You are on page 1of 10
—™ | } 4 [ESE TRANSACTIONS ON SYSTENS, MAN, AND CYBERNETICS, VOL 84-13, No, 3, MAY/7UNE 1983 287 Skills, Rules, and Knowledge; Signals, Signs, and Symbols, and Other Distinctions in Human Performance Models JENS RASMUSSEN, SENIOR MEMBER, TEE Abstract—The introduction of information technology based on digital ‘computers for the design of man-machine interface systems has led to a requirement for consistent models of human performance in routine task ‘environments and during unfamiliar task conditions. A discussion Is pre- sented ofthe requirement for different types of models for representing perforosnce atthe skill, ule, and knowledge-based levels, together with review of the different ways in which Information is perelved at these “ferent levels in terms of signals, signs and symbols. Pariular attention 's paid to the diferent possble ways of representing sytem properties hich underlie knowledge-based performance and which can be char- Aetorized_at several levels of abstraction—from the representation of physical form, through functional representation to representation in terms of intention or purpose. Furthermore, the role of qualitative and quanti. tive models inthe design and evaluation of interface systems is mentioned, and the ned to consider such distinctions carefully discussed, Intropuction ANY technical systems in modern times are highly automated and do not rely on human intervention in the control of normally planned functions. Yet their existence depends on extensive support from a human staff 10 maintain the necessary conditions for satisfactory opers- tion and to cope with all the badly structured and probably unforeseen states of affairs in the system. Due to the high risk involved in the potential for accidents in large central- ized production units, concern with being able to predict human performance during complex rare events has in- creased. We therefore need systematic descriptions of human performance in total, from the observation of infor- ‘mation to the physical actions on the process plant, and the descriptions should cover a wide range of work situa- tions from daily routine to stressed encounters with acci- dental events. ‘We need tools for reliable prediction of human perfor- mance and of the various error modes for this purpose. A. long tradition exists within vehicle control to use quantita- tive models for systems design and performance analysis, such as the models based on optimal control theory. Dur- ing recent years, attempts have been made to extend these models to higher level human decisionmaking to conform ‘with the increasing levels of automation in aviation, and to transfer such models for process control applications. Whether or not this approach is fruitful depends on the Manuscript received Apeil 1, 1982; revised December 10, 1982, ‘The author is with the Riso! National Laboratory, DK 4000 Roskié Denmark nature of the human task. The optimal control part of the ‘model may not be needed if the manual acts are no longer an integral part of the control task but merely a general interface manipulation skill. In that case, independent de- velopment of a decision model may lead to a more direct approach. What we need is not a global quantitative model of human performance but a set of models which is reliable for defined categories of work conditions together with qualitative framework describing and defining their cover- age and relationships. In some areas, particularly in reli- ability engineering, several premature attempts have been made to quantify human performance due to the pressing need for prediction. This tendency to rush to measurement and quantification is, however, not only a modern trait of engineers. Indeed, the stranger in Plato's Statesman re- marked: ‘There are many accomplished men, Socrates, who say, believing themselves to speak wisely, that the art of mea surement is universal, and has to do with all things.... But these persons, because they are not accustomed to di tinguish classes according to real forms, jumble together two widely different things, relating to one another, and to 1 standard, under the idea that they are the same, and also fall into the converse error of dividing otter things not according to their real parts ‘The aim of the present paper is to discuss some basic distinctions which are useful in defining the categories of human performance for which separate development of models is feasible. In this effort we have to consider that humans are not simply deterministic input-output devices but goal-oriented creatures who actively select their goals and seek the relevant information. The behavior of humans is teleological by nature. In their classical paper Rosen- bluth and Wiener {1} define teleological behavior as behay- ior which is modified during its course by signals from the goal. This restrictive definition seems, however, to be due to an inadequate distinction between two concepts: causes of physical events and reasons for physical functions, a distinction which has been discussed in detail by Polanyi [2], Teleological behavior is not necessarily dependent on feedback during its course but on the experience from previous attempts, ie., the reason for choosing the particu- lar approach. Reasons act as the classical “final causes” and can control funetions of behavior systems by selection, 0018-9472 /83 /0500-0257801.00 ©1983 