990. Theory of Knowledge. London: Routledge.
- - - . 1997. Self-Trust: A Study of Reason, Knowledge, and Autonomy. Oxford: Oxford
University Press.
- - - . 2000. Theory of Knowledge. Second, substantially revised edition. Boulder: Westview
Press.
Levi, Isaac. 1967. Gambling with Truth. New York: Knopf.
- - - . 1984. Decisions and Revisions: Philosophical Essays on Knowledge and Value. Cam-
bridge: Cambridge University Press.
McCarthy, Gerald D. (ed.). 1986. The Ethics of BeliefDebate. Atlanta: Atlanta Scholars Press.
Nayak, Abhaya C. 1994. "Foundational Belief Change." Journal ofPhilosophical Logic 23:495-
533.
Nebel, Bernhard. 1989. "A Knowledge Level Analysis of Belief Revision." In Ronald Brachman,
Hector Levesque and Ray Reiter (eds.), Proceedings of the 1st International Conference
on Principles ofKnowledge Representation and Reasoning, Morgan Kaufmann, San Ma-
teo, Cal., pp. 301-311.
Nozick, Robert. 1981. Philosophical Explanations. Cambridge, Mass.: Belknap Press, Harvard
University Press.
Olsson, Erik J. 1998. "Competing for Acceptance: Lehrer's Rule and the Paradoxes of Justifica-
tion." Theoria 64:34-54 .
..- ...~. 1999. "Cohering With." Erkenntnis 50:273-291.
Plantinga, Alvin. 1990. "Justification in the 20th Century." Philosophy and Phenomenological
Research 50, Supplement Vol., 45-71.
Plato. 1994. Meno. In Jane M. Day (transl. and ed.), Plato's Meno in Focus, Routledge, London.
242 LEHRER'S DYNAMIC THEORY OF KNOWLEDGE
Gordian Haas
Universitat Konstanz
Abstract:
According to the analysis of knowledge proposed by Lehrer, knowledge
equals undefeated justified acceptance. Undefeated justification is then spelled
out as coherence with every element of the ultrasystem, as Lehrer calls it. The
question arises how this key-notion should be defined. Two definitions of the
ultrasystem which have been proposed by Lehrer are investigated. An
argument is presented that both definitions are flawed. A formal proof is given
that a third and simpler way to define the ultrasystem is preferable.
This is not, however, the way in which Lehrer defines the ultrasystem. In his
1990 book Theory of Knowledge, he instead gives the following slightly
different definition: 2
This definition differs from (DefUltra) only in the additional requirement that
every statement r which is logically entailed by q and is also accepted and
false should be treated just the way q was treated. In his 1997 book Self- Trust,
Lehrer also gives basically the same definition of an ultrasystem. 3 A
somewhat more informal characterization of the ultrasystem that is equivalent
to the more formal definition (DefUltra') can already be found in Lehrer's
1988 paper Metaknowledge: Undefeated Justification. 4 But in Theory of
Knowledge, Lehrer also describes the ultrasystem in terms of the ultra
justification game, which is a heuristic to illustrate the ultrasystem: 5
She [the skeptic] may require the claimant to eliminate anything the claimant
accepts that is false, and the claimant must eliminate the specified item from his
acceptance system and at the same time eliminate anything he accepts that
logically implies the eliminated item. Or the skeptic may require the claimant to
replace anything the claimant accepts that is false with the acceptance of its
denial and at the same time replace anything that logically implies the replaced
item with acceptance of its denial. [italics GH]
Since (DefUltra') and (DefUltra' ') are obviously not equivalent by definition,
it seems odd that Lehrer gives these two different characterizations of the
ultrasystem in the same book without presenting any philosophical argument
for their equivalence. Worse still, it seems to be pretty obvious that both
definitions are in fact not equivalent except for some trivial cases. Although
there is certainly, therefore, a serious tension between these two
characterizations of the ultrasystem given by Lehrer, I shall not inquire which
one could or should be regarded as the authorized one. Instead, I shall argue
against both (DefUltra') and (DefUltra"). My claim is that one should adopt
the most simple definition of the ultrasystem, that is (DefUltra). I will thereby
argue against all the characterizations of the ultrasystem which Lehrer gave in
[1], [2] and [3]. It is worth mentioning that in the second edition of his Theory
of Knowledge, Lehrer gives a simplified characterization of the ultrasystem 6
that is not affected by the criticism in this paper, and to which this author is
actually quite sympathetic. Also, one can understand the following argument
for (DefUltra) as a partial motivation for Lehrer's new characterization of the
ultrasystem.
But before I will argue for (DefUltra), let us first consider the merits of
(DefUltra') and (DefUltra"). A motivation can be given for both of them.
Although Lehrer has not, to my knowledge, published a motivation for the
introduction of the additional constraints of (DefUltra') and (DefUltra"), he
has orally reported to this author that the reason for doing so has been a worry
that could be called 'the problem of the creation of artificial justifications'.
Let me try to explain what this means as well as I can before arguing that this
worry is groundless. Let us first consider an example supporting (DefUltra)
Suppose someone accepts p and also accepts P/\q, but for some reason fails to
accept q. Now suppose further that p is false and hence that P/\q is also false.
According to (DefUltra) there would be an element M of the ultrasystem
where p, -,(p/\q) E M.7 According to the system M, one would be justified in
accepting -,q since this is a logical consequence of p and -,(p/\q). Since the
person initially did not have any epistemic attitude toward q or -,q, this
creation of a justification for -,q seems to be artificial. A more drastic case
would be one in which the person initially accepts q as well. Then, according
to (De fUltra) , there would be an element M of the ultrasystem with
p, q, -,(p/\q) E M. Since M is logically inconsistent, every proposition would
246 LEHRER'S ULTRASYSTEM
be justified relative to M So, according to (DefUltra), it is possible that in
some elements of the ultra system artificial justifications are created. At least,
prima facie, one can conceive of a worst-case scenario in which a person
originally has a faulty justification for some statement p that gives rise to
Gettier-type problems and in which in every element of the ultrasystem as
defined by (DefUltra), there will be created an artificial justification for p as
well, so that the justification for p will remain undefeated. According to the
analysis of knowledge as undefeated justified acceptance, one would therefore
know that p, contrary to what an adequate definition of the ultrasystem should
accomplish.
The additional constraint of (DefUltra ') seems to be appropriate to
avoid this unwanted effect. Considering the initial example, the system M
would not be an element of the ultrasystem defined by (DefUltra') because
P/\q implies p and p is false and also accepted. Therefore, the constraint of
(DefUltra') would require us to form a system M, instead of M, where we
replace not only P/\q by -,(p/\q) but also replace p by ---,po So we would have
---,p, -,(p/\q) E M. Now -,q is no longer deducible and therefore no artificial
justification has been created. In the modified example from above,
(DefUltra') would similarly require us to form a system M, instead of M,
which is not inconsistent.
A similar consideration can be given to motivate (DefUltra"). Suppose
there is a person who accepts both p and P/\q. Suppose further that p is false
and therefore P/\q is false as well. According to (DefUltra), there is an
element M of the ultrasystem where p ~ M and P/\q E M Though we have
eliminated p in M, p would still be justifiable in M on grounds of accepting
P/\q. Again we would have some unwanted artificial justification. (DefUltra")
seems to be appropriate in dealing with this case. Since p is logically entailed
by P/\q and P/\q is also accepted, the constraint of (DefUltra") would require
us to form a system M' instead of M, where we not only eliminate p but also
eliminate P/\q. Thus on M' we no longer have an unwanted justification for
p. 8
So there seems to be some quite good reasons to define the ultrasystem
by (DefUltra') or (DefUltra") rather than by (DefUltra). Despite this, I will
argue for (DefUltra) because I believe that the goals (DefUltra') and
(DefUltra") are aiming at are already brought into effect by (DefUltra). Ifthis
is true, all one would need is the more simple (DefUltra). Although the proof!
want to give to support my claim is somewhat technical, the underlying idea
is very simple. Therefore, it might be helpful to give a sketch of the idea first,
before we turn to the detailed proof. In the foregoing examples, (DefUltra)
gave rise to systems M of the ultrasystem on which too many beliefs are
justified. By stating additional constraints, (DefUltra') and (DefUltra")
required us to form a system M (M '), instead of M, on which these unwanted
GORDIAN HAAS 247
justifications are no longer available. But I will try to demonstrate that, in
addition to M, the system M (M ') is also a member of the ultrasystem
defined by (DefUltra). Since it is necessary for a justification to be undefeated
not to be defeated by any member of the ultrasystem only those statements
that are justified on both systems M and M (M ') will have an undefeated
justification. Therefore, the unwanted justifications relative to M do not
increase these justifications that will remain undefeated. The claim that an
arbitrary member of the ultrasystem defined by (DefUltra') or (DefUltra") is
also a member of the ultrasystem defined by (DefUltra) of course needs to be
proven. I will first prove that we should prefer (DefUltra) over (DefUltra');
the proof concerning (DefUltra") will be perfectly analogous. For such a
proof, we need to give a formalization of the somewhat informal definitions
of the ultrasystems stated above. Obviously, a recursive approach seems to be
appropriate. But first, some remarks on notation are in order:
Lemma 1: U'S;;; U.
Proof: We have to show that if M E U' then M E U. So let us
suppose that M E U'. We will show by induction on the degree
of Mthat M E U.
degree(M) = 0: It follows M = A and therefore M E U (by clause
1 of Def 1.1).
degree(M) = n: o.k.
degree(M) = n+l:
Case 1: M = N-' q for some N, q where N E U', degree(N) = n,
q E A, q is false.
Because N E U' and degree(N) = n, we have by the
induction hypothesis that N E U.
We have to show that M= N\ {r IrE A and q I- rand r
is false} is also an element of U.
Let {r IrE A and q I- rand r is false} = {rl> .. .,rk}.9
We can then rewrite Mas:
M= (... «M{rJ})\{r2})\... )\{rk}'
Note further that all the rJ, .. .,rk are trivially elements of
A that are false. Because of this, and because N E U, we
have by clause 2 of Def 1.1 that M{rJ} E U. By the
same reasoning, we have (M{rd)\{r2} E U etc.
Therefore we have ME U.
Case 2: M = N*' q for some N, q where N E U', degree(N) = n,
q E A, q is false.
Because N E U' and degree(N) = n, we have by the
induction hypothesis that N E U.
We have to show that M is also an element of U where
M = [N \ {r IrE A and q I- rand r is false} ]
GORDIAN HAAS 249
U {r I-,r E A and q f- -,r and -,r is false}.
Let {r IrE A and q f- rand r is false} = {rl> .. .,rk} 10, we
then have {r I -,r E A and q f- -,r and -,r is false}
= {-,rl> ... ,-,rk}' We can then rewrite Mas:
M = (( ... ((M{rj} )u{-,rj} )\u ... )\{rk} )u{-,rk}'
Note further that all the rj , ... h are trivially elements of
A that are false. Because of this, and because N E U, we
have by clause 3 of Def1.l that (M{rj})u{-,rj} E U.
By the same reasoning, we have
(((M{rd)u{---,rd)\{r2})u{-,r2} E U etc. Therefore we
have ME U.
o
Now we need to introduce some new notations. If M is an element of the
ultrasystem then Js(.M) shall denote the set of all beliefs justified on M. That
is:
If U is an ultrasystem, then Js(U) shall denote the set of beliefs that are
justified on all members of U.
That is:
If U = U{Mi} then
Js(U) = {p Ip coheres with every Mi E U} or more precise:
Js(U) = nJS(Mi).
iEJ
Js(U2) = n [n
iE/VJ
JS(Mi) =
iE/
JS(Mi)] n [n
iEJ
JS(Mi)] =
o
Theorem 1: Js(U) ~ Js(U').
Proof: By specializing Lemma 2 on UI = U' and U2 = U we get:
If U' ~ Uthen Js(U) ~ Js(U').
Since by Lemma 1 we have U' ~ U, we get by modus ponens:
Js(U) ~ Js(U').
o
What is the upshot of all this? The interpretation of Theorem 1 is this: all
justifications that remain undefeated if one defines the ultrasystem by
(DefUltra) also remain undefeated if one defines it by (DefUltra'). The worry
that (DefUltra) could create unwanted justifications, thereby leaving too many
justifications undefeated, was without good reason. Every justification we
wanted to exclude by the additional constraint of (DefUltra') is already
excluded by (DefUltra)! Therefore, if the only reason for introducing the
additional constraint of (DefUltra') is to avoid unwanted justifications, then
we can just abandon this constraint from our definition of an ultrasystem. That
is, we should adopt (DefUltra) because we are benefiting from this in three
ways: (1) We do not have to introduce a constraint for which the motivation is
somewhat controversial; (2) We have exactly what we intended to get by
introducing the additional constraint; (3) We can even simplify the theory by
so doing!
It remains to be demonstrated that (DefUltra) is also preferable over
(DefUltra"). If we give a recursive definition of U" = {MI ", M2", ... }, and
the degree of its members that is analogous to the one given for U and U',
then we are able to prove:
Lemma 3: U" ~ U.
Since the proof for this is also perfectly analogous to the one of Lemma 1, it
would be unforgivable pedantry to present it in detail. But it should be noted
that every element of the set {r IrE A and r f- q} is trivially an element of A
GORDIAN HAAS 251
that is false if q is false. Therefore, all elements of this set fulfill the
requirements of clause 2. and 3. of Def 1.1. This is the reason why we can
treat the case of (DefUltra' ') parallel to the one of (DefUltra') in the induction
step. Using Lemma 2 again, we can immediately infer from Lemma 3:
ENDNOTES
1 An earlier draft of this paper has been presented to Keith Lehrer. I am indebted to him for his
helpful comments. It is worth mentioning that Lehrer welcomed the proposal of this paper. An
earlier version of this paper was also read at the Lehrer-workshop in Konstanz in 2000. I wish
to express my indebtedness to the participants of the workshop for the helpful discussion, in
particular to Wolfgang Spohn and Erik Olsson who also read a first version of the paper.
2 See [2], p. 149 and also p. 194 (n.!l) where Lehrer defines the ultrasystem exactly as in
(DefUltra') [where I added the italics].
3 See [3], p. 44. There are some minor differences between the formulations in [2] and [3]
including the usage of the term 'evaluation system' instead of 'acceptance system' but the
crucial point is the same. In both definitions of the ultrasystem Lehrer states the additional
constraint of (DefUltra').
4 See [1], p. 343.
5 See [2], p. 14l.
6 See [4], p. 171, definition D8.
7 Here and further expressions like 'p EO M' are to be understood as shorthand for
"S accepts that p' EO M'. This notation simplifies our formulas considerably.
