(return to Bartlett index)

Human control systems

Citation: Bartlett, F.C. (1951). 'Human control systems', Transactions of the Society of Instrument Technology 3: 134-142.

Read on 24th April, 1951 at the Royal Society of Tropical Medicine and Hygiene, Portland Place, London, W.I.

Mr. D. A. Oliver (President) in the Chair

Summary

All human operation of machine tools, scientific instruments, or any other source of varied display which must be dealt with by a succession of more or less adjusted movements, yield illustrations of the working of human control systems. An initial stimulus sets up a movement. The movement itself sets up impulses from muscles, joints, tendons and so on, which tell the operator something about what he has done, and may at the same time alter the external stimulus for the next move in a behaviour sequence terminating only when what is taken to be the purpose, or issue, of the task involved has been as near as may be achieved. Human control systems are (1) essentially cyclic; (2) have a monitoring feed-back which normally has both external (exteroceptive) and internal (proprioceptive) components, and (3) are broadly classifiable into systems whose input is directly sensory and systems requiring mental operations

not capable of analysis into direct sensory process. The information on the basis of which human control systems act is thus given normally partly by external stimuli and partly by internal cues. The external information is usually predominant, especially in relatively simple behaviour control systems, and when the internal proprioceptive signals do act, they seem to do so in a cumulative, intermittent manner. Behaviour control systems in everyday life almost always operate in "continued exercise", and then, at every stage after the first, the input (or feed-back) is apt to possess characters that cannot be expressed in direct sensory terms, and involve, in particular, anticipation, immediate recall, and remote memory. The typical ways in which these work are illustrated by reference to experiments, and some of their implications for instrument design and operation are indicated.

1.   The Control of Behaviour.

It is well known that during the last ten or fifteen years there have been many discussions, of a more or less public character, about the part played by the human operator in various kinds of control systems. At first these discussions were closely tied up with particular practical problems, especially with the tracking of targets. It is clear, however, that whenever a person is trying to operate any instrument or mechanism in accordance with a varying display questions of precisely the same fundamental character arise. Very soon, therefore, the discussions began to range freely in widely different fields; and even, as in some of the current developments of what is called "theory of information," or in parts of that wide area which my friend Professor Norbert Wiener has ingeniously named "cybernetics," to have no specific reference to any particular practical application. In these discussions considerations belonging to electronic theory and practice, to mathematics, to neurophysiology, to experimental and general psychology, and even sometimes to logic and philosophy have been mingled, not infrequently in a somewhat bewildering manner.
All the time nobody has ever doubted that the human operator can act as a link in such systems only because his own behaviour is also, in its own right, a collection—a more or less organised collection—of control systems.

Hick and Bates (1) have defined the term "control system" as "an arrangement of elements inter-connected in such a way that the operation of each depends on the result of the operation of one or more other elements, and the purpose [I should myself prefer to say the issue] of which is to control some condition of a body, process, or machine."   Here, as usually and no doubt con- [135]veniently, the operator's actions are being considered as a partial series only in a chain of events, the rest of which is contributed by the mechanical or electrical instruments which he is operating. Hick and Bates therefore say that they are ready to "adopt the convention that the term 'control system' means the mechanical or electrical part of the total system." In their particular context this usage seems to me to be fully justified, and it is probably the one with which instrument designers and engineers are most familiar and which they find most useful. But the definition proposed is applicable without further modification, directly to the behaviour chain itself whenever an operator is attempting to deal with developing displays and related controls. I am therefore going to ask you to try to think, for the present, of "control systems" strictly as behaviour sequences in which a performer is using streams of information, which come to him from a variety of sources and in different media, as a check to produce balanced response, and to meet, with greater or less success, the current demands of his working environment.

The reason why I am making this demand, which may well turn out to be difficult, and to appear abstract, is that whether for the purpose of engineering design, or for a fundamental understanding of the ways in which machines and machine tools must be used, we need to know all that we can discover by direct experimentation about the behavioural phenomena themselves. Without this knowledge it is fatally easy to present very difficult, and perhaps insoluble problems to the operator, or to build neurophysiological models which have exceedingly little relation to the facts of human performance.