IEEE 238 swans, sensony PUT Fe 1 [EEE TRANSACTIONS ON SYSTEAS, MAN, AND CYBERNETICS, OL. SMC-13, NO. 3, MAY/JUNE 1983 Sexson Horor ‘Simplitied lustration of throe levels of performance of skilled hurnan operators, Note that levels are not ateratives but interact in a way only rudimentarly represented in diagram. be it natural selection in biological evolution or through human design choices for man-made systems. Causes, on the other hand, control functions through the physical structure of the system, Since all technical systems are designed for very definite reasons, it follows directly that teleological explanations—in the classical sense—of the functions of man-made systems derived from their ultimate purpose are as important as causal explanations based on engineering analysis. The same is the case for explanations of purposive human behavior. Actually, even human position and movement in the physical environment ate only occasionally directly con- trolled during the course of action by simple feedback, It may be the case in unfamiliar situations calling for accu- rate and slow time-space coordination, but in more com- plex rapid sequences, the sensory equipment is too slow for direct feedback correction, and adaptation is based on means for selection and regeneration of successful patterns of behavior for use in subsequent situations, ie, on an internal dynamic world model ‘AL a higher level of conscious planning, most human activity depends upon a rather complex sequence of activi- ties, and feedback correction during the course of behavior from mismatch between goal and final outcome will there- fore be too inefficient, since in many cases it would lead to a strategy of blind search. Human cctivity in a familiar environment will not be goal-controlled; rather, it will be oriented towards the goal and controlled by a set of rules which has proven successful previously. In unfamiliar situations when proven rules are not available, behavior may be goal-controlled in the sense that different attempts are made to reach the goal, and a successful sequence is then selected. Typically, however, the attempts to reach the goal are not performed in reality, but internally as a problem-solving exercise, ie, the successful sequence is selected from experiments with an internal representation or model of the properties and behavior of the environ- ment, The efficiency of humans in coping with complexity is largely due to the availability of a large repertoire of different mental representations of the environment from which rules to control behavior can be generated ad hoc. An analysis of the form of these mental models is im- portant to the study of human interaction with complex ‘man-made systems, Basically, meaningful interaction with an environment depends upon the existence of a set of invariate constraints in the relationships among events in the environment and between human actions and their effects. The implications of the foregoing discussion is that purposive human behav- ior must be based on an internal representation of these constraints. The constraints can be defined and repre- sented in various different ways which in turn can serve to characterize the different categories of human behavior. Skits, RULES, AND KNOWLEDGE When we distinguish categories of human behavior according to basically different ways of representing the constraints in the behavior of a deterministic environment or system, three typical levels of performance emerge: skill-, rule, and knowledge-based performance. These levels, and a simplified illustration of their interrelation are shown, in Fig. 1 The skillbased behavior represents sensory-motor per- formance during acts or activities which, following a state- ‘ment of an intention, take place without conscious control as smooth, automated, and highly integrated patterns of behavior. Only occasionally is performance based on sim- ple feedback control, where motor output is a response to the observation of an error signal representing the dif- ference between the actual state and the intended state in a time-space environment, and where the control signal is derived at a specific point in time, Typical examples ate experimental tracking tasks. In real life this mode is used rarely and only for slow, very accurate movements such as assembly tasks or drawing. In most skilled sensory-motor a | | Te ASMUSSEN: DISTINCTIONS IN HUMAN PERFORMANCE MODELS tasks, the body acts as a multivariable continuous control system synchronizing movements with the behavior of the ESfitamment Peformance i based on cedtorvard contol depends upon a very flexible and efficient dynamic internal world model, Feedforward control is necessary to explain rapid coordinated movements, for instance, in handwriting, sports, ete. The role of feedforward control for industrial control tasks has been demonstrated experi- ‘mentally by Crossman and Cooke [3]. Pew {4} found a shift from error correction mode to pattern generation mode between 0.5 and 1 Hz in sinus tracking. ‘The control of voluntary movements is even more com- plex, Since the success of rapid movements is independent Of the initial positions of limbs, and since the topology of movements can be transferred to other metric proportions and limbs, the function must depend on schemata for generating complex movements with reference to a dy- namic internal map of the environment. Sensory input