8 This case parallels the case of a contraction discussed in the theory of belief revision. If one
contracts ones belief set ('" acceptance system) by p, one not only has to eliminate p but one
also has to take care that p is no longer deducible from the belief set. Therefore everything that
entails p and is also an element of the belief set must be eliminated too just as (DefUltra")
requires.
9 In order to write {r I r EO A and q f- rand r is false} as {rl' .. .,rk} we made the assumption that
{r I r EO A and q f- rand r is false} contains only finitely many elements. This assumption has
been made to simplify the proof. But note that this assumption is not essential. If one abandons
this assumption one has to make a sub-induction on the number of elements in
{r I r EO A and q I- rand r is false}, in addition to the main induction on degree(M).
10 Cf. last footnote.
REFERENCES:
[1] Lehrer, K.: Metaknowledge: Undefeated Justification. In: Synthese 74 (1988). Reprinted in
his: Metamind. New York 1990.
252 LEHRER'S ULTRASYSTEM
ON LEHRER'S SOLUTION TO
THE GETTlER PROBLEM!
Jacob Rosenthal
University of Bonn
(1) p is true,
(2) S believes that p, and
(3) S is justified in believing that p.
It seems clear that the conditions (1 )-(3) are indeed necessary for knowledge.
But as some well-known examples (by Edmund Gettier, among others) show, the
stated conditions are not sufficient. The problem is that S's justification for his
belief that p may involve a false belief in an essential way. In such cases we
typically do not speak of knowledge, although the stated conditions are fulfilled.
So the task is to discover a fourth condition such that (1 )-(4) together are
necessary and sufficient for knowledge.
The following is Keith Lehrer's solution of the problem. 2 Take the set of
all beliefs of S. This is called the acceptance system of S. Condition (3) can be
read as stating that the belief that p is justified relative to this system, or, as
Lehrer says, on the basis of this system. Now, what Lehrer demands in addition
to justification relative to the acceptance system is justification relative to
certain modifications of it. Namely, if the acceptance system contains false
beliefs, and some of them are deleted from the system or even replaced by the
corresponding true beliefto the contrary, the belief that p must still be justified
relative to this new system to count as knowledge.
253
E.J. Olsson (ed.), The Epistemology of Keith Lehrer, 253-259.
© 2003 Kluwer Academic Publishers.
254 JACOB ROSENTHAL
More exactly: Suppose the acceptance system ofS contains n false beliefs.
Now you decide for each of them independently ifit is to be (a) deleted from the
system, (b) replaced by the corresponding true belief, or (c) left untouched. In
this way you can obviously get 3n modifications of the acceptance system (one
of which is the acceptance system itself). S's belief that p is knowledge if and
only if it is justified relative to each of these 3n belief systems.
The idea behind this is that, to count as knowledge, S's belief that p must
survive corrections in the system of all beliefs of S. Make any corrections you
want, either weak (delete a false belief) or strong (replace a false belief by the
corresponding true one }-you always get a system relative to which the belief
that p is still justified. Then, and only then, S's belief that p is knowledge.
This proposal for solving the Gettier problem has great aesthetic appeal,
which is, however, somewhat diminished by a complication introduced by
Lehrer I didn't mention in order to keep things simple. Namely, the deletion or
replacement of false beliefs in the acceptance system is not entirely
unconstrained. If q and r are false propositions, and both are believed by S, and
q logically entails r, then, if you delete the belief that q from the acceptance
system, you must also delete the belief that r, and if you replace the belief that
q by the belief that not-q, you must also replace the belief that r by the belief that
not-r. So, your decisions what to do with the false beliefs are not totally free and
independent from each other. You have to respect relations oflogical entailment
in the indicated way. But this constraint is the only one and, having mentioned
it, the presentation of Lehrer's account of knowledge is complete. 3
I think that Lehrer's conception of knowledge is too demanding. Take the
following example: A reliable person has told me that the senate of my
university has elected Cohen for rector, which is indeed the case. So I know that
Cohen is rector. Most of what we know we get to know in more or less this way.
Now I remember a clause in the constitution of the university to the effect that
the rector is also the chairman of the Research Committee. I conclude that
Cohen is chairman of the Research Committee. But this is, in fact, wrong. The
senate of the university has, on the very same meeting, changed the constitution
and separated the positions. Cohen was only prepared to become rector if he
need not also be chairman of the Research Committee. My source of information
has told me nothing about this (nor should he have). This has the effect that if
I got to know that Cohen was definitely not the chairman of the committee, I
would also doubt his being rector and no longer believe it. There are two false
beliefs in my acceptance system: first, the belief that the constitution of the
university still contains the rule that the rector is also the chairman of the
Research Committee, and second, the belief that Cohen is chairman of this
committee. If the second belief is replaced by the corresponding true belief,
while the first false belief is left unchanged, the belief that Cohen is rector is no
longer justified, i.e., not justified on the basis of this modification of the
THE GETTlER PROBLEM 255
acceptance system. (At least you can construe the case in this way.) So Lehrer's
criterion is not fulfilled, and my opinion that Cohen is rector would not count as
knowledge. This is clearly counterintuitive. I know that Cohen is rector. All I
have done is to draw from this true proposition as one premiss and a false
proposition as second premiss a false conclusion. Normally such an act should
not destroy the knowledge status of the true belief. But Lehrer's condition is
such that this is regularly the case.
I think that Lehrer's idea to consider modifications of the acceptance
system of Sin order to decide whether S's beliefthatp is knowledge is the right
idea, but his condition is too strong. It is sufficient, but not necessary for
knowledge. The problem in the example arises because in modifying the
acceptance system, you are allowed to correct just some of the false beliefs,
while leaving others untouched. In my opinion, Lehrer should have said the
following: the modifications of the acceptance system of S relative to which the
belief that p must be justified must not contain any false beliefs any longer. So,
ifthere are n false beliefs in the acceptance system, you have to decide for each
of them whether it is deleted from the system or replaced by its true counterpart.
There are 2n possibilities to do so. A true belief of S is knowledge if and only if
it is justified relative to the acceptance system of S and relative to these 2n
corrections of it. I think this is a better proposal for solving the Gettier problem.
(And the above-mentioned constraint is now in any case superfluous, because
the modified systems contain only true beliefs.)
Is it really necessary to take into account so many different belief
systems? I don't really know, but with criteria of the Lehrer type it is definitely
not enough to consider just one modification of the acceptance system of S. It
would be much easier, of course, if you could say that S's true belief was
knowledge iff it was justified, first, relative to S's acceptance system, and
second, relative to a certain correction of it. But conditions ofthis type turn out
too weak. I consider the two most natural proposals along this line.
(a) S's true belief that p is knowledge if and only if it is justified relative to the
acceptance system of S and relative to the system that results from deleting all
false beliefs from the acceptance system.
past three, but has just happened to acquire a true belief to this effect. This belief
is justified on the basis of the pedestrian's acceptance system. The acceptance
system contains the false belief that the clock is moving and working in the usual
reliable manner. But this false belief is no indispensable part of a justification
for the pedestrian's opinion about the time, and therefore his opinion is still
justified if the false beliefis deleted from the acceptance system. Ajustification
might run as follows: "This is a clock. It shows ten minutes past three. Most
clocks work properly most of the time and therefore show the right time most of
the time. So I conclude that this clock shows the right time right now and believe
that it is ten minutes past three." The false belief that this clock works properly
is not involved in this reasoning, and so this reasoning is not blocked by merely
deleting the false belief from the acceptance system. Therefore the stated
criterion is fulfilled and gives the false result that the pedestrian's belief about
the time is knowledge.
The remedy seems obvious. If you not only delete the false belief that this
clock works properly from the acceptance system, but replace it with the true
belief that this clock does not work properly, then the justification just sketched
is blocked or, as Lehrer says, defeated. So, what about the following condition?
(b) S's true belief that p is knowledge if and only if it is justified relative to the
acceptance system of S and relative to the system that results from replacing in
the acceptance system all false beliefs by their true counterparts.
correction of his acceptance system? Indeed it is, but the correction gives the
wanderer another way to justify his belief, namely, that the second animal is a
sheep, although it does not look like one. So the conviction that there is a sheep
on the meadow is justified before and after the correction of the wanderer's
acceptance system. But it is justified in two different ways.
Now we have hit on the reason why the original and the modified Lehrer
criterion have to take into account so many different modifications of the
acceptance system of the subject S. If you consider just one or a few of these
modifications, one can always dream up examples where S's original
justification for his true belief is defeated, but where the defeating modifications
open up other ways to justify the belief that p. In such cases we typically do not
count S's belief as knowledge. It is merely a lucky coincidence that the subject,
although his original justification is defeated, has now other, new ways to justify
the true belief in question. In order to make such examples impossible, one has
to put up criteria which refer to a multitude of corrections of the acceptance
system, in a sense, to all possible corrections of it.
But obviously there is another possibility to deal with the problem such
examples pose. You just have to demand that the subject's original justification
is preserved when the acceptance system is corrected. 6 Instead of demanding that
S's belief that p be justified relative to very many different belief systems, you
demand that the belief is justified relative to just a few, but always in the same
way. I propose the following condition as a solution to the Gettier problem along
the indicated lines:
S's true belief that p is knowledge if and only if among the reasons S has for his
belief that p there are reasons r l , r z, ... ,rm with the following properties:
acceptance system. Then, and only then, is the true belief knowledge. In
comparison with Lehrer-type proposals, this proposal for solving the Gettier
problem has the advantage of involving just two belief systems, whereas the
former have the advantage of using merely the concept of a justified belief
(relative to a system of beliefs) and not the more demanding concept of
(sufficient) reasons for a belief (relative to a system of beliefs).
So we have arrived at two proposals for solving the Gettier problem: first,
the modified Lehrer proposal, and second, the one just mentioned. I am not sure
whether they are equivalent. But they could both be satisfactory solutions to the
Gettier problem and yet not be equivalent, as long as they agree in all clear
cases. There are borderline cases of belief in which one does not know whether
to call the belief in question knowledge, because the intuitions are unclear or
divided. No proposed criterion can be dismissed just because it decides a
borderline case in this or that way. As long as it gets the clear cases right, it may
count as a solution of the Gettier problem, and so there may be many
nonequivalent solutions. But I am afraid that sooner or later a clear example will
come up for which the two proposals considered here fail, as was the fate of so
many of their predecessors.
ENDNOTES
I These considerations were first presented on a workshop on Keith Lehrer's epistemology and
related topics, held at the University of Constance on June 16th , 2000. I am grateful to Keith Lehrer
for a discussion of these topics, to Wolfgang Spohn for several valuable remarks, and to
Christopher von Bi1Iow for improving my English.
2The presentation follows Lehrer's Theory o/Knowledge, Boulder 1990. His account has remained
essentially the same since the middle of the 80s and can be said to be the most prominent
internalistic proposal to solve the Gettier problem. (Compare also his book Self-Trust, Oxford
1997.) But recently Lehrer has changed his mind, as can be seen in the second edition of Theory
o/Knowledge, Boulder 2000. His new proposal is similar to proposal (a), discussed below, which
is also unsatisfactory.
3 Actually, I have some difficulties with this constraint. I would expect it to be the other way round.
If q logically entails r, and you delete the beliefthat r from the acceptance system, you should also
delete the beliefthat q, because otherwise the beliefthat r remains in the system in an implicit way.
After all, r is logically entailed by q. The same holds for the case of replacement. So the constraint
should be that in case you delete or replace the belief that r you must do the same with the belief
that q. But that is not important here, because the constraint, whatever it is, will play no role in
what follows. A constraint ofthis type is a half-hearted step into the direction ofthe AGM-theory
of belief revision (see Peter Gardenfors: Knowledge in Flux, Cambridge (Mass.) 1988), and
therefore unsatisfactory anyway. Either you should accept the whole AGM-apparatus (or
something similar), or you should try to make do without any proviso of this form.
4 Bertrand Russell: Human Knowledge. Its Scope and Limits, London 1948, p. 170.
6 As Volker Halbach pointed out to me, John Pollock makes a similar proposal in the Appendix
of his book Contemporary Theories of Knowledge, Totowa 1986. The difference is that Pollock
believes (mistakenly, I think) that there are examples ofthe Gettier type in which the subject does
not believe anything false. So he does not speak of a correction ofthe subject's acceptance system,
but of adding truths to it. (In the clock example the subject does believe something false: namely,
that the clock is working properly. This false belief is no indispensable part of the subject's
justification for his opinion about the time, but it is nevertheless connected with this opinion. That
the subject's justification for his belief does not, or need not, include any false beliefs in Gettier
problem examples does not mean that there is no false belief involved at all.)
SKEPTICISM
Chapter 16
John W. Bender
Ohio University, Athens
I. BASIC SUSPICIONS
Let's begin with some old Lehrerian themes that continue to be central to
the latest version of his theory of knowledge. I Epistemic justification is a matter
263
E.1. Olsson (ed.), The Epistemology of Keith Lehrer, 263-280.
© 2003 Kluwer Academic Publishers.
264 JUSTIFICATION AND TRUSTWORTHINESS
admits he cannot prove the skeptic wrong, but he does think that he has a
convincing response to her and argues that ultimately we not only have
knowledge but that we can (and must, since discursive knowledge is
metaknowledge) know that we know (Lehrer [2000a), 205, 229)· His desire to
establish this makes his recent discussions return to the skeptic again and again,
as ifhe senses the inherent weakness of the theory's two games to really answer
skepticism.
The "keystone" of this effort is the concept of self-trust. And this
keystone is put into place in our defenses against skepticism by the deployment
of an argument Lehrer labels the "trustworthiness argument." The
trustworthiness argument is touted and alleged to have the epistemic ability to
loop our justification back on itself, guaranteeing coherence and saving us from
the surd beliefs of foundational ism and the infinite regress of reasons which
threatens if there is no loop back (Lehrer [2000b), 643). Lehrer now apparently
admits that coherence within an acceptance system that fails to contain an
acceptance that one is trustworthy in what one accepts will not be able to be
elaborated in any way that will yield knowledge, or at least, discursive
knowledge (Lehrer [2000b), 638.) This paper will examine the trustworthiness
argument and Lehrer's more wide-ranging discussion of skepticism and
conclude that the argument is impotent and the discussion ultimately
unsatisfying. Although Hume's objection is, as stated, a type of infinite regress
argument, its spirit still applies to Lehrer's theory: there is a serious doubt
whether the "new judgement" of trustworthiness is, in fact, sufficiently justified
to support our claims of knowledge.
justification. But I do not want to enter a general and probably useless verbal
discussion about the use of 'justification.' However, there is a related
substantive objection that foundationalists have urged against coherentists, and
which is one that any good skeptic would urge as well. Coherence with a
background system of beliefs will justify a target belief only if the background
beliefs themselves are justified. But a personal evaluation system contains all
acceptances the person holds (with the objective of accepting somethingjust in
case it is true); it is not a system of justified beliefs alone. So the skeptical
question is, "Are we really justified in accepting those other things we need to
accept if we are to be able to meet a critic's objections?"