2.   Some Guiding Ideas.

In the present state of knowledge—or lack of knowledge—we must approach our problems with certain guiding ideas in mind. The first of these that I wish to use is that every human control system is fundamentally cyclic with a monitoring feed-back which goes on operating throughout the whole sequence of responses that make up the cycle. Sometimes indeed this can be treated as a formal consideration only. When we are trying to achieve simple measurements of sensory thresholds of the absolute kind, or to measure relatively isolated activities such as simple reaction-times, no great violence is done if we treat each measure as independent and complete in itself. But the control systems with which we are normally concerned always have to do with the continued manipulation of instruments or tools, or with adjustments to the rapidly changing displays of everyday environment. They have to be expressed in that sort of performance which we called "skilled"; and whatever else may be the character of skilled performance nobody ever dreams of doubting that it consists ofa number of interconnected steps, or moves, towards whatever is regarded as the required issue. With a very large number of instances, as, for example, with all repetitive industrial operations, and with connected manoeuvres such as are required for driving a car, piloting an aircraft and tracking a target it is possible and natural to think of a group of steps as completing a stage, after which the system can be treated as making a fresh start. All the instances of the operation of human control systems that I am going to discuss are alike in their possession of this cyclic character.

If we concentrate on the psycho-physical skills, as we must when we are principally concerned with instrument control, the initial input, that which starts off the whole series of bodily movements, is always some form of physical stimulus which is received by one or more of the senses that tell us about the external environment. These are what are called the exteroceptive—the outward-directed—senses, the eye, the ear, touch, less often smell and taste, and perhaps vibration. This exteroceptive initial input produces a first step or move which then, sometimes with and sometimes without a consequent change of external display, acts as a signal for the next step, and so on.

Our second guiding idea is that the signals which are associated with all steps in the performance following the first can operate in some manner as a monitoring feed-back and be used as a check on the efficiency of the moves already made relative to their required issue. We have to try to discover what are the sources, functions and mechanism of operation of this monitoririg feed-back. It is perhaps unnecessary to point out that the monitoring feed-back can be regarded, from another aspect, as continued steps of input. But a good many of these will be initiated within the control system itself, and will not directly tell us anything about the external environment. In . technical language they belong to the internal, proprioceptive sensory systems and not to the exteroceptive mechanism.

The third guiding idea is concerned with the nature of the input elements. There is a broad working distinction between instances in which all the input elements are directly a matter of immediate sensory perception, and others in which some or all of the input elements have a character which, at the moment at which it is effective, cannot be treated as immediately sensory.   The [136]nature and importance of this distinction will, I hope, become clear as we go on to consider the evidence from specific experiments which I shall now attempt to expound.

3. A Simple Human Control System.

Let us begin by considering what can be learned from experiments involving what looks like a very simple human control system. The operator has to press a lever so as to bring about an assigned displacement of a seen signal pointer. Thereafter he must repeat, or hold, his pressure. The experiments were designed and carried out by G. C. Grindley, S. Macpherson and V. Dees (2) working at Cambridge. "The lever was invisible to the operator and was pressed by the first three fingers of the left hand, the wrist being supported on a steady platform." During the learning stage the operator could see the result of his pressure and movement on a milliameter scale. The required pressure for a "bull's eye" was nearly 2 lbs., and the finger movement about 2 inches.

The operation was started by an initial input consisting of a light or a sound. But as soon as the movement began, a complex set of internally initiated sensory impulses must have gone speeding along a feed-back loop from tendons, muscles, perhaps joint-surfaces, from tactile and pressure senses, to whatever neural end-stations were scanning, selecting, or controlling whether the movement was to proceed with acceleration, or with deceleration, or to stop. There was also the visual feed-back, from the perception of the advancing pointer on the milliameter scale. What. if anything, was predominant in all this mass of information which contributed stage after stage of input signals, or, if we like to look at it that way, stage after stage of monitoring material, as the operation proceeded?