is probably not used to control movements directly but to update and align this internal map (see Bernstein [5] and the excellent review by Pew [4]. The case in point is that the behavioral complexes necessary to perform an inten- tion to “pick up a glass” or “place finger on nose” (6) are integrated wholes which cannot be decomposed into sep- arate elements (without changing the level of description to neurophysiology). From this discussion the constraints in the behavior of the environment at the skill level appear to be represented only by prototypical temporal-spatial pat- terns Characteristically, skilled performance rolls along with- out conscious attention or control. The total performance is smooth and integrated, and sense input is not selected or observed: the senses are only directed towards the aspects of the environment needed subconsciously to update and crient the internal map. The man looks rather than sees. In some cases, performance is one continuous integrated dynamic whole, such as bicycle riding or musical perfor- mance, In these cases the higher level control may take the form of conscious intentions to “modulate” the skill in general terms, such as “Be careful now, the road is stip- pery,” or “Watch out, now comes a difficult passage.” In other cases, performance is a sequence of rather isolated skilled routines which are sequences of a conscious “execu- tive program.” In general, human activities can be con- sidered as a sequence of such skilled acts or activities composed for the actual occasion. The flexibility of skilled performance is due to the ability to compose, from a large repertoire of automated subroutines, the sets suited for specific purposes. ‘At the next level of rule-based behavior, the composition of such a sequence of subroutines in a familiar work situation is typically controlled by a stored rule or proce- dure which may have been derived empirically during previous occasions, communicated from other persons’ know-how as instruction or a cookbook recipe, or it may be prepared on occasion by conscious problem solving and planning The point here is that performance is goal- oriented but structured by “feedforward control” through oO a9 4 stored rule. Very often, the goal is not even explicitly formulated but is found implicitly in the situation releasing, the stored rules. The control is teleological in the sense that, the rule or control is selected from previous successful experiences, The control evolves by ‘survival of the fittest” rule, In effect, the rule will reflect the fumetional properties, which constrain the behavior of the environment, but usually in properties found empirically in the past. Fur- thermore, in actual life, the goal will only be reached after @ long sequence of acts, and direct feedback cor rection considering the goal may not be possible. Feedback correction during performance will requite functional un- derstanding and analysis of the current response of the environment, which may be considered an independent concurrent activity at the next higher level (knowledge based). ‘The boundary between skill-based and rule-based per- formance is not quite distinct, and much depends on the level of training and on the attention of the person. In general, the skill-based performance rolls along without the person's conscious attention, and he will be unable to describe how he controls and on what information he bases, the performance. The higher level rule-based coordination is generally based on explicit know-how, and the rules used can be reported by the person. ‘During unfamiliar situations, faced with an environment for which no know-how or rules for control are available from previous encounters, the control of performance must move to a higher conceptual level, in which performance is goal-controlled and knowledge-based. In this situation, the goal is explicitly formulated, based on an analysis of the environment and the overall aims of the person, Then a useful plan is developed—by selection—such that different plans are considered, and their effect tested against the ‘goal, physically by trial and error, or conceptually by means of understanding the functional properties of the environment and prediction of the effects of the plan considered. At this level of functional reasoning, the inter- nal structure of the system is explicitly represented by a “mental model” which may take several different forms. We will return to this point in discussion of reasons and. causes Tater. Similar distinctions between different categories of hu- man behavior have been proposed elsewhere, Fitts [7] distinguishes between three phases of learning a skill: the early or cognitive phase, the intermediate or associative phase, and the final or autonomous phase. If we consider that in real life a person will have a varying degree of training when performing his task depending on variations and disturbances, the correspondence with the three levels in the present context is clear, Whitehead (8, pp. 92-98), discussing symbolism, oper- ates with three categories of human performance: instine- tive action, reflex action, and symbolically conditioned action, which are also related to the present discussion; ure instinct i the most primitive response which is yielded by organisms to the stimulus of their environment. Reflex action is a relapse towards a more complex type of <<

You might also like