More clearly now than previously, Lehrer admits the force of this line of
argument. His answer, given in the passages above, is that personal evaluation
systems, if they are ultimately to provide us with discursive knowledge, must
contain meta-acceptances of the form, "My acceptance that p is a trustworthy
guide to truth." It is easy to see the necessity of these acceptances in the
justification game. For example, if, on the basis of taste, I accept that the wine
I am drinking is a Bordeaux, I must be able to meet (answer or neutralize)
objections such as, "You have no ability to distinguish Bordeaux from
Californian or Italian Cabernet," or "People often confuse California Cabernet
and Bordeaux." And, intuitively, it seems correct that unless I believe that I can
trust my taste in these types of matters, I will not be justified in accepting that
this wine is a Bordeaux.
But the obvious problem is that the necessary meta-acceptance is now a
member of the evaluation system, and therefore subject to the same doubt that
it itself may not be justified. Lehrer deploys two responses to this latest doubt.
In both Lehrer [2000a] and [2000b] a straightforward inductive argument from
successful past acceptances of the same or similar sort is thought to support the
meta-acceptance that my acceptance of p is trustworthy. Although it seems to
me that this is the crucial and natural argument, there is little more than a
mention of it in Lehrer's discussions. (Cf. [2000a], 138, 142 and [2000b], 642f.)
His real interest seems to lie with a different kind of argument, based upon an
even more fundamental principle of self-trust:
If a person accepts (T), then her acceptance of (T) itself will have
the result that it is reasonable for her to accept that (T) by an
application of the trustworthiness argument to (T) itself as the
target acceptance p. The principle applies to itself. It yields the
results that ifshe accepts (T) with the objective of accepting it just
in case it is true, then she is trustworthy in accepting it, and by the
trustworthiness argument, to the conclusion that she is reasonable
in accepting it (Lehrer [2000a], 142).
If acceptance of (T) has the power to make (T) reasonable, it has the power to
make (T*) reasonable. But, that (T*) is reasonable on the basis of the
trustworthiness argument is completely absurd. The mere acceptance of (T),
even assuming (T)'s truth, has no inherent justificatory power, and no amount
of "looping" through the trustworthiness argument can generate what is not
there.
6. A SECOND CIRCLE?
Coherence within one's personal evaluation system, plus the truth of the
members of that system, such that they ascend to the ultrasystem, is supposed
to yield justification, according to Lehrer. I have been arguing that personal
coherence plus truth cannot equal epistemic justification, because the skeptical
question whether the personal acceptances are truly justified has been side-
274 JUSTIFICATION AND TRUSTWORTHINESS
stepped. We have seen that principles that Lehrer requires in any justificatory
system, such as principles (T) and (TR) are problematic. (T) seems unjustified
by the trustworthiness argument, and (TR) appears to be false in the demon
situation, even though its truth is required for epistemic justfication in that case
on Lehrer's theory. The defenses against skepticism appear to be depleted.
A problem also arises with the claim that, even though no proof can be
offered that skepticism is false, we nevertheless know things and know that we
know those things. The problem has deep roots in the very structure of the
ultrasystem. To see this, consider that everything that a subject accepts with the
objective of accepting it just in case it is true, i.e., everything in his personal
evaluational system, is something he will believe to be a member of his
ultrasystem, or, more precisely, a member of his "T-system," (the subset of the
ultrasystem containing only those states, acceptances, preferences and
reasonings that are true) (168). Since no subject is omniscient in regard to the
possible errors he may have made, he cannot know the membership of his T-
system. (If he did know that his evidence was in error, then, given that he is
motivated to accept things if and only if they are true, he would expunge the
error from his evaluation system.) Yet, Lehrer says,
A person lifting her hand before her eyes accepts that she has a
hand, and she also knows that her justification for accepting this
does not depend on any error of hers. She might not know what
the members of her ultrasystem are, but she knows that whatever
they are, they will leave her justified in accepting that she has a
hand. So, she may know and know that she knows ([2000a], 169).
But how can this be? Precisely because of the fallibility of our quest for truth,
we sometimes accept things that are false or based upon error even when we are
doing our epistemic best. In the famous Gettier-style example, I justifiedly
believe that Mr. Nogot owns a Ferrari because of the evidence I possess for that
acceptance. But I do not know that my conclusion is based upon an error,
because I do not know that Nogot is playing a practical joke on me. Hence,
although I accept and justifiedly accept that my evidence is not deceptive, I am
wrong. So, I do not know the membership of my T-system. Whenever the
skeptical critic commands me to eliminate an acceptance in the ultrajustification
game, it will come as a surprise to me, since, of course, I hold that acceptance
with the objective of accepting something just in case it is true.
Now, if the situation is that the necessary principles (T) and (TR) are
either unjustified or false, and the person does not know the membership of his
ultrasystem, how can it be that the person knows, let alone knows that he knows
that the skeptical objections are less reasonable than his own knowledge claims?
It seems that he cannot.
JOHN W. BENDER 275
Even to know that I know that I have a hand would require, on Lehrer's
theory, that I know that my acceptance that I have a hand is invincible in the
ultrasystem. But how can I know this, if! do not have access to the membership
of my T-system? The conclusion can only be that Lehrer has not made out his
claim that knowledge and meta-knowledge are possible even in the face of
skeptical objections.
It is easy to see the crucial role played by principles (T) and (TR) in this
context. Without a justified acceptance of these principles, it is difficult to
imagine any defense against the skeptical objection that it is not more reasonable
for a person to accept what he does rather than to accept that he is systematically
deceived, or at least in error, in what he accepts. If one's evidence is completely
consistent with the skeptical scenario, how can it be thought to positively
confirm the person's preferred acceptance? To cite some independent
implausibility ofthe skeptical hypothesis appears, again, to be question-begging.
Other acceptances of ours may imply such an implausibility, but they, in their
turn, are open to the skeptical criticism. Lehrer admits that reliance upon
principle (T) is question-begging in any attempt to prove that the skeptic is in
error. But he believes that relying upon (T) in explaining why it is reasonable
to accept what we do is not vicious but virtuous. He says,
[H)ow can we know that we are not deceived when the reasons we accept
for concluding we are not deceived are exactly the same reasons we would
accept for concluding we are not deceived if we were deceived?
The reply is that though the content of the reasons we accept would be the
same, the reasons we accept would not be the same. In the one case, in the
actual world, the reasons would be true and in the other case, in the demon
world, they would be false. This is a crucial difference ([2000a),212).
But, first of all, my reasons for believing what I do in the demon world are
not false, if those reasons are construed as acceptances of the sort, "It appears
to me now that there is a zebra in front of me." The demon makes it false that
there is a zebra in front of me, but does so while giving me the misleading but
true evidence that there appears to be a zebra here. Secondly, and far more
importantly, Lehrer's suggestion that we individuate reasons by considering
their truth-values is highly counter-intuitive, and leads immediately to
unacceptable results. My reasons for believing that Nogot owns a Ferrari are the
same whether or not he owns the car or is playing a practical joke. In either
case, they have to do with the likelihood that someone would have official-
looking papers, assert that he owns a Ferrari, and be seen driving a Ferrari ifhe
did not own the car. My reasons are, as in the demon-world scenario, actually
true: I did see official-looking papers, hear Nogot assert that he owns the Ferrari,
and see him driving the car.
If Lehrer means that it is not the evidence that is false, but, rather, the
conclusion they support, then we are still left with unacceptable results. I take
it that in the ultrasystem, we are to eliminate any false reason I have for my
ultimate acceptance. But then, eliminating my acceptance that Nogot owns a
Ferrari as my reason for believing that someone in my office owns a Ferrari is,
allegedly, not to eliminate the same reason I have for believing that someone
owns a Ferrari if, in fact, Nogot does own the car. This seems, simply, wildly
counter-intuitive. Reasons must be individuated on their content alone, and not
JOHN W. BENDER 277
This passage, however, says only that ifwe are trustworthy in a truth-connected
way, then we can neutralize and defeat any skeptical challenge, and ifwe are not
trustworthy and reliable, then our justification will be defeated.
278 JUSTIFICATION AND TRUSTWORTHINESS
12. CONCLUSION
ENDNOTES
1 Latest expressions of Lehrer's view can be found in Lehrer [1996], Lehrer [2000a] and Lehrer
[2000b]. My attention will focus most upon chapters 6-9 of Lehrer [2000a].
JOHN W. BENDER 279
2A person's evaluation system is a newer, more robust conception of what Lehrer used to call the
acceptance system of a person. The acceptance system contained only acceptances (belief-like
states), but the evaluation system now contains not only acceptances, but preferences to accept one
thing over another, as well as reasoning structures or strategies. The additions have been made to
allow doxastic preferences to count towards judgments of comparative reasonableness, and to meet
objections that the acceptance system did not reflect the inferential structure of a person's
reasoning and therefore could not reflect whether the person was justified in her beliefs.
280 JUSTIFICA TION AND TRUSTWORTHINESS
REFERENCES
Lehrer, K: 1996, Self Trust, A Study of Reason, Knowledge, and Autonomy, Oxford University
Press.
---- :2000a, Theory of Knowledge, Second Edition, Westview Press.
----: 2000b, "Discursive Knowledge," Philosophy and Phenomenological Research, Vol. LK, No.
3,637-653.
Chapter 17
COHERENCE, KNOWLEDGE
AND SKEPTICISM*
Peter Klein
Rutgers University
1. INTRODUCTION
About twenty years ago Keith Lehrer wrote this about how he does
philosophy:
Now, we could all say such things. But Lehrer exemplifies this ideal. I have
always admired him for that. My comments about his most recent account of
knowledge should be taken as part of a long series of discussions we have had
over the years. I have always learned a great deal from them. My bet is that his
reply to my comments will amply demonstrate that he is not about to take to
writing fiction.
281
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 281-297.
© 2003 Kluwer Academic Publishers.
282 COHERENCE, KNOWLEDGE AND SKEPTICISM
Lehrer's account of knowledge has evolved over the years but there have
always been two central concepts designed to grapple with two central problems
in epistemology. 2 Justification is held to be a matter of coherence and knowledge
is taken to be true, undefeated, justified belief. Those two central concepts have
provided evolving responses to the Gettier Problem and skepticism. 3
Those two problems are not unrelated. The Gettier Problem pointed to the
felicitous, coincidental fulfillment of the justification and truth conditions of
knowledge in rather mundane situations. Skepticism-at least of the academic
as opposed to the pyrrhonian variety-is motivated by envisioning scenarios in
which we have fulfilled our epistemic responsibilities to accept only those beliefs
that pass the requirements for justification but nevertheless fall short of
knowledge because even if the beliefs are true, it is a lucky accident because,
given what justified our beliefs, it could easily be that our beliefs are false. If
the demon were sleeping, perhaps there would be hands on some occasion when
we are justified in believing that there are hands. But that would be a lucky
break for us.
I want to consider the adequacy of Lehrer's most recent treatment of the
Gettier Problem and skepticism, but it is first necessary and useful to sketch the
main outlines of Lehrer's accounts of knowledge and justification. There are
places where I might have misunderstood them and laying them out will give
him an opportunity to teach me once again. Let me say, first, that I agree with
Lehrer that knowledge is true, undefeated justified belief. But since the devil is
in the details, we must look at some of the more important features of the
account. Here are some of them [TK 129-137]:
I come to accept that Mr. D. Pendable will attend the next Pacific
AP A meeting roughly a year from now because I accept that he is
scheduled to give a paper at the meeting, I accept that he says he is
going, and I accept that he has kept his commitments in the past.
So, it looks like the proposition Mr. D. Pendable will attend the Pacific APA
meeting is at least prima facie justified for me. But is it personally justified
according to Lehrer's view and, if so, could it also be knowledge? If it is
knowledge, then if Closure holds, do I know that between now and next year Mr.
D. Pendable will not have a fatal heart attack?
The answers appear to be,jirst, that on Lehrer's account I am personally
justified in accepting the proposition that D. Pendable will go to the meeting a
year from now because it coheres with my E-system. I could answer (in Lehrer's
terms) the obvious objection that some people do not fulfill their commitments
even when they intend to by invoking the same strategy that Lehrer invokes in
replying to the skeptic to be discussed later, namely, by claiming correctly that
it is more reasonable for me to accept that D. Pendable will go to the APA than
to accept that he will have a fatal heart attack. Second, on the assumption that
there is no relevant false belief in my acceptance set, the ultrasystem would
sanction the personally justified belief as knowledge. And if so, unless Closure
is false, D. Pendable and I would know that he won't have a fatal heart attack (or
that he won't die in a car crash or that he won't be struck by a meteorite or ... ).
I will assume that closure holds for Lehrer, since a coherentist seems almost
honor-bound to accept it.
I think this is just a specific instance of what seems to be a general
problem with many forms of coherence theories of justification. It is this: On
this view, it appears that whether some proposition, say q, is justified, Le.,
reasonable to believe in the face of objections, will depend on the epistemic
relationships that q has to the other elements of my current evaluation system,
E - including that system's resources in answering or neutralizing objections to
q. So, whether I am permitted to or should accept a new proposition into E
depends solely upon what is currently in E. E-systems can expand incrementally
always using the immediately prior E-system. And I can use the existing E-
System to answer objections to a new, candidate proposition to be added to the
existing E-System. But what worries me about this is that it would seem to
sanction the addition of a proposition to some distant ancestor E-system in a
PETER KLEIN 285
step-by-step process as the E-systems evolve that would not have been
permissible in one giant step from the ancestor system. But that shouldn't be.
I take it that this is the lesson from considering the Sorites situations. It could
be reasonable to add p to some system, E, and to add q to the (E + p) system, but
not reasonable to add q to E. The objection that it would not have been
reasonable for you to accept q on the basis of the original E-system can be
neutralized by the proposition that it is now reasonable on the basis of the (E +
p) system.
The case of Mr. D. Pendable and the more general possible problem with
coherence theories were presented just to get the theory before us. I will return
to them briefly at the end of the next section. Now, I want to turn to the two
main items I wish to discuss: The Gettier Problem and Skepticism.
First, to show that the conditions are not necessary: Suppose there is
some false proposition,/, that I accept. Further, suppose that I know that I accept
/ on the basis, no doubt, of introspecting my acceptance set. I am personally
justified in accepting that I accept! But since the proposition "I accept thatf'
would not be in the ultrasystem (since/is false), I would have no way to answer
or neutralize the objection that I don't accept/and, hence, "I acceptf' would not
be justified in the ultrasystem. 9
I am quite sure that with a little chisholming that problem can be fixed,
although I don't quite see how to do it because, if I understand the account
correctly, "I accept thatf' cannot remain in the ultrasystem and it seems that it
must in order for me to be able to answer the objection that I don't accept!