The learning phase showed nothing unusual. All operators, some quickly, some more slowly, achieved and held a relatively steady output. Then the operator's milliameter was removed of immobilised, but the experimenter still had his measure of movement and: pressure. The operation was continued for several days. At once and universally there was a great change. Pooled records showed a picture of fairly regular oscillation between long periods of very bad performance (whether by undershooting or by overshooting, but with a so far unexplained predilection for the latter), and longish periods of relatively good performance. These oscillations were much more regular than could be expected by chance. But individual records seemed strongly to indicate that amount of pressure and movement, though not their direction, was almost or quite haphazard within wide limits.

The procedure was modified. An isometric lever control was adopted, the movement elements of the response being reduced well below discrimination threshold. When the required pressure had become stabilised, as before the operator's milliameter was cut out, but the operator was required to hold the pressure for as long as he could. Drift began at once and usually continued, either with diminishing or with increasing (once again more often with increasing) pressure. But the drift appeared to have tolerance limits, differing from person to person though always large. When the limit was neared or reached direction of pressure was reversed and this was continued until the opposite limit was neared or reached. The result as before was a series of long wave oscillations, too regular to be produced by chance, but within these oscillations die moment-to-moment proprioceptive feed-back was apparently having little or no monitoring function. In all these experiments any estimate made by the operator of his own performance was found to give no accurate indication of the nature of that performance within wide limits. In this elementary case it became apparent that much of the information which must have been set up within the human control system itself was not being used.

This is no isolated situation. Wapner and Witkin, for example, working in America (3), seem to have demonstrated that the maintenance of balance on an unstable platform is most successful when a full, stable, and highly structured visual field is also present; less with a limited or weakly structured field; very much less with no visual field, and worst of all with an unstable field full of unpredictable change.

Their results have been several times called in question, but generally only where conflicting visual and proprioceptive cues are present, and this does not really affect the general principle (4).

In all this there is nothing very remarkable, though there is something which is often overlooked by the design engineer and by the people who build theories. The body and the mind, exercising their own control systems, do not and probably cannot use all the information that is available, any more than people do who read newspapers, or children do who are taught lessons. In particular those streams of information which are set up within the control systems themselves seem as if they are normally used intermittently only and in special circumstances. The conditions of their effective operation cannot possibly be established in terms directly of any continuous rate of change of each item of information or of small clusters of items. There is some little under- [137]stood cumulative principle at work, the elucidation of which may well affect radically some of the current notions and criteria of time constants. Moreover other experiments strongly suggest, if they do not yet prove, that the relative priorities of different sources and types of information suffer regular change with increasing age (5).

Direct behavioural facts of this kind have an obvious bearing upon problems of instrument design. They mean that whenever accurate control has to be exercised by a human operator, it  is of the greatest importance to provide unambiguous, graded exteroceptive display. It is unsafe to rely upon a purely proprioceptive, or internal estimation of direction, speed, acceleration and extent of bodily movement, or of pressure. There is indeed some extraordinarily interesting, but little understood, evidence that pressure control, if it is supplementary to exteroceptive display, is superior in most respects to free movement (6). But concerning this much more has yet to be discovered.

Furthermore it seems fair to say that facts of this kind must be taken into account if we are trying to develop a mathematical analysis and formulation of the basic properties of human control operations, or seeking to set up that kind of neurophysiological model which is often of great value in this field.

4.   Control Systems in Continued Exercise.

Except under closely festricted experimental conditions there are few human control systems for which the input, especially in terms of the exteroceptive stimuli, is contemporaneous with its matched output Usually there are a number of related changes of display, presented as discriminable steps in temporal sequence. Each display item, or step, must be appreciated by one or more of the exteroceptive senses, and matched by the appropriate bodily movement, the whole activity constituting a progressive receptor-effector interchange.