Indeed, the proposition "I accept f' would seem to ill-cohere with what else
would remain in my ultrasystem since it would seem to include a proposition like
the following: The best way, and a very good way at that, to determine what I
accept is to introspect my acceptance set of beliefs.
Second, the more difficult issue, and the one directly related to the
traditional Gettier Problem narrowly construed, has to do with sufficiency.
Consider a case we can call the Closed-Box Case:
The following propositions are in my acceptance set and are personally justified
for me:
Now add to the story that Sally didn't say that there was a drawing of a
triangle in the box. I seriously misheard her. She actually said that there wasn't
a drawing of a triangle in the box. Presumably, then, I didn't know that there
was a drawing of a triangle in the box, even though it was personally justified for
me. But, now add one more twist to the case. Sally was wrong. There really
was a drawing of a triangle in the box!
The following propositions are in my ultrasystem:
The problem is that 2 is justified or coheres with the evaluation system because
3 provides a perfectly good, indeed an entailing, reason for 2!
Lehrer might perhaps suggest that in the acceptance set and ultrasystem
there will be such meta-evaluative propositions as: I actually (formerly)
accepted both that there was a drawing of a triangle in the box and that there is
a drawing of a plane, closed figure whose interior angles sum to 180 degrees
because I actually (formerly) accepted that Sally said there was a triangle in the
box. I might note that I no longer accept that Sally did say that and hence, I
might conclude that I am no longer justified in accepting the propositions about
what is in the box.
There are four replies to this. First, I might not have such meta-beliefs
about the reasons for my actual acceptances. If it is required that we must have
such beliefs in order to possess knowledge, too much of our knowledge would
be lost. Think of the person who believes that the Battle of Hastings was in 1066
but has no recollection of the basis on which she believes it. Does she forfeit
knowledge?lo A second and related point is that this requires that we have meta-
beliefs about acceptances and their evidential relationships in order to have
knowledge. So, in order to have knowledge that there is drawing of a triangle
in the box, we would all have to be epistemologists and even employ the notion
PETER KLEIN 287
4. SKEPTICISM
Let us turn to skepticism. Once upon a time, Lehrer defended the
plausibility of academic skepticism by suggesting that when we are doing
philosophy no hypothesis should be epistemically privileged over another unless
some reason, presumably a non-question begging reason, could be provided.
Here is what he said then:
... generally arguments about where the burden of proof lies are
unproductive. It is more reasonable to suppose that such questions
are best left to courts of law where they have suitable application.
In philosophy [emphasis added] a different principle of agnoiology
[the study of ignorance] is appropriate, to wit, that no hypothesis
should be rejected as unjustified without argument against it.
Consequently, if the sceptic puts forth a hypothesis inconsistent
with the hypothesis of common sense, then there is no burden of
proof on either side ... [WNS, 53]
The passage is a bit ambiguous, but suppose that it means that when doing
philosophy there is no presumption in favor of any two conflicting hypotheses
and that in order to accept one of them one has to first have good evidence
against the other. If that is what was meant (and I'm not at all sure it was),
Lehrer no longer seems to hold this view since he now believes that we can use
"It is more reasonable for me to accept that I see a zebra than that I am asleep
and dreaming that I see a zebra" in our response to the skeptic/critic who poses
the objection, "You are asleep and (merely) dreaming that you see a zebra." In
other words, I don't have to first eliminate the hypothesis that I am asleep and
merely dreaming that there is a zebra in order to be justified in believing that I
see a zebra.
I think rejecting the requirement that there is some obligation to eliminate
all contrary hypotheses to some proposition, say h, prior to our being justified
in believing h is surely correct. For if it were a requirement-even for doing
philosophy-the road to skepticism would be way too short to be worth traveling.
Consider any two contraries, C 1 and c2 • In order to be justified in believing c l , S
would first have to eliminate c2 • And in order to be justified in believing that c2,
S would first have to eliminate c l . So, of course, S could never be justified in
believing either c 1 or be justified in believing c2. And since every contingent
proposition has a contrary, if this requirement were accepted, knowledge of all
contingent propositions would be automatically prohibited. 14 In so far as
skepticism remains a disputable and interesting philosophical position, the
skeptic cannot impose such an outrageous departure from our ordinary epistemic
practices.
PETER KLEIN 289
So, Lehrer was right in rejecting that stringent requirement. Here is what
he now thinks we can say in response to the skeptic:
I can tell that I am awake and not asleep and dreaming now. My
experience does not feel at all like a dream and I have a distinct
memory of what preceded my present experience, leaving my hotel,
taking a cab to the zoo, buying a ticket, all of which is trustworthy
information that I am now at the zoo looking at a zebra and not
asleep and dreaming. (TK, 133)
Thus, in response to the skeptic we appeal to two types of beliefs: (i) particular
beliefs about our experiences and (ii) a general belief that such information is
trustworthy. This strategy will put off the skeptic for a while, but not for long.
The skeptic will quicky note that in answer to all of her objections, the general
proposition that my information is trustworthy keeps recurring. And she will see
that her better objection is this: "What makes you think that you are trustworthy
in the way that you evaluate information?" Or more simply: "What makes you
think that your belief acquisition methods are trustworthy?"
This should be reminiscent of Descartes' strategy on behalf of the skeptic
in the "First Meditation." There, he considers and rejects many grounds for
doubting the objection that he has trustworthy information about the world. For
example, he considers the objection "your senses have misled you in the past"
and, in Lehrer's terminology, neutralizes it by adding to it the proposition that
he can distinguish those occasions when his senses are trustworthy from those
when they aren't. Later Descartes considers the objection that perhaps he is
dreaming now and neutralizes it by responding that dream images are like
"painted representations" and have been formed as the "counterparts of what is
real and true." So, if we limit our acceptances to those "simple" properties
shared by dream-images and waking images, we still can arrive at knowledge.
Thus, Descartes answers these initial objections by neutralizing them in the way
recommended by Lehrer.
But then, the Cartesian dialectical skeptic raises this doubt:
In other words, the skeptical objection now is this: "You do not have trustworthy
belief acquisition methods."
This is the heart of academic skepticism. The academic skeptic requires
that there be some way within the E-System to answer or neutralize this
290 COHERENCE, KNOWLEDGE AND SKEPTICISM
says that the "loop" explains why it is "reasonable" to accept T, and hence the
viciousness associated with an argument for T does not arise. Thus, a central
question becomes this: What is the crucial difference between an explanation
that shows why it is reasonable to accept a proposition and an argument for the
truth of the proposition?
Lehrer is correct that there is a crucial difference between explaining p and
giving an argument for p. Doing the first is completely consistent with assuming
that p. Both I and my critic agree that p is true. In asking for an explanation of
p, the critic is asking why p is true. The rhetorical situation is such that p's truth
is not at stake. So, explaining that p will not beg the question-since the
question is not whether p is true, but why p is true.
But that is not parallel to the worries concocted by the skeptic. The
skeptic is not willing to grant that T is true. She wants to know whether T is
true; not why T is true. In other words, giving an explanation of why T is true
will neither answer nor neutralize the skeptic's challenge. It will not answer it,
because it presupposes that T is true. It doesn't neutralize it because it does not
provide a basis for accepting ~ T along with something else that overrides the
impact of ~T alone. In short, the distinction between explaining and giving an
argument for will not help here. The skeptic isn't asking for an explanation of
T. The skeptic is challenging the truth of T.
That being said, it still might seem that the more satisfyingly coherent
thing to do is to accept T. And coherence is the essence of justification,
according to Lehrer. After all, wouldn't there be something deeply incoherent
in thinking that my belief formation methods are unreliable or even just that they
might not be reliable. Ernest Sosa has argued for just that point. Here is what he
says (this is a close paraphrase):
But now suppose that by using way W of forming beliefs ... we
arrive at the conviction that W is our way of forming beliefs. Now,
so long as we do not go back on that conviction, does that not
restrict our coherent combinations of attitudes? Take: (e) Believe
that W is my overall way of forming beliefs. And compare (f)
Believe that W is reliable, (g) Deny that W is reliable and (h)
Withhold that W is reliable. It is not evident that (e) & (f) would
be more satisfyingly coherent than either of (e) & (g) or (e) & (h)? 17
Even if we assume that we accept some propositions that the skeptic challenges,
we are not forced either to answer or to neutralize the skeptic's objection. There
is another, I think better, strategy available, namely to question the basis on
which the skeptic believes or even raises the objection. The skeptic asserts that,
or at least seems to be questioning whether, our methods of belief acquisition are
unreliable. She dares us to come up with an argument for T that doesn't
presuppose T. But why are we limited to either answering the objection or
neutralizing it on the basis of something else we believe? Why can't we simply
ask the skeptic for her reasons for thinking that T is false?
Suppose that we thought that all stated arguments about what is a subject
of genuine controversy were either question-begging or based upon arbitrary
premises-ones which cry out for a defense before the conclusion is accepted.
That was what the Pyrrhonians thought and their strategy in dealing with the
dogmatists-whether they were the type who accepted that we had knowledge
or the Academic Skeptic type who accepted that we didn't have
knowledge-was to ask for the reasons for the accepted proposition in order to
show in case after case that either the offered reasons begged the question or
begged for further defense.
The Pyrrhonians would point out that either the skeptic will or will not
have some argument like the following to present:
:. Not T
Consider the first alternative: If the skeptic has no argument with this
form, why should we take the challenge seriously? After all, we have already
said that we need not eliminate all contrary hypotheses before we are justified in
believing a given hypothesis. And if! don't have to eliminate all of them before
PETER KLEIN 293
arriving at the view that the animals are zebras, why would I have to eliminate
all of them after I have arrived at the belief that the animals are zebras?
So, there must be something special about this challenge that requires that
we answer it. It cannot merely be, as some contextualists have suggested, that
a hypothesis becomes a relevant alternative to h (one which must be eliminated
prior to being justified in believing that h), whenever it is raised-even raised
with a serious tone. IS Consider the zebras again. Suppose the skeptic says,
"Well how do you know that those things are not cleverly disguised aliens from
a recently discovered planet outside our solar system? Or that they are not newly
invented super-robots? Or that they are not members of the lost tribe of Israel
who have been practicing the zebra-disguise since the 8th century BC, and
consequently, have gotten very good at it?" Those objections are so far-fetched
so that even if someone advancing those alternatives happens to believe them,
there appears to be no reason why one should have to rise to the bait and
eliminate those alternatives either before or after one has arrived at the belief that
the animals are zebras. And finally, isn't the skeptical hypothesis-that we are
not in the actual world but rather in one which just seems identical to it-just as
far fetched? What is so special about that far-fetched hypothesis?19
Now to the second alternative: If the skeptic does have such an argument,
we can examine it. And we might just find that the Pyrrhonians were right, at
least about this argument. 20 Thus, I think we are not forced either to answer or
to neutralize the skeptic's claim on the basis of some of our other acceptances.
Just because the challenge is raised imposes no burden on us to answer or
neutralize it.
This is the first step in sketching the alternative response alluded to earlier.
Suppose we prefer to answer the skeptic. After all, Lehrer thinks what we are
justified in believing depends, in part, on what we prefer. Consider this
argument schema:
some methods other than M which she thinks are not reliable? That would make
her practice inconsistent with her beliefs.
The fallacy of begging the question in reasoning consists in using a
conclusion of an argument as a basic premise-whether overt or suppressed.
Question-begging arguments can provide no additional reason for accepting the
conclusion beyond those already available for the basic premises. But it is not
a fallacy of reasoning to employ what one takes to be the reliable belief
formation rules in arguing for the claim that those very belieffonnation rules are
reliable. Indeed, to do otherwise would be to fail to practice what one preaches.
Now, the academic skeptic might deny one or both of the premises of the
argument. And if the pyrrhonian skeptic is correct, any such argument will
either beg the question or rely upon premises that require further defense.
Nevertheless, the point here is that the argument for the reliability of methods M
need not beg the question.
I have already partially addressed that question by arguing that even if we accept
some proposition, say p, we need not answer or neutralize the skeptic's objection
to p. We can challenge the skeptic's basis for accepting the objection. But, here,
I mean to be asking whether we need ever be placed in the position of defending
acceptances at all.
Of course, we have that burden only if we accept some propositions that
could be challenged by the skeptic. My question is this: Why is it assumed that
we must accept anything-at least anything that requires inferential support?
It is absolutely crucial to note that for Lehrer to accept p is not the same
as to believe p. Knowledge entails acceptance not belief. To accept something
is to hold it true precisely because doing so promotes our supposed goal of
accepting a proposition iff it is true. According to Lehrer we can even know
something to be true which we believe is false (again paraphrase and quotation):
But what if we thought that our epistemic equipment was not trustworthy enough
to arrive at unobjectionable (in Lehrer's sense of that term) truths in matters as
complex as evaluating the weight of reasons on each side of a complex issue?
What if we thought that we are only trustworthy enough to identifY what seems
true? By "seeming true" I do not mean some sort of hesitating endorsement of
the proposition's being true. Rather, I mean that the proposition has the look of
truth about it. It seems to satisfY what appears to make a proposition true.
Further, what if it seems to us that our goal is not and should not be
unobjectionable propositions but rather provisionally justified belief in what
seems true? That is a goal that seems reachable and does not bring with it the
dangerous flirtation with dogmatism that easily accompanies the preference for
being in a situation in which we are able to answer or neutralize all objections on
the basis of what we currently accept.
Recall part of what I cited that Lehrer wrote twenty years ago: "Should J
ever reach the point at which I am disinclined to seek criticism and amend my
views in light of it, I shall take to writing fiction . . . and give up writing
philosophy altogether." Now I suppose that one could seek criticism in the way
that an arrogant gunslinger would seek out opponents believing that victory will
be his. Such macho philosophic gun fighters have a virtue of their own. But this
isn't Lehrer's motivation. He seeks criticism because he expects that it will
improve the epistemic worth of his beliefs. That sounds like a person who does
not accept that he is ever in a position to answer or neutralize all objections to
what he believes. Such a person would never accept the meta-view that he had
personally justified acceptances.
5. ALTERNATIVETOANSWERINGORNEUTRALIZING
THE SKEPTIC'S CHALLENGE
NOTES
* I want to thank Anne Ashbaugh and especially Keith Lehrer for their help with this paper. Keith
and I have been talking and emailing about some ofthese issues for many years. In particular, an
ancestor of this paper was given at the Pacific AP A meeting (March, 200 I). He replied to it there
and also kindly sent me some further comments later. Keith has taught me a lot over the years and
I expect that his replies to this version of the paper will continue to make me think more about the
issues. That's the fun of it all.