All continued exercise of target tracking is fundamentally of this nature, and so also are lots of everyday activities like crossing a crowded street, or driving a car, or piloting an airplane, and so also are an increasing list of industrial skills. In all such cases it is found that if the successive timing of the input exteroceptive elements lies inside the limits of the overall reaction time of the effector elements, the operator very soon, without knowing it, is receiving ahead of his point of action. With a simple type of effector response, like pushing or releasing a lever or switch, or making almost any kind of rapid small adjustment, when the receptor stimuli succeed one another at a rate of about two a second (+ or -within limits) it will be found that a normal and healthy human control system is working, so far as the exteroceptive receptor components go, one, two or three, but practically never more, stimulus steps ahead of the point reached in action.

In this there is no learning process. It is inherent in the timing relationships of human receptor and effector response. It is the simplest and most natural of all illustrations of the operation of anticipation in human performance. It has been shown also that normally the function of this sensorial anticipation is, not to increase the speed of effector response, but to diminish the resting time between one such response and the next: Action becomes smoother.   Effort is reduced (7).

Sensorial anticipation, however, has three serious drawbacks. Its range, or span, is small and sharply cut off. It can deal best with successive discrete stimuli; but if the succession has an inherent structure or order, these properties play no part in the resulting action. Its load capacity is very limited, so that it comes to grief when two or three streams of input signals have to be dealt with simultaneously.

George Eliot once said: "Even our failures are a prophecy," and this is a case in point, for the shortcomings of sensorial anticipation set the stage for a new start.   The new start is achieved not by [138] complicating or reduplicating the existing mechanism, or by a change of speed, or range, or direction in any of its elements, but by something much more radical than these. In sensory anticipation the receptor functions look into the immediate future and use it to prepare for action; now the control system looks back into the immediate past, and uses what it has found there to project action into the immediate future.

E.C. Poulton (8) working at Cambridge, used an experimental tracking technique. The display consisted of simple and complex harmonic courses presented at controlled speeds. The operator must predict particular positions, or follow the target course, when he has access to the whole, or to parts only of the display, which can be blacked out for periods of from 0.5 to 5 seconds.

Under these conditions anticipation came into play so rapidly that very little, if any, learning could have been heeded, even when the intermittent exposures made it impossible for sensory reception to run ahead and pre-date the action of the moment. In the case of simple harmonic displacement "speed" anticipation was predominant. It required response to position, direction, velocity and acceleration, all related together, but, in this order involving a duration of presentation of from a little over 1000th sec. to a little over 1/2sec.

As the display became more complex "courses anticipation" replaced "speed anticipation," and depended upon the use, a little after its actual perception, of a perceived order in the pattern of display. In both "speed" and "course " anticipation the immediate future was being built upon the immediate past, and the working assumption was that, relative to speed and pattern, the objective situation would stay constant.

In theory both speed and rate, or order, are capable of indefinite repetition, and so it might be supposed that anticipation based upon these would operate any number of steps ahead without limit of time. But in practice this does not, indeed cannot, happen. The world we know is not built upon a principle of repetitive pattern, and the moment any discrepant stimulus breaks in, this ' type of control is apt to fall into disorder. Further, immediate recall is severely limited as regards load; in these experiments the combination of three simple harmonic displacements was beyond its range. And finally it appears that decay in immediate memory, when manipulation is also involved, may be unusually rapid.

Thus although the range of this second type of anticipation is greater than that of the first, it is still small. More important, like all attempts to cope with'the future in terms of the past, it is relatively inflexible. Once again the effort to develop a type of control which can match the needs of a swiftly marching and qualitatively varying world has, though for different reasons, met only with partial success.

But another way of dealing with the difficulty, of great interest and importance, proves possible. One of the outstandingly good illustrations of how a human control, system can work may be seen in the direction of traffic, say at a busy airport, or over a large railway region; or again in navigator operations based upon radar displays. This type of situation has been experimentally studied, with synthetic display and control, by Dr. N. H. Mackworth (9).