I. Keith Lehrer: Profiles (Dordrecht: D. Reidel Publishing Company, 1981), ed. Radu Bogdan, p.
II.
2. In this paper I will be relying primarily on these works by Lehrer:
Self-Trust: A Study o/Reason, Knowledge and Autonomy (Oxford and New York:
Oxford University Press, 1997) [ST]
3. By "skepticism" I will be referring to Academic Skepticism unless otherwise noted. Using
Lehrer's terminology, the Academic Skeptics thought that it is more reasonable to accept that we
don't (or can't) have knowledge in those areas in which it is generally thought that we do have
knowledge than to deny or withhold such a judgement about the scope of our knowledge. This is
to be contrasted with the Pyrrhonian Skeptic who would withhold such an acceptance (and its
denial).
4. One might worry here that it could never be more reasonable for S to accept a conjunction than
it is to accept one of the conjuncts (unless it entails the other) because the probability of the
conjunction will be less that the probability ofthe conjunct. But I think the answer is this: Since
the goal is to gain truth and avoid falsehood, by accepting the conjunction one has a greater chance
of accepting more truths than one would by merely accepting the one conjunct.
j I am using "valid" in the standard way to indicate an argument whose pattern is such that any
argument with that pattern not possible for the premises to be true and the conclusion false. I am
using "cogent" in the way defined by Richard Feldman in Reason and Argument (Upper Saddle
River, New Jersey: Prentice Hall, 1999 second edition) to refer to good patterns of inductive
reasoning.
6. The best source for a full and illuminating discussion ofthe various counterexamples connected
with the early Gettier Problem literature is Robert Shope's, The Analysis o/Knowledge (Princeton,
New Jersey: Princeton University Press, 1983).
7 It was unreliable at least in this case because too many of its reports were false.
8. This example, or one very similar to it, was suggested to me as a possible counterexample to
some forms of closure by John Hawthorne. I don't think it works against closure because I think
a proper account of justification would have it that you are not justified in believing that D.
Pendable will attend the meeting. But this is not the place to present that account.
9. This example, or one very much like it, was developed by Robert Shope, op cit, pp. 52, 71, 102.
It was also used by Alvin Plantinga in Warrant: The Current Debate (New York and Oxford:
Oxford University Press, 1993, pp. 220-1.) I argued that this counterexample would not work
against my version of the defeasibility theory in "Warrant, Proper Function, Reliabilism, and
Defeasibility" in Warrant in Contemporary Epistemology (London: Rowman & Littlefield
Publishers, 1996), pp. 121-2.
PETER KLEIN 297
10 Examples like this one were given by Colin Radford, "Knowledge - By Example," Analysis,
27,1966,1-11.
11. Of course, I would be very pleased if some epistemologist other than I thought infinitism is the
67.16,1971,471-482.
13. I think neither a coherentist nor foundationalist approach to justification is correct. The third
17. Ernest Sosa, "Philosophical Skepticism and Epistemic Circularity," Proceedings of the
Standards," Synthese, 73 (1987), 3-26 and his "How to be aFallibilist" Philosophical Perspectives,
2 (1988), 91-123, David Lewis in "Elusive Knowledge," Australasian Journal ofPhilosophy, 74
(1996),549-67 and Keith DeRose in "Solving the Skeptical Problem," Philosophical Review, 104
(1995), I-52 and his "Contextualism and Knowledge Attributions," Philosophy and
Phenomenological Research, 52 (1992), 913-929.
19 Let me add that Descartes does not simply raise the hypothesis-he gives a reason for thinking
that it is at least plausible; namely, that the degree of perfection of his epistemic equipment cannot
be higher than the degree of whatever is its cause and, at that point in the Meditations, he has no
reason to think that its cause is sufficiently perfect. Thus, in order to see whether we should take
this challenge seriously, we would have to examine the presuppositions ofthis skeptical challenge.
This is obviously not the place for that.
20. See my paper "Skepticism and Closure: Why the Evil Genius Argument Fails," Philosophical
Topics, 23.1, Spring 1995,213-236, and "How a Pyrrhonian Skeptic Might Respond to Academic
Skepticism," op. cit.
21. This does have the consequence that neither foundationalism nor coherentism is the correct
account of reasoning. As mentioned in fn. 13, I have argued for infinitism in several papers.
Chapter 18
David A. Truncellito
Arkansas State University
1. INTRODUCTION
Personal justification, then, arises when one has the resources within one's
acceptance system to reply to any challenge a skeptic might raise. This
dialectical formulation of personal justification suggests the justification game. 7
In short, the game is played between S (the claimant), who claims to have
knowledge, and a critic (or skeptic), who challenges S' s claims to knowledge by
raising objections. S wins a round by adequately answering or neutralizing the
critic's objection; S wins the game, and achieves personal justification, by
winning every round.
Notice, though, that victory in the justification game does not entail that
S knows that p. This is for the obvious reason that while S' s acceptances were
formed with the goal of attaining truth and avoiding error, he might have failed
to meet that goal in some cases; indeed, given that S, like the rest of us, is
fallible, it is likely that he did fail to meet this goal in some cases. But since
victory in the justification game, and thus personal justification, are based on S' s
acceptance system, S's belief that p, although personally justified, might be
based on falsehoods. And, while personal justification is a noteworthy
achievement, belief based on falsehoods cannot constitute knowledge. This is
the motivation behind the' isolation objection' against coherentist-and, indeed,
any purely internalistic-accounts of justification. 8
DAVrD A. TRUNCELLITO 301
Lehrer's notion of the ultrasystem clearly serves to avoid the worry that
accompanied personal justification, namely the possibility of acceptance being
based on error. However, it exposes him to a new form of objection-one which
might perhaps seem to be less central to the matter of knowledge, but an
important objection nonetheless. Namely, it runs the risk of jeopardizing one's
knowledge of one's own mental states, which surely no account of knowledge
should do.
Robert Shope has suggested a common sort of mistake made by a number
of contemporary philosophers, which he calls the conditional fallacy and
characterizes as follows:
Shope then claims that Lehrer is among those many who are guilty
of committing this fallacy. IS Others, including Peter Klein l6 , Alvin
Plantinga and John Pollock, have offered similar objections to Lehrer's
theory. Let me attempt to unify these objections by describing the
purportedly offensive sort of case.
Let us suppose that S accepts some p, such that p is false. Then, of
course, it follows that S does not know that p. But let us suppose further
that S accepts that S accepts that p. This latter acceptance (call it S's
acceptance that q), intuitively, constitutes knowledge. 17 However, because
ofthe way in which Lehrer has characterized undefeated justification, his
account has the consequence that S does not know that q, which is a
serious mark against the view. Recall that undefeated justification-and
knowledge, which is equivalent -requires the ability to win the ultra
justification game. We could, then, characterize knowledge in terms of a
subjunctive conditional analysis, so as to see just how it is meant to be
vulnerable to the conditional fallacy:
But in the case we are considering, we see a problem for Lehrer's account. For,
p is among the falsehoods that are removed from S's acceptance system in
constructing S's ultrasystem! It follows, then, that S does not know that q (i.e.
does not know that he accepts that p).
4.2. Q is true
In order to evaluate the truth of S's acceptances about the world, we need
to note whether they hold true of the actual world. In particular, in order to
evaluate the truth of S's acceptances about S's acceptance system, we need to
note whether they hold true ofS's actual acceptance system. But q is indeed true
(i.e. S does accept that p), for p is a member of S's acceptance system, even
though it is not a member of S's ultrasystem. We need to recall that the
ultrasystem is a new system, and that falsehoods have been removed from the
ultrasystem, but have not actually been removed from the acceptance system.
Indeed, Lehrer explicitly makes note of this point in his latest book (although
not, to my knowledge, in any of his earlier work): "The ultrasystem,
nevertheless, is required to acknowledge the existence of the eliminated states
of acceptance ... in the original evaluation system, for it is true that they are states
of the person.,,19 In other words, the ultrasystem is nothing more than a
theoretical construct, designed with the purpose of helping us to evaluate the
justificatory status of S's acceptances. But even while we are conducting this
evaluation, S's acceptance system remains unchanged: it continues to include
any falsehoods which might have been in it previously.
Let us suppose that S's acceptance that p is personally justified. This is,
I take it uncontroversial, since personal justification depends upon the
acceptance system but not the ultrasystem. However, it will be informative to
304 ULTRASYSTEM AND CONDITIONAL FALLACY
The exact details might be different in the case of a particular agent. The
crucial point to note, though, is that however the correct account might
proceed, S's acceptance that q is not justified by p.
Now, let us imagine a round of the ultra justification game. The
ultra system will contain q, ql' and q2' but not p, since all falsehoods have
been removed in constructing it.
The critic's move is illegitimate, since not-g is false (see section IV.2 above),
and thus not available to the critic as an objection.
So, the only possible objections one might imagine that the critic would
raise are not-p, which does not constitute an objection to S's acceptance that g,
and not-g, which is false and thus not a legitimate objection. S does, after all,
have undefeated justification and knowledge that p.
DAVID A. TRUNCELLITO 305
5. CONCLUSION
ENDNOTES
6 Lehrer (2000), p.130; see also Lehrer (1990), p.117. In Lehrer (1974), this was referred to as the
corrected doxastic system (p.190).
7 For a detailed description, and several rounds, of the justification game, see Lehrer (2000),
pp.132ff; Lehrer (1990), pp.119ff. I'll rehearse a few rounds of the game in section IV below.
S Lehrer is explicit that, although his account is often taken to be an internalistic theory of
justification, it is not; it is what he prefers to call a 'match theory'. "There must be a match
between what one accepts as a trustworthy guide to truth and what really is a trustworthy guide to
truth ... To obtain knowledge one needs the right mix of internal and external factors." (Lehrer
(2000), p.I72) However, personal justification is a purely internalistic component of Lehrer's
account, in that it is indexed to a subset of S's internal states (viz. S's acceptance system), and that
is why it is vulnerable to the worry raised above.
9 Lehrer (2000), p.l53. See also Lehrer (1974), p.215.
10 Lehrer (2000), pp.l53-4. In Lehrer (1974), this was referred to as the verific alternative to the
12 She may also disallow unsound reasonings (Lehrer 2000), p.160), although that will not concern
us here.
13 Lehrer actually added a third, intermediary stage, in Lehrer (1990), viz. verific justification, with
its attendant notions of the verific system and the verific justification game. Also, in that work,
the ultrasystem was a set of systems rather than a single system. Given that this was not the case
in Lehrer (1974), and no longer is the case in Lehrer (2000), I shall ignore it for the purposes of
my discussion. In any event, the spirit of the proposal is consistent throughout the development
of Lehrer's theory, and that should not be obscured by this wrinkle.
14 Shope (1978), pp.399-400. This is actually Version 2 of the fallacy, which is the version Shope
accuses Lehrer of committing.
15 See Shope (1983), pp.48ff, pp.53ff, and especially pp. 73-74.
16 Most recently in his "Coherence, Knowledge and Skepticism", in this volume.
17 The exact details of the way in which S comes to be justified in his acceptance that he accepts
that p can be ignored here. Presumably, though, the explanation will have to do with the fact that
S introspected such an acceptance and that introspection is a trustworthy source of knowledge of
one's own mental states. If one were to deny that S knows that S accepts that p, then of course the
objection is rendered irrelevant. I shall discuss the details ofthis case somewhat more thoroughly
in the following section.
18 Strictly speaking, since knowledge reduces to undefeated justified acceptance, I do not need to
show that q is true. However, I shall do so nonetheless for the sake of clarity and perspicacity.
Since the objection of the previous section could be interpreted so that the undesirable
consequence of Lehrer's theory is that q is false or that q is not completely justified, I shall respond
to both possibilities.
19 Lehrer (2000), p.160; see also Lehrer (2000), p.168. Note that the evaluation system is the
system composed of all of S's acceptances, as well as his preferences and reasonings. What is
relevant for our purposes, though, is that the acceptance system is a subset of the evaluation
system.
20 Klein, in the very paragraph in which he raises his objection in Klein (forthcoming), agrees that
this must be the case, so this is a point of agreement between the objector and myself.
21 Presented, under the same title, at the occasion of the 2001 Pacific APA session honoring
Lehrer.
22 And, given that Lehrer does not insist that coherence requires consistency, it might not even be
conditions for this paper: namely, developing the coherence theory of knowledge and helping me
to develop my philosophical abilities in general, and the particular thoughts which led to this
DAVID A. TRUNCELLITO 307
paper, via his guidance as my advisor and mentor. I consider myself extremely fortunate to have
been Keith's student, and am extremely happy to have him as a friend.
I might note that as I was working on this paper, Keith came to the Mid-South to give a
talk at my University, and we spent several days talking about philosophy, walking along the
Mississippi River, listening to blues, and generally enjoying each other's company. The pleasant
memories of his visit are stilI foremost in my mind as I type these words.
308 ULTRA SYSTEM AND CONDITIONAL FALLACY
REFERENCES
Klein, Peter, forthcoming. "Coherence, Knowledge and Skepticism." (to appear in this volume)
Lehrer, Keith, 1974. Knowledge. Oxford: Clarendon Press.
Lehrer, Keith, 1990. Theory of Knowledge. Boulder, CO: Westview Press.
Lehrer, Keith, 1997. Self-Trust: A Study of Reason, Knowledge, and Autonomy. Oxford:
Clarendon Press.
Lehrer, Keith, 2000. Theory of Knowledge (2nd ed.). Boulder, CO: Westview Press.
Plantinga, Alvin, 1993. Warrant: The Current Debate. Oxford: Oxford University Press.
Shope, Robert, 1978. 'The Conditional Fallacy in Contemporary Philosophy." Journal of
Philosophy 75: 397-413.
Shope, Robert, 1983. The Analysis of Knowing: A Decade of Research. Princeton: Princeton
University Press.
Chapter 19
Keith Lehrer
University ofArizona
2. CIRCULARITY
There are a number of philosophers who have objected to the circularity
in my account of the application of the principle of trustworthiness and other
principles in the explanation of justification. These principles are not part of my
analysis or definition of knowledge and do not render the analysis or definition
circular. I contend that the explanatory loop is a virtue and not a vice. I have
drawn a distinction between explanation and proof. I have conceded that you
cannot prove something to someone who doubts your conclusion by using it as
a premise. That is not legitimate and all the rules of proof disallow it. But
reasoning has different purposes, and only one of those is proof Another is
explanation. If I seek a complete theory of justification for the purpose of
explaining why I am justified in accepting each thing that I am, in fact, justified
in accepting, then the theory, if! am justified in accepting it, must explain why
I am justified in accepting it.