An operator has control of up to 20 mobile units, each of which must be moved, according to a predetermined random order, so as to reach a particular position at a particular time, and to avoid all interference with, or from, any other unit. A simple unit move sequence contains five steps, and we will take the case in which the steps or moves are made at a rate of one a second. It will be clear that once the experiment has begun very soon there may be a number of units actively engaged and the operator must decide at each step which to move and which to retain where they are. A critical point in the whole situation, therefore, is where and when new units are to become active. How far ahead is it worth while trying to give the operator knowledge of the point of entry of a new unit in the developing situation? Such foreknowledge can be given by underlining, or spotlighting those units which are going to be affected when an appropriate action point has [139] been reached. With decision speeds, or unit moves, of one a second, and a load'of up to 20 units, Mackworth was able to show that underlining, or spotlighting, produced maximum significant favourable effects 2, 3, or 4 units ahead of action point, that is, 10, 15, or 20 seconds, or 10, 15, or 20 decisions or moves beyond the point reached. This is a very marked increase of anticipation span. There are still speed and load limits, but a range of forward facilitation of action has been achieved far beyond anything that either of the other types of anticipation we have considered can possibly compass.

At the same time, since the whole method depends upon keeping in touch with current events, flexibility has been regained.

Now underlining, or spotlighting, tells the operator nothing concrete about just what is going to happen. It tells him only that something important is going to occur at a marked place and time, and he must be on the watch. They are both, in fact, fulfilling a symbolic function. The same spotlighting or underlining can be used whatever may be the particular descriptive character of the unit or the precise direction and timing of the associated move sequence. We have reached a stage which is beyond the range of direct sensory or immediate recall anticipation, both of which are tied up closely with the particular characteristics of the input that must be dealt with.

4.   Anticipation and Instrument Technology.

I have deliberately chosen to talk at some length about facts which may at first appear to have little to do with the more familiar problems of instrument technology, for two main reasons. First, because the experimental facts about the functions and mechanism of anticipation in the development of balanced and economic human performance are now known, but not widely known. Secondly, and particularly, because it appears to me that these facts have a bearing upon problems of instrument design and working conditions which are still given less consideration than their importance warrants. Wherever machine-controlled performance consists of a sequence of related display signals and associated movements, the timing relations of the two are a matter of very great importance. There is a rate of successive display which lies within the optimal limits for smooth, accurate and economic performance. If the display has an inherent structure or pattern, as it frequently has, the instrument designer needs to know within what limits of complexity or of load, this character can be used by the operator who has to depend upon a type of immediate memory which can be made effective in short term exercise. When, in a developing situation, information about the later stages has to be accumulated, while action is being taken about the earlier stages, the designer must consider at which points and in what form he can provide symbolic indicators which will keep the operator alert just when alertness is most needed.

5.   Basic Theory.

It seems to me also that this whole story may have extremely important implications for basic theory, but these are at present far more disputable. Take the question of the neurophysiological model. It may be that the search for a single all-inclusive model is at present premature. Not only in regard to anticipation but in regard to the whole matter of input-output relationship in human control systems it may be that there are three key problems. First there are the questions of the parts played by current and basically continuous sensory perception of different types and sources. Second there are those of short-term storage and immediate recall having a fundamentally intermittent operation. Third there are those (always and inevitably associated with the development of symbols) of long term storage, with its possibilities of relatively distant forecasting and of memory proper. Perhaps the human way of dealing with them is from time to time to make a real new start, and to use principles of check [140] and feed-back which are different from any that have preceded them.

All the time the whole development of human control systems can be seen as a prolonged struggle between the claims of accuracy on the one hand and flexibility on the other. Everybody looking at how the mind and body do their work must at some time have been impressed by the fact that for very nearly everything that we have to do we seem to have a lot more ways of doing it than any regard for economy would demand. At any moment there is a vast lot of "noise" in all the information upon which we must base action. If we get rid of it we gain definiteness but lose range; if we try to make it all completely unambiguous we are caught up in a state of unsatisfactory indecision. Perhaps, therefore, the watchword for the instrumental control system must be accuracy, for the human control system flexibility, and we must pursue and understand both as well as we can, until it may turn out that they are less incompatible than they appear to be. With that perhaps rather lame conclusion, I must stop.