I have no proof that one should seek for such a complete explanation of
justification. I affirm only that it is my goal. Now some, Bender, for example,
seem to think that I should try to prove to myselfthat I am justified, and he notes
that I cannot do that with an argument that is circular, nor can I present any
circular argument to myself that ought to convince me that I am justified I agree
with him. I hope he will not be disappointed to find me so agreeable, but I agree
with him that just as I should not expect to argue another out his doubts by a
circular argument, so I should not expect to alleviate my own doubts by a
circular argument. But suppose, as is the case, that I have no doubts about
whether I am justified in accepting some obvious claims of common sense and
only seek to explain why I am justified. Then I allege that some explanation is
obtained by arguments that are circular and that circularity is necessary in a
complete theory.
The general point might be conceded and the particular forms I have used
repudiated. Some of my critics suggest that rather than circularity of explanation
one might be better advised to affirm that some beliefs or claims or whatnot are
immediately justified. That is the strategy of the foundationalist. I have affirmed
that some beliefs are justified without argument, or if you prefer, they are
non inferentially justified. So where does a coherence theorist disagree with a
foundationalist who claims that some beliefs are immediately justified? A
coherence theorist disagrees about whether immediacy is a satisfactory
explanation of why the belief is justified. Those who claim to be satisfied with
such an explanation often have a reason for thinking special beliefs are
immediately justified. That is a kind of background theory about immediate
justifications. They have background beliefs, exactly what they are may vary,
that explain why some beliefs, the foundations, are justified in themselves. But
once that explanation is made explicit, they will be revealed as closet coherence
theorists. They will assume that such beliefs are ones that are worthy of a
KEITH LEHRER 313
person's trust without argument because they are successfully truth connected or
for some related reason. Once that assumption is made explicit and the role of it
in explanation of why we are justified in accepting those immediately justified
beliefs is acknowledged, they will emerge in their true glory as philosophers who
allege that justification can always be explained in terms of the background
system of the subject, that is, they will emerge as coherence theorists. If they
eschew that noble title, I would not dispute over the word. Call themselves what
they may, their explanations, if traced far enough, will tie their explanation
together in some virtuous loop rather than leaving immediate justification
unexplained. If not, if they choose a theory of justification in which the
immediacy of justification is unexplained, then we differ in our philosophical
objectives. They are satisfied with unexplained first explainers, and I am not. I
do not claim that they should have my goals of explanation, only that ifthey do,
they will be led to the virtuous loop.
objection because the evidence supporting the latter claim is the evidence that
Mr. Nogot owns a Ferrari. Can the objection be met in the ultrasystem? I would
argue that it cannot be met because the truth constraints on that system preclude
t-acceptance that Nogot owns a Ferrari, t-preference for accepting that Nogot
owns a Ferrari over accepting that he does not, and any t-reasoning to the
acceptance of the conclusion that Nogot owns a Ferrari.
Olsson notes that the change in the conception of the ultrasystem is
complicated by the introduction of preferences and reasonings in the evaluation
system. That is, of course, correct. It is perhaps useful to note here why I have
introduced these notions into the evaluation system. I introduced preferences to
provide an explanation for the comparative evaluations of reasonableness. If a
person prefers accepting p to accepting q and is reasonable in her preference,
then it is more reasonable for her to accept p than to accept q. So, the
reasonableness of preference can explain the comparative reasonableness of
acceptance. The addition seemed to me important because it may be more
reasonable for a person to accept p than to accept q even though the person
accepts neither of them. Thus, it was not obvious to many how an acceptance
system by itself could account for the comparative reasonableness of acceptance
even with the assumption that the person was reasonable in what she accepted.
I added reasonings to the evaluation system because it was unnatural to assume
inferential deductive and inductive closure over the contents of the acceptance
system to determine what a person knows. A person might know something if
they were capable of deducing some consequence that they do not know
precisely because the deduction is beyond their capabilities. What a person
knows depends on how they actually can reason rather than on idealized closure
relations. So, the addition of reasonings to the evaluation system is essential for
the characterization of what a person knows.
On the matter of skepticism, Olsson contends that there is a conflict in my
views. Olsson contrasts the sophisticated common sense philosopher, which he
says I am at the beginning of my book, with the moderate skeptic, which he says
I am at the end of my latest edition. In part, he has misunderstood the
sophisticated common sense philosopher, for he says,
In fact, I have never held the position that we have knowledge even if we
are systematically deceived. Truth is a necessary condition of knowledge on all
the analyses of knowledge I have offered, and it follows that if we are
systematically deceived, then what we accept is false, not true, and we lack
knowledge. So, Lehrer as a sophisticated common sense philosopher assumed
that we were not systematically deceived and, indeed, he assumed we know that
we are not systematically deceived.
I think that what misled Olsson is that he, contrary to Lehrer, holds the
view that we cannot know that we are not systematically deceived, and thinking
Lehrer must agree with that clearly correct view, altruistically attributes it to
Lehrer. Olsson writes,
Now, there is where he gets it wrong about what I think and why he thinks there
is a conflict in my view. I am a falliblist, so I agree that it is logically possible
that we are systematically deceived, but I contend that we can be both
trustworthy and reliable even though we are fallible. As a result, I think we can
know that we are not systematically deceived, that we can have empirical
evidence that we are not systematically deceived that provides us with personal
justification and undefeated justification because we are not systematically
deceived. We are fallible, admitted, and our evidence is fallible, admitted again,
but that does not mean that we have failed. We have empirical evidence that we
are not deceived, even though it is logically possible that we are deceived, for
empirical evidence is inductive only and leaves open the logical possibility of
error.
The point put simply is this. There is a possible world in which we are
systematically deceived. In that world the empirical evidence we would have that
we are not deceived would be defeated because we are deceived and ignorant of
the deception. In the actual world, the empirical evidence we have that we are
not deceived is undefeated because we are not deceived, and we know that we
KEITH LEHRER 317
are not deceived. It is wrong, though tempting, to conclude from the fact that
empirical evidence does not logically entail a conclusion, that is, that there is a
possible world in which we have the evidence even though conclusion is false,
that, therefore, the empirical evidence fails to give us knowledge that the
conclusion is true. As Olsson himself notes, an inductive argument, no matter
how strongly it supports a conclusion inductively, does not have the salient
feature of deductive validity, to wit, that it is logically impossible that the
premises are true and the conclusion false. Our evidence that we are not
systematically deceived and that the knowledge claims of the sophisticated
common sense philosopher are correct is in no way refuted or undermined by the
logical possibility that we are deceived.
Lehrer the moderate skeptic alleges that, though we are fallible and
systematic deception is logically possible, we have undefeated justification and
knowledge, as Lehrer the sophisticated common sense philosopher affirmed, of
common sense matters, including that we are not deceived. Some will say that
since we would believe that we are not demonically deceived even if we were,
our belief fails to track truth and, therefore, our evidence is inadequate. But that
is a mistake. It is the mistake of thinking that no evidence can provide us with
justification and knowledge unless the evidence logically entails what we know
or at least that the belief for which it is evidence tracks truth. Inductive evidence,
though it does not logically entail the conclusion known or insure that the belief
for which it is evidence tracks truth, can provide us with undefeated justification
and knowledge, nonetheless. Coherence combined with truth in the right way is
sufficient for us to know and know that we know even though we are fallible.
Hence I accept the consistent combination of sophisticated common sense,
embracing knowledge, including that we are not systematically deceived, with
moderate skepticism, embracing falliblism, including the logical possibility that
we are deceived. We can reach across the gap of our fallibility to the goal of
knowledge.
reply is that supervenience must hold, and we must be mistaken in thinking that
the skeptical position is possible. It is like thinking that it is possible that
identities do not hold, when it is necessary that they do. Well, why must
supervenience be true? The usual reply is justification must be based on
something. I agree it is based on something, but why must the basing relation be
metaphysically necessary supervenience? Justification is contingently based.
That is a typical way in which one thing is based on another.
Here is my view about how justification is based, when it is, on
naturalistic conditions, though, again, this view is not essential to the coherence
theory. The basing is, as far as I can see, causal. Epistemic properties are part of
the causal order. For example, sometimes a naturally formulated sequence of
formulas proving a theorem in set theory makes the theorem evident to me. So
sometimes a natural condition causes an epistemic one, and sometimes the
causation runs in the other direction. Sometimes it is the evidentness of a
proposition that causes me to believe it, for example, when the evidentness of the
theorem makes me believe it. In short, I see no reason to deny that
exemplification of epistemic properties are part of the causal order, causing and
being caused by the exemplification of other properties.
Can supervenience construed in terms of metaphysical necessity provide
a place for epistemic properties in the causal order of things, however? Should
we say that an object is caused to have epistemic properties if they supervene on
base properties a thing is caused to have? Suppose that it is necessarily the case
that some belief is evident if the belief has a base property. Suppose that,
necessarily, if I believe that I am appeared to some way and am appeared to in
that way, then it is evident to me that I am appeared to in that way. It does not
follow from this that my believing that I am being appeared to in some way and
being appeared to in that way caused it to be evident to me that I was appeared
to in that way. So evidence is not, thereby, included in the causal order.
To see that the conclusion does not follow, consider another example in
which one thing is a necessary consequence of another. Something causes me
to move by pushing me. If! move, then, necessarily, someone moves and 2+2
equals 4. Should we say that what caused me to move caused me to move and
2+2 to equal 4 just because that is a necessary consequence? I would not say so.
In general, it is incorrect to claim that if something causes an object to have a
property, then it causes it to have the necessary consequences of having that
property. So, were it a fact that epistemic properties supervened on base
properties, I would not say that something that caused a thing to have some base
property, therefore, caused it to have the supervenient property just because that
was a necessary consequence. There are replies to this objection I cannot deal
with here.
However, what is more important, and more crucial, is that supervenience
cannot offer any explanation of causation in the other direction, that is, it cannot
explain how something being evident to me, for example, how it being evident
320 LEHRER REPLIES
has no idea whether he or his thoughts about the temperature are reliable." So I
find it a mystery why he should think I would allow that Truetemp in my
example would assume that he is trustworthy in accepting these thoughts about
the temperature.
My Truetemp example suffices then to show that reliability is not
sufficient for knowledge. I concede that it is necessary. In a similar way, given
what Greco concedes, he should admit that an adequate account of agent
reliability will have to bring in background considerations about how the agent
views himself and how what he accepts coheres with these views so that he is in
a position to answer objections to what he accepts. Such agent causality
reliablism will adopt coherence as a condition of knowledge, and such reliablism
may be even be true.
the conclusion drawn by others that she lacks knowledge. But the conclusion is
incorrect, and why it is incorrect is also explained by the theory. The
explanation is that the doctor can meet the objection by neutralizing it. She is in
a position to admit that she believes what she does out of prejudice, that is, she
believes what she does in a way that is not trustworthy, but to neutralize the
objection by adding that the evidence that she knows to be true shows that the
disease is genetically specific to the race.
So far, it appears that her background system enables her to meet the
objection that the way in which she believes things is not trustworthy. Now,
however, a somewhat deeper objection arises, namely, that the doctor is not
intellectually trustworthy because of her prejudice which must be admitted to be
an objection. Can she meet that objection by neutralizing it? What can she say?
It seems to me that she can admit that she is not intellectually trustworthy
because of prejudice but that she understands in a trustworthy way that the
evidence she knows to be true shows that the disease is genetically specific to
race. Some will refuse to accept this reply as true because they will think that no
trustworthy understanding is compatible with prejudiced belief. Others will
think her reply is just. I think it is just.
My reason for thinking so is based on my distinction between the first
order state of belief and the higher order states of evaluation which include
acceptance. Belief is driven by nature, and is not immediately responsive to
reason or evidence. Acceptance is the positive evaluation in terms of evidence
and is responsive to it. It is acceptance, consequently, that I take to be what is
necessary for knowledge and which, when backed by undefeated justification,
converts to knowledge. People may be trustworthy in their understanding of
evidence and how to employ it in evaluation when belief stands fixed and
inflexible before the court of evidence. It is not only the prejudiced doctor and
gypsy lawyer whose belief formation and retention is sometimes unresponsive
to reason. The best we can do is to evaluate belief, accept it or reject it in a
trustworthy way as we consider the evidence, and that is the way to knowledge.
To expect that we shall cease to believe what he reject in a trustworthy way is,
unfortunately, naive, and to expect that we shall always come to believe what
we accept in this way is also naive. I confess that I did not appreciate how belief
and acceptance may separate. Contrary to what I said, we may accept something
without coming to believe it. Belief sometimes lags behind the positive
evaluations of reasons, or worse yet, ignores them.
One final question might be raised by the distinction between acceptance
and belief. It might be asked whether acceptance on the basis of evidence
requires that the evidence cause or causally sustain acceptance. As in the case of
belief, evidence or our attitude toward it, may cause us to accept what we do, but
it need not. Acceptance is positive evaluation and understanding ofthe nexus, the
relationship, between evidence and what is accepted. Evaluation suffices for
acceptance. However, as I note in reply to Stewart, there remains a question
324 LEHRER REPLIES
win. Of course, one is not obliged to adopt that view about what is reasonable,
and one might find a justification for stopping at some point. But the point is
that one needs a justification, as Ross would agree.
Now, however, one can see the advantage of the account of justification
in terms of neutralization, for it allows us to explain a difference in views, a
difference in intuitions, in the case of the lottery. Some are inclined to affirm, in
spite of my reasoning to the contrary, that they are justified in accepting that the
ticket they hold in a large lottery will not win, and, indeed, that they know the
ticket is not the winning ticket. Others hold another view and have a different
intuition. The difference, I suggest, depends on whether their views about
reasonableness allow that the competitor or objection that it is just as reasonable,
and on the same grounds, to accept that each other ticket will not win (including,
of course, the ticket that will, in fact, win) can be neutralized. So, in fact, I take
it to be an advantage of my account that it explains the views about justification
and knowledge that differ from my own in terms of whether the neutralization
succeeds.
The questions Ross poses concerning what is reasonable raises issues
about impartiality and consistency. My view is that it may be reasonable to
accept an inconsistent set of statements or even a complicated statement that is
self-contradictory, for, as I have noted before, it may be reasonable, though false,
to accept that such a set of statements is consistent. Our most reasonable efforts
to be consistent can fail. We are fallible about consistency. This does not mean,
however, that consistency is not an objective of rationality. I think it is.