REFERENCES

(1)  '. W. E. Hick, and J. A. V. Bates, " The Human Operator of Control Mechanisms." Ministry of Supply Monograph No. 17204, 1950.
(2)    G. C. Grindley and others, " The Effect of Knowledge of Results on Learning and Performance. Some Characteristics of Very Simple Skills." Quarterly J. of Exp. Psych., 1948, 68-78.
(3)    S. Wapner and H. A. WrnciN, "The Role of Visual Factors in the Maintenance of Body-Balance." Amer. J. of Psych., 1950, 385-408.
(4)    e.g., George E. Passey, "The Perception of the Vertical: Adjustment to the Vertical with Normal and Tilted Visual Frames of Reference." /. of Exp. Psych., 1950, 738-745.
(5)    See e.g., A. T. Welford and others, " Skill and Age: An Experimental Approach." Oxford, for the Nuffield Foundation, 1950.
(6)    C. B. Gibbs and J. C. Baker, " Free Moving versus Fixed Control Levers in a Manual Tracking Task." Proceedings of the Conference on Automatic Control, 1951, to be published by D.S. & l.R.
(7)    M. A. Vince, " Rapid Response Sequences and the Psychological Refractory Period." Brit. J. Psych., 1949, 23. J. A.,Leonard, "Experimental Studies in the Temporal Relation between Information and Action." {Unpublished Thesis, Cambridge), 1951.
(8)    E. C. Poulton, "Speed Anticipation and Course Anticipation in Tracking." M.R.C. Unit for Research in Applied Psychology, Cambridge. A.P.U. Report 123.
(9)     N. H. Maocworth. "Researches on the Measurement of Human Performance." (Unpublished Thesis), 1948.

DISCUSSION

The PRESIDENT introducing the lecturer said: Sir Frederic Bartlett is the Professor of Experimental Psychology at the University of Cambridge. If our audience is really up-to-date most of you will be engrossed in Sir Frederic's fascinating book which has just appeared entitled "The Mind at Work and at Play." I have enjoyed dipping into it and am now reading it seriously; as a cure for any form of self-adulation or egotism it is a wonderful tonic. Sir Frederic Bartlett is probably the most eminent worker in the field of experimental psychology and was created a C.B.E. for outstanding services during the war in the selection of personnel. Judging from his recent book as distinct from his more serious works, I am sure he is a very charming human personality as well.

Opening the discussion the PRESIDENT said: After listening to Sir Frederic and following the eloquence with which he generalised and drew differences between so many everyday occurrences with which we are partially familiar, it is difficult, first of all, not to say with the Psalmist that we are "fearfully and wonderfully made." It is fearful the mistakes we can make and it is marvellous that any of us deserve our salary cheques at the end of the month!

I was interested in Sir Frederic Bartlett's remarks concerning the anticipatory mechanism. .About twenty-five years ago the Institution of Industrial Psychology set up a unit in one of the offices in Bushey House, and called for volunteers from the National Physical Laboratory to lend themselves for an impartial test. The usual thing that happened was, you went into a room where you were kindly received, and then you were asked your name, those of your parents, whether you were frightened of the dark and so on. Then they took you to what was then quite new, a machine with a rotating spiral and with little circles on the spiral. Starting in the middle you had to put a dot in each of the circles. You could do this fairly easily for about a third of the record, but after that it became more difficult, and reached the stage where one found it impossible.

As a slight aside, a young man was shown into the room and went through this process, and after a little while a secretary came in and said that the Director was ready to interview him! He had been dotting furiously to try and get the job! Whether he got the job or not, he was certainly impressed by the way in which such organisations selected their personnel.

On the other hand, it was my first introduction to experimental psychological tests of this kind. I naturally felt humiliated that I could not finish this spiral, but I made enquiries as to what could be deduced from the fact that people failed near the end. I should like Sir Frederic to confirm that they were really looking for someone with the wit to dot every second, third or fifth one rather than stab madly at them all?

Sir FREDERIC BARTLETT: I cannot confirm that, because that particular instrument was used for all sorts of different purposes. They may have been trying to find someone who would leave a lot out; I do not know.

PROFESSOR A. TUSTIN: I should like to offer a few comments on the bearing on the problem of manual tracking, as it arises in industry and in the services, of some of the points of Sir Frederic Bartlett's lecture. [140]

It is important to keep in mind that the "characteristics of the human operator" that are referred to in discussions of such questions are "learned" or acquired characteristics arising from practice in the skill in question. The process of learning consists in the reinforcement by success of the likelihood of certain kinds or patterns of response and the inhibition by failure of other kinds. This process is largely unconscious. The learner cannot say in what his growing skill consists. Sir Frederic placed great stress on the contribution from failure in the .learning process. I would be interested to know whether this was intended or whether he thinks that the stimulus or reinforcement of successful types of response by the experience of success can be equally important.

The lecturer tended to question the usefulness or value of attempting to represent the complexities of the behaviour of a human operator by such grossly simpli­fied "mathematical models" as are now in common use in the discussion of problems of manual control. Is it sufficient, or at least useful, to say for example that a skilled operator in a tracking task gives a control handle speed that is proportional to some combination of error and rate of change of error, with a fixed time delay, possibly also with some random disturbance? My own opinion is that such models are of the greatest value in such problems as, for example, the devising of the best type of response characteristic for the mechanical part of the control. I would even go further, and say that the implications of the approximate representation by a simple model of this kind have never yet been seriously followed up, and that probably improvements in accuracy of control and ease of learning would result very rapidly from a programme of experimental development based on this conception. It is surprising that so little has been done in this direction. The fact that the response includes a certain element of randomness need not discourage the use of a model An identical problem occurs in automatic control in the form of the presence of "noise," and this has been handled successfully.

A further question that I would like to put to Sir Frederic is whether he can tell us anything about the nature of the mechanism by which the human operator perceives speeds (e.g., is able to respond to the "rate of change of error"). It has been argued that to perceive speed one must snap-shot, as it were, the position at two separated instants and take the difference. This argument is used by those who wish to support the view that the behaviour of a human operator is essentially of an intermittent nature.

I feel, from subjective impressions, that this view is wrong, and that we have some faculty of becoming aware of movement quite directly. Could it be that the organ for such awareness is in the "servo-control" that enables the eye to follow moving objects? Such control certainly exists and seems to have great speed of action and precision. If not, what explanation of this remarkable faculty is at present the accepted one?

Sir FREDERIC BARTLETT: I do not think that I can fully answer the first point raised as to whether failure or success is the greater incentive to further learning. In terms of general development of the human mechanism it seems certain to me that what has happened over and over again is that a particular method of dealing with information has been tried and has been found wanting, and it is that which has proved to be the principal incentive for trying something else. In that broad sense, failure seems to me to  be the principal spur to further effort

If we are considering the operations of an individual who is learning a particular task, then it seems as if the whole process is different, and that there is no learning to speak of unless there is the knowledge, following every effort, of the degree to which the performance has approximated to the desired issue. Then there is no doubt that, as far as the individual is concerned, the principal spur to increased effort is success. I should have thought that the answer is that in the very broad sense, in considering the development over a long period of years of different methods of attack, failure is the stimulus, but that so far as the individual process of learning is concerned, success is the principal stimulus.