Similarly, there may be cases in which one is reasonable to accept a set of
statements that are not accepted with impartiality. For, it may be reasonable,
though false, to accept that one has accepted what one has with impartiality. Our
most reasonable efforts to be impartial can fail. We are not infallible about
impartiality either. Moreover, our objectives contain internal stress, for example,
our objectives of not accepting what is false and our objective of accepting what
is true. So there are multiple objectives that may come into conflict as we seek
to satisfy them all. What is the reasonable policy? It is to avoid any general
strategy that we know with certainty will defeat our objectives. That means we
should avoid any strategy that we know with certain will lead us into
inconsistency or impartiality. We may not be able to avoid them, since we are
fallible in our objectives, but that does not mean that it is reasonable to adopt a
strategy that we are certain will lead to failure. We do not have to be infallible
to be reasonable, but we should seek to avoid the certainty of failure in seeking
the objectives of reason in order to be reasonable. That is why I prefer to forego
accepting that my ticket will not win the lottery. If! do accept that my ticket will
not win the lottery, I am certain I will either fail to be consistent or fail to be
impartial. I probably will fail anyway, but I do not want to make that certain.
Finally, what about the claim that viewed with sufficient generality,
acceptance is always a lottery, at least with respect to the totality of what we
KEITH LEHRER 329
to
to
4.p
knew that he was Russell and much else beside on the basis of his background
system which, because of what he accepted about set theory, was inconsistent.
We know things on the basis of inconsistent background systems, that is, on the
basis of background systems in which the claims that we accept are inconsistent
as a set.
Moreover, it is obvious, once inconsistency is allowed, that deductive
closure is not harmlessly imposed but epistemically disastrous. If Russell had
deductively closed his inconsistent system, he would have included the claim
that he was not Russell along with the claim that he was and would have fallen
ignorant about even the simple fact that he was Russell. Deductive closure is not
harmless, however, even when a background system is consistent. A reasonable
person, one who is interested in proceeding in a reasonable way, must constrain
the impulse to deduce. A reasonable person will be ready to consider objections
to what she considers and to consider the acceptance of new information, in
short, she will be ready to revise. Now, the actual person, unlike the idealized
one, is hindered rather than assisted by increasing the size of her acceptance
system. So, leading the life of reason, the model person will not deductively
close her system. She will, instead keep the system as simple as she reasonably
can to facilitate a survey of what is contained therein.
To be sure, there is a conflict within reason between keeping her system
simple and deducing the consequences thereof in dealing with revision in a
reasonable way. Ifshe does not deduce the consequences of what she accepts, of
some simple representation of the content thereof, she may fail to note that what
she accepts is inconsistent with some new information and fail to revise when
reason calls for it. Consistency, though not a reasonable constraint on a
background system, is, nevertheless, a desideratum, and deduction of
consequences is important for noticing inconsistency. I am afraid that the upshot
may seem disappointing to Spohn. Consistency is a desideratum but not a
necessary condition for a background system that generates reasonable
acceptance,justified acceptance and knowledge. As a result, deduction, however
important it may be in the life of reason, can be harmful to reasonable revision
of a background system when used to close a background system which is
inconsistent, though effective for generating knowledge, and render it useless for
generating that knowledge. As I have illustrated elsewhere, moreover, there are
cases in which a background system that is inconsistent may be more reasonable
to accept than any consistent subset of it because the choice of a subset would be
arbitrary and would ignore important information.
So how can coherence generate justification and knowledge when the
background system of acceptances is inconsistent in terms of the accepted
claims? It is because coherence is not a property of the background system but
a relationship of the background system to some target acceptance. The target
acceptance coheres with the background system just in case the background
system suffices to meet all objections to the target acceptance. The inconsistency
KEITH LEHRER 333
The formal result that Wagner has contributed in his essay reminds me
how much I enjoyed our previous collaboration even though we have gone in
somewhat different directions subsequently. The book that we wrote together is
one that I continue to find gratifying and continue to use in my philosophical
writings. I am very grateful that we were able to collaborate on the book together
334 LEHRER REPLIES
and that I was able to benefit from his exceptional mathematical and
philosophical talents. His focus on probabilism in the present paper affords me
with the opportunity to explain how I see the connection between our work
together as well as his present work and the theory of knowledge.
A number of philosophers, social scientists and mathematicians have
tended to see theories of categorical or rational acceptance and theories of
degrees of acceptance or personal probability as competing explanatory theories,
Richard Jeffrey, most especially. On the face of it, they are complimentary rather
than competing for the simple reason that degrees of acceptance or probability
would be a determining factor in categorical acceptance or rational acceptance.
Looked at from the standpoint of cognitive psychology, probabilistic vectors are
plausibly regarded as a component in categorical decisions whether those are
decisions to act in the external world or to arrive at some internal choice, for
example, to accept something. The assumption that probabilism competes with
a theory of rational acceptance results from adding to probabilism and
Bayesianism the addendum, and that is all there is to human thought, action and
rationality. The addendum seems to me to be false. The addendum is supported
if one accepts behaviorism, but that is not something I would accept. So, I think
that probabilism has an important contribution to make, but that is not all there
is to it. Science as well as everyday life is rife with categorical claims expressing
acceptance. These must, of course, be supplemented with impartial
consideration of objections to those claims and the attempt to meet those
objections in an impartial manner to avoid dogmatism and error. That is part of
the structure of rational thought and discourse, and that structure is what I have
sought to refine in the theory of justification and knowledge I have articulated.
So why do I think that probability theory is not adequate to model all this
with only minor modification? The reason is grounded in our psychology,
however admirable or deplorable that might be. We do accept some claims and
reject others. But there is a theoretical point of interest, namely, if we seek to
avoid error in what we accept considering only probability, then we should take
no risk in what we accept and restrict ourselves to what has a probability of one.
But we do not want to do that in science or everyday life. So, something other
than probability is entering into our rational decisions of what to accept. Some
have thought that informational content is the other factor, and I was inclined
to concur, but I now think that more systematic considerations, systematic
explanatory power and coherence, for example, are important as well.
Moreover, just as in practical matters, we ultimately choose to accept one action
or another, so in intellectual matters we choose to accept one claim or another.
Whether the choice is practical or intellectual, we avoid dogmatism by remaining
open to reconsideration, but there is an advantage to making a choice, and choice
is categorical, even if subject to revision and reconsideration. The advantages
of choice are illustrated very nicely by the choice of Bayesian ism by those who
KEITH LEHRER 335
have developed the theory. Acceptance of the theory motivates them and drives
their creative construction. And that is how we work.
What is the contribution of probabilism to my theory of knowledge? One
factor in comparative reasonableness of acceptance, which corresponds to
reasonable preference over acceptance, is probability. Comparative
reasonableness of acceptance cannot be equated with degree of probability
because that would lead us to the result that it is always more reasonable to
accept the more probable hypothesis, to stick to the data rather than to
extrapolate to some unifYing explanation of it. The data will be more probable
than the theory which, like all theories, has a high probability of turning out to
be in error. Nevertheless, probability is an important factor, because it gives us
a measure of our risk of error. We do seek to avoid error. Whatever other
objectives we have and no matter how much weight we give to those objectives,
we must give some weight to avoiding error, and, hence to probability.
Moreover, as Wagner notes, if the superior theory of probability proves to be one
employing nonadditive upper and lower probabilities, the advantage of such a
theory is that it allows us to employ and socially aggregate a wider base of
information, including consensual information. Such probabilism imposes an
important constraint on reasonableness. Moreover, probabilism has provided us
with an illuminating account of the dynamics of probability change to which
Wagner has brilliantly contributed. It remains an open question in my mind how
much of the story of reason, justification and knowledge can be accounted for by
the development of probability theory.
him to think that the truth about the world and ourselves should not be assumed
to conform to our wishes for simplicity. That may also have been one reason why
he did not seek to reduce all his principles of contingent truth to principle 7 or
any other for the sake of simplicity.
I am much indebted to Van Cleve for the clarity he has brought to these
issues in his discussion and hope that I may have done something by way of
answering his inquiries. I acknowledge that I have not replied to all of them, and
that is only because I do not have the answers to them all. Van Cleve provides
us with an interpretation of Reid of great cogency, and where we disagree, I am
inclined to find the textual evidence inconclusive. I find the multiple
interpretations of Reid to be a consequence of both the subtlety and modesty of
Reid. When his intellect could take him no further, he fell silent.
Manning has argued in detail and with careful attention to my text that my
account is defective. I am impressed with his philosophical acumen. His
arguments are challenging and perceptive, but admit of reply. Manning argues
against my position in various ways, but there are three central issues, and I
believe all his arguments depend on these. He claims that coherence theories in
general, and mine in particular, fail to fulfill a condition of adequacy for a theory
340 LEHRER REPLIES
assume some of the blame, but my definitions of coherence are not open to the
objection that coherence, on my theory, is defined in terms of inference, though
inference may yield coherence and justification in some cases. So I can embrace
the advantages of the foundation theory to the extent that I share the claim with
the foundationalist that some beliefs are noninferentially justified. I disagree
with the foundationalist who claims that such justification is independent of
other things that the person accepts. I contend that the justification depends on
coherence with other things the person accepts. I assume that is a disagreement
of philosophical substance.
Now let us tum to the truth connection objection. Here I can be brief.
There are two sides or aspects to justification that I have separated in my theory
ofjustification, the personal aspect and the undefeated or, as I now prefer to say,
the irrefutable aspect. These two aspect are, in fact, combined in undefeated or
irrefutable justification, for undefeated or irrefutable justification is personal
justification that does not rest on errors in the background system which yields
that justification. It is, however, irrefutable justification that is the condition that
converts acceptance of truth to knowledge. If one chooses, as Plantinga does, to
call the conversion condition warrant, then my view is that warrant is irrefutable
justification. So what is the connection with truth? The basic idea is that there
is a subjective aspect and an objective aspect to the truth connection in
irrefutable justification. The subjective aspect is that I aim at accepting things
just in case they are true. Moreover, I accept that I proceed in a trustworthy way
that reliably leads me to truth. That is what I accept and is the connection with
truth on the subjective side. The objective side is that my acceptance of this must
be true, that is, it must be true that I proceed in a trustworthy way that reliably
leads me to truth. For, if that is untrue, then the personal justification I have
will be defeated or refuted by errors, untruths, in my background system.
Manning says that truth conduciveness must come into justification and
notes that I agree. My explanation of how truth connection comes in is, therefore,
twofold. On the subjective side, the subject of knowledge accepts that there is a
truth connection to yield personal justification, and, on the objective side, the
subject must be right in accepting the truth connection to have an undefeated or
irrefutable justification which converts to knowledge. If Manning demands
more of a truth connection than that, I must leave him to satisfy his own
demands, for I think there is no more satisfaction to be had.
On the matter of circularity, notice, first of all, that there is no circularity
in the definitions I give ofpersonaljustification and irrefutable justification. The
analysis is not circular. I have claimed, however, that the principle of
trustworthiness in what I accept has the merit of explaining why I am trustworthy
in accepting what I do. Therefore, when I accept that I am trustworthy in
accepting what I do, this principle of trustworthiness explains why I am
trustworthy in accepting it as well as the other things I accept. So, what is his
objection? As far as I can see, it is that he disallows circular explanation. Well,
342 LEHRER REPLIES
that is his fiat, but why should I allow that to settle the matter? We need an
argument. Here is one that I gave and give that has special relevance to
justification. Suppose you are seeking a complete theory to explain why a
person is justified in accepting everything that they are justified in accepting.
Assuming the theory is finite and a person is justified in accepting it, then the
theory must explain why the person is justified in accepting it. Now, as to the
question of whether using a theory of justification to explain why you are
justified in accepting the theory is vicious, I do not see why I should find it
vicious if I am only seeking to explain the matter to myself. For I consider
myselfjustified in accepting the theory. Why shouldn't I use the theory to explain
to myself why I am justified in accepting it? I use it to explain why I am justified
in accepting everything else I consider myself justified in accepting
Manning objects to this appeal to my trustworthiness because he objects
to the inference from
to
2. p.
The argument is not a valid deductive one. Moreover, he may object even if! add
the premises
and even,
that the argument from all the premises above to the conclusion that p is still not
a valid deductive one, and he would be right. However, when I reflect on the
matter for my purposes, when I reason to see where I am led from what I accept,
then, I may appeal to what I accept in my reasoning. For what I accept represents
my best efforts to obtain truth and avoid error. My reasoning from what I accept
has all the hazards of nondeductive reasoning, but those risks are a good gamble
if I am circumspect. It is, of course, a defeasible kind of reasoning, and there are
times when the inference is defeated by further considerations, but that is the
character of defeasible reasoning.
The matter is of general interest, so let me elaborate further. Manning
objects in detail to deriving any conclusions about the truth of any claim from the
KEITH LEHRER 343
from it, or I refuse to venture beyond the premises affirming that I accept various
things and limit my conclusions about the world to egocentric observation about
my own acceptances. I hope he will excuse me if I venture forth to the
conclusions beyond my acceptances. Whatever he says in his philosophy, I am
sure that in practice he does the same.
I have said more about circularity in reply to others as it arises concerning
trustworthiness. In conclusion, however, I wish to reiterate that my definitions
of knowledge and justification are not themselves circular. I could retain those
definitions and give up the principle of trustworthiness and the use of that
principle to explain why a person is justified in accepting it. I defend the
principle of trustworthiness and the application of that principle to itself because
it seems to me to be true and the application of it explanatory. It is not necessary
to my analysis of knowledge.
Rott has developed a very valuable theory of the dynamics of belief which
he has sought to connect with the synchronic theory of knowledge that I have
constructed. I read his paper after replying to some of the others because of
difficulty opening his file, and some of his objections and my replies are
contained in other replies to which I refer the reader to avoid some duplication.
He has clearly found my 1990 theory of knowledge useful to his own purposes,
and he is welcome to work with the earlier account of knowledge and to amend
it for his own purposes. It may, after all, turn out that the earlier work will have
an importance that I did not perceive and a useful application that I did not
intend. I would welcome that.
However, I did not, in fact, intend the use of the members of the
ultrasystem in the 1990 and earlier accounts to be viewed dynamically, I
regarded each member of that system, which corrected errors in the acceptance
system in different ways, to provide tests at a time for whether the justification
a person has at a time depends on his accepting things that are false. There is
nothing dynamic in that account, though, again, Rott is welcome to reinterpret
what I have said in any way that he finds useful. I shall not, however, comment
on Rott's discussion of the 1990 account ofthe ultrasystem, for I think that Haas
has shown that it is defective formally and Rosenthal has a good counterexample
to it. I refer the reader to those articles and my replies.
Now Rott raises the objection that in my earlier account, 1990, I did not
say enough about what is involved in it being more reasonable for a person to
accept one thing than another, and especially, on some assumption the person
makes. I hoped to offer some illumination in 2000 by introducing the notion of
preferences over acceptances, that is, preferring accepting one thing to another,
as a component in the evaluation system. I then added the argument, made more
KEITH LEHRER 345
giving priority to one over the other. What we accept rationally influences what
methods we accept just as what methods we accept influences what we accept.
Neither comes before the other. They are tied together in a loop of our
trustworthiness.
This is a clock. It shows ten minutes past three. Most clocks work
properly most of the time and therefore show the right time most of
the time. So I conclude that this clock shows the right time right
now and believe that it is ten minutes past three.
clock is running to accepting that the clock is not running or the person is not
justified. It is that preference that makes it more reasonable for the person to
accept that it is more reasonable for him to accept that the clock is running than
to accept that the clock is not running. But that preference, the preference to
accept that the clock is running over accepting that it is not running is not a t-
preference, because it involves preferring to accept something false, that the
clock is running, over something true, that it is not running. Therefore, this
preference, a crucial one for meeting the objection that the clock is not running,
will not be contained as a t-state in the ultrasystem, and the objection that the
clock is not running cannot be met on that system. So the justification would be
defeated.
Now Rosenthal might wonder why, if the reasoning he suggests above
remains in the ultrasystem, that would not suffice for justification? My answer
is that it might suffice for some kind of justification, it does supply some
inductive confirmation for the conclusion, but more than that is required for the
kind ofjustification that converts to knowledge. It does not suffice that one have
some reasoning to support the target acceptance. One must also be in a position
to meet objections to one's reasoning on the basis of one's ultrasystem. One
objection to the reasoning is that the clock is not running, and that cannot be
answered on the basis of the ultrasystem. It is important to notice, moreover, that
states of acceptance, preference over acceptance and reasoning with acceptance
need not be something that is occurring or has occurred to the subject. They are
functional states that provide a capacity to respond to objections which, having
not been raised, have not elicited response. But the system must provide the
materials, it must have the right evaluative stuff, to answer objections. That is
missing in the example from Russell as Rosenthal presents it.
Moreover, a similar point, which I have made elsewhere, arises with
respect to the Chisholm example of the man who takes a Terrier to be a sheep.
One objection to what he accepts, namely, that there is a sheep in the meadow,
is that what he takes to be a sheep is not a sheep. To meet the objection, the
person must prefer accepting that what he takes to be sheep is a sheep over
accepting that what he takes to be a sheep is not a sheep, and that preference is
not a t-state because it involves preferring accepting something false over
accepting something true. Again, the absence of the t-preference in the
ultrasystem will lead to the defeat of the justification on that system.
Finally, what should we say about the proposal that Rosenthal makes? It
is a very good one. The idea is that one of the original justifications a person has
must remain when the acceptance of what is false is systematically replaced with
acceptance of what is true. One problem, which led me to require that t-states
and unmarked states both remain in the ultrasystem, is the problem of knowing
that you accept things, when, unknown to you, what you accept is false. This
problem is raised by Klein and answered by Truncellito in this volume. Suppose
that a person seeing the clock in Russell's example accepts that the clock is
350 LEHRER REPLIES
running. He would not know that the clock is running, but he would know he
accepts that the clock is running. Yet Rosenthal would require that the man be
justified in accepting that he accepts that the clock is running relative to a system
in which acceptances of falsehoods are replaced with acceptances of truths, and,
therefore, in which he accepts that the clock is not running. How can the person
know that he accepts that the clock is running relative to a system in which he
accepts that the clock is not running? Rosenthal might well have an answer to
this, but I do not see what it is. Such problems have led me to abandon the
requirement of replacing false acceptances with true acceptances and to turn to
an improvement on accounts in which the acceptance of falsehoods are
eliminated.
Bender, whose philosophical talents I have long admired and enjoyed, has
written a studied critique based on careful reading of the texts. So it is a special
challenge to try to deal with it. I think that some misunderstanding persists, and
that is, no doubt, my fault. At any rate, his citation of what I have said provides
anyone with a clear basis for deciding whether my reply is already contained in
what I had written previously or contains some novelty. My basic claim is that
internal coherence yields knowledge when combined with systematic truth that
shields the coherence from refutation based solely on errors contained within the
background system. I argued, as Bender accurately notes, that the subject must
accept that she is trustworthy in what she accepts and, moreover, that this
trustworthiness is truth connected in a reliable manner. Furthermore, I have
contended that the person must, in the case of a particular target acceptance that
p, be right because of the trustworthy way in which she accepted it.
Bender focuses on an argument from the general trustworthiness of
acceptance to the trustworthiness of a particular acceptance that p and, further,
to the reasonableness of accepting that p. Now Bender objects that the argument
begs the question against a serious philosophical skeptic. But I agree with his
claim and have said so. As an argument against a serious philosophical skeptic,
it is a petitio to argue for the reasonableness of accepting that one is trustworthy
in what one accepts by appealing to the premise that one is trustworthy in what
one accepts. That is why I have insisted, contrary to such an admirable defender
of common sense as Moore, that one cannot prove that the skeptic is wrong.
Bender says that I have not established the claims needed to prove that the
philosophical skeptic is wrong. I completely agree with him, for I see no
difference between proving that skeptic is wrong and establishing what would
be required to prove that the skeptic is wrong. That is not what I was trying to do
in my later work, and I agree it cannot be done.
KEITH LEHRER 351
However, I now deny that I must be able to prove that the skeptic is
wrong, when he denies that I know that p, in order for me to know that p.
Though I cannot prove that the skeptic is wrong, I can know that the skeptic is
wrong, and, indeed, I can know that I know this. Bender also wishes to deny
this. His argument seems to rest on agreement that the principle of
trustworthiness of acceptance is needed for the evaluation system, containing
what a person accepts, but on his denial that a person can be justified in
accepting that he is trustworthy by the reasons that he has for accepting that he
is trustworthy. He puts this by saying that a person is not epistemically justified,
by which I think he means, I am not certain of this, justified in a way that would
satisfy a skeptic with opposing views, that is, satisfy the skeptic by showing the
skeptic that he was wrong. But again, I agree that a person is not justified in this
sense, for I agree that we cannot prove the skeptic wrong without begging the
question. So, there just is no issue about this between us. What, then, is the
issue? In part, I think there is a verbal problem which I created by an unfortunate
use of the word "skeptic" in the first edition of Theory of Knowledge which I
corrected in the second edition, namely, the use of the word "skeptic" in the
justification games that I used as a heuristic to assist a student reader to
understand personal justification and ultrajustification. I rectified this by
speaking of a critic rather than a skeptic in the second edition. I did not intend
that the critic be regarded as a philosophical skeptic. I can examine my views
critically without taking on the heavy mantel of a philosophical skeptic. To be
justified, my views must pass the test of my own critical examination of them.
They need not pass the test of Descartes exercising hyperbolic doubt, however.
But verbal misunderstandings aside, there is an issue of fundamental
importance between us. I think of justified acceptance, personally justified
acceptance, in my technical expression, as acceptance that has the capacity to
answer the objections to it in terms of the background system of the subject.
Suppose I consider whether I am justified in accepting something. What will I
do? I will consider other things that I accept, my preferences concerning
acceptances and my reasonings in the matter. What else can I do? Now, as I
consider critically my acceptance, I consider objections, those I might imagine
raised by in internal critic, and how I can reply in terms of what I accept, prefer
to accept and my reasonings about the matter. Again, what else should I do? If
I cannot appeal to what I accept, for example, I am immediately rendered mute,
and my attempt to proceed critically ended before it begins. Now, I agree that in
proceeding in this way, I am accepting that I am worthy of my trust in what I
accept, in my preferences over acceptances and my reasonings concerning
acceptance.
One question I might critically raise is what reasoning I might use to
defend my acceptance of my trustworthiness. There are two answers. One is my
success in obtaining truth and avoiding error in what I accept. This is induction
from other cases. The other answer is one that arises when I consider why I need
352 LEHRER REPLIES
The person may use the premise that the Biblical acceptances are
trustworthy as one of his reasons, to be sure, but it is not a reason for accepting
the conclusion because it is false. In one crucially important sense of "reason"
the person has failed to provide a reason for accepting his conclusion. For one
defect of a premise that undermines its claim to be a reason for accepting a
conclusion is simply the falsity of premise. We must be careful here not to get
entangled in verbal disputes over the word "reason," for it is an ambiguous
notion, but an unsound argument does not establish the conclusion, and false
premises make an argument unsound. Bender knows that. Truth matters. I
allege that it matters enough to convert personal justification into knowledge.
The truth of the premise that one is trustworthy in what one accepts and that
trustworthiness is successfully truth connected can make the difference that
converts internal coherence into knowledge.
I do not claim to have refuted Bender's claim that this does not prove the
skeptic wrong. I do claim to have refuted Bender's claim that my account of the
conditions of knowledge is insufficient for knowledge if it does not prove the
skeptic wrong. Coherence plus truth, especially concerning trustworthiness and
reliability, though not only concerning that, explain why we know.
Finally, I do think that Bender has put forth a position and defended it
admirably concerning the relationship between genuine skeptical objections and
knowledge. We might call it the impartiality condition, namely, that the reply to
any objection sufficient to defend a claim to knowledge must be impartially
effective. It is a sufficient reply to one person only if it is a sufficient reply to
another. This is a most plausible claim, but one I reject. Whether a reply to an
objection is sufficient for a person depends on what that person accepts. What
one person accepts, another does not. So my reply to an objection might be
sufficient for one person but not another. More specifically, my reply to an
objection I consider may be sufficient for me because of what I accept though
not sufficient for the genuine philosophical skeptic.
to what I have written as Klein is misled in some way, there must be some fault
in what I said, and it is useful to attempt to correct that.
Here are my replies. First of all, consider the problem about knowing that
one accepts something one does, in fact, accept, that p, when, as it turns out, p
is false. The reply I would offer and endorse is the one that Truncellito supplies.
The uitrasystem does not replace my acceptance system but is, instead, a
theoretical construct used to test whether justification depends on error and is,
consequently, defeated by the error. As Truncellito notes, I do, in fact, accept
that p, that acceptance is part of my acceptance system, so when I accept that I
accept that p, what I accept is true, not false. The problem Klein raises is an
interesting one, and it has led me to make explicit that justification based on the
uitrasystem must acknowledge the acceptances in the original acceptance system,
for it is true that I accept those things. The restriction placed on acceptances used
in the ultrasystem to meet objections, that is, for coherence with the system, is
that the content falsely accepted, the content that p, when p is false, cannot be
used to meet objections. But my acceptance that p, though not the content, p,
itself, can be used, for it is true, not false, that I accept that p.
Now for the case ofD. Pendable. First of all, as I note elsewhere, I do not
assume that acceptance systems are deductively closed. That would be
unrealistic. No one deduces all the deductive consequences of what they accept.
Nevertheless, it would be a small modification of the example to suppose that
someone deduced from the claim that D. Pendable will be at the meeting that he
will not die of a heart attack before the meeting. Now is this a problem for my
theory? It is not. I think that people would divide in their intuitions about what
they know about the future, especially in such cases. Some would think that,
though there is no reason to think the man will die of a heart attack, there is an
objection to the claim that he will not, namely, that people do die of heart attacks
even when there is no reason to expect that they will. Now, I would say that such
an objection to the claim that the person will not die of a heart attack before the
meeting is also an objection to the claim that the person will attend the meeting.
Dead men do not attend the AP A. So, do we know that D. Pendable will not die
of a heart attack and will attend the AP A meeting, or should we allow that the
objection that people do die of heart attacks, even when there is no reason to
expect that they will, cannot be met and, therefore, that we lack knowledge?
I think that the answer depends on the success of neutralization for the
subject. Some will be inclined to reply to the objection-people do die of heart
attacks when there is no reason to expect they will-that D. Pendable will not
have a heart attack before the meeting and think it just as reasonable to accept
that in conjunction with the objection as to accept the objection alone. They will
consider the objection neutralized and affirm that they know. Others will say
there is no way of telling about whether D. Pendable will have a heart attack, and
so they will say it is more reasonable to accept the objection then to accept the
conjunction of it with the disclaimer that D. Pendable will not have a heart
KEITH LEHRER 355
attack. All I will say in defense of my account is that it explains in the right way
differing intuitions in terms of what a person thinks about what is reasonable and
how reasonable it is to accept that D. Pendable will not have a heart attack before
the meeting. That seems a merit of the account.
Now for the objection that my account cannot deal with the closed box
example. Klein assumes, though I would not, that if a person accepts something,
then, if what they accept is true, they can justify anything that follows from it.
I would say that what matters is whether they can meet objections to the claim
in terms of their evaluation system. One objection that is relevant in this case is
that the person does not have any reason for accepting that there is a triangle in
the box. The person can only meet that objection by replying that Sally told him
there is, but, in fact, she did not, and that justification would be defeated. It is not
adequate for the purposes of justification that the person accept that there is a
triangle in the box and be right about that when he lacks a justification that is
undefeated.
Now Klein thinks I cannot make this reply because there are cases, those
of distinct memory, I suppose, when a person knows something even though they
have forgotten the evidence. But I would say that even in such cases acceptance
is not sufficient. In addition, the person must evaluate the memory as being based
on correct information, even if the person no longer remembers the source of it.
Again, that is required for the purposes of justification, for it is an objection to
any memory belief that it is not based on correct information. Now if the
memory is based on correct information, the justification will be sustained in the
ultrasystem, otherwise it will be defeated.
Finally, concerning skepticism, I do hold, contrary to what Klein alleges,
that my evaluation system must contain the capacity to meet all objections, but
that might in some case rest primarily on my preferring to accept what I do, that
I perceive a zebra, rather than accepting that I am deceived, or any other
objection contradicting what I accept. I agree with much of what Klein says
about skepticism. Some of his criticism of my views on skepticism are due to a
misunderstanding I caused by an ambiguous use of the word "skeptic" in the first
edition of Theory of Knowledge. I used the term to represent a kind of internal
critic raising objections, as well as to represent a serious external skeptic. In the
second edition, I changed the expression used to characterize the internal critic
to "critic" hoping to avoid the misunderstanding.
My view about the external skeptic is that we can know, and know that we
know that he is mistaken, but that we must appeal to what we accept as a source
of reasons to argue against him. I assumed that the external skeptic would note
that I must accept that I am worthy of my own trust in what I accept to argue
against him. If the skeptic denies this, I may, as Klein suggests, ask him for his
reasons. Suppose he has none but asks if we have any for thinking the opposite.
Ifwe have none, and he has none for his opposing position, then, it seems to me,
we should concede that the issue is a draw. If we attempt to argue against him,
356 LEHRER REPLIES
ENDNOTES
* I would add that another article appeared in an undergraduate journal of philosophy, Stoa, Vol.
2, Issue 2, 2001, written by Jonathan Tweedale, entitled, "Epistemic Humility and Undefeated
Justification," arguing in a similar way that my account did not fall victim to the problem about
knowing that one accepts something when, unknown to one, what one accepts is false. He does
not discuss Shope, but his defense, though that of an undergraduate, is professionally written.