The second question concerned the use of neuro-physiological models. I believe it is highly desirable that all those people who are interested, whether they are engineers or physiologists, who can find advantage from constructing neuro-physiological models, should do so. In fact, I do not think that it is possible to make a great advance in this field unless such models are constructed, although most of them will fall a great deal short of the complexity which we know to be involved in the case of human responding systems. What I have been trying to point out is that if we take any one field of development, such as, for example, the question of anticipation which I discussed in detail, we should need not one model but at least three, operating rather different principles with rather different time constants and other things coming in. It seems likely that we cannot have a single model which is capable in any one field of human endeavour of dealing with all the kinds of things which happen.
With regard to the third question of how we can deal with speed, I can attempt, briefly, only a very speculative answer. There are plenty of psychologists who would say that the estimation of movement, of speed of movement, and of direction and course of movement are all immediate; that we have special mechanisms for all of these; and that that is all we can. say about it. I do not really believe that. I think that evidence suggests that in the case of speed, which is not really very accurately estimated by direct observation, it does mean that there will have to be related response to different positions, to something which has to be called 44 direction," to velocity and to acceleration, and they are all different. These are then somehow combined together to produce the complex result which we call estimated speed. I do not know what are the actual mechanisms by which all this is done; but perhaps Dr. Hick can answer that question.

Dr. W. E. HICK: I agree with Professor Tustin that the direct perception of speed need not be, and commonly is not, simply and literally a matter of identifying two positions and dividing the distance by the time. Speed, in the general sense of a rate of change, can undoubtedly be registered in living systems by approximate differentiating devices of the "slow leak" variety, roughly analogous to those we are all familiar with in the mechanical and electrical fields. Discontinuous devices which measure a particular displacement and the corresponding time and indicate the quotient may be placed at one end of the scale; at the other end we have the completely continuous "slow leak" mechanisms. Biological mechanisms tend to be of a mixed type. There is evidence that the nerve network in the retina may give some, indication of the speed with which an image passes over it; in the frog, for example, it has been shown that the retinal network acts as a predictor of [142] the future position of a fly or other insect prey, which otherwise the frog would be unable to catch. In man, the visual judgment of speed is more complicated and comparatively little is known about the physiological mechanism of it. However, all speed indicators require to collect information over a finite time, and therefore must commit errors depending on how the instantaneous speed varies in that time. With regard to the speed judgments involved in operating machines and instruments, such factors would have to be taken into account in attempting anything more than the crudest deductions. A beginning has been made—for example, the least perceptible sudden change in speed has been measured— but there are many more quite simple experiments waiting to be done.

Dr. A. M. UTTLEY: On the question of perception of movement, I think there is some doubt as to how far it is due to perception of the movements of one's own eye. If a swinging pendulum is set up in a dark room with a small light at the fulcrum and another on the bob, then with no other visual reference it is not possible to tell which light is fixed and which is oscillating. The subjective experience alternates from one to the other.

We perceive movement in a number of ways other than through seeing that present position differs from remembered past position. If one point is fixated then objects moving relative to it produce trailing after images. If one fixates a road then a car moving at 30 m.p.h. will be associated with trailing after images one or two feet long. Recent work on the frog's eye suggests that a mechanism may exist for immediate detection of movement through the agency of special cells which respond only to changes in light intensity. A similar idea is suggested by the "streaming phenomenon" in which we clearly perceive movement without change of position.

On the question of success and failure one may refer to the theory of homeostasis, in which it is suggested that living things do not react to a normal acceptable situation but only to deviations from it, when they react so as to reduce this deviation. Deviation is failure, either in servo-mechanisms or in living things; and the natural reaction to success, i.e., zero deviation, is for the organism to become quiescent.

Sir FREDERIC BARTLETT: I think that this question of speed is a very difficult one. It seems to me clear that if we identify an object in one position with an object which is subsequently in another position, this may, within certain limits of interval, give us a perception which is interpreted to mean movement from one position to another; but it will not by itself give us anything at all about speed. It seems to me that there are two possibilities. It might be that speed is an experience which can occur only if we are able to compare two objects, one of which moves from one position to another position, while the other is moving from one position to a different position. Then, the difference being greater in the one case than in the other, the inference is that the one has moved quicker than the other. It is possible that all our experience of speed is based on something which is of the nature of comparison between objects which are treated as identical, all of which arrive at different positions at the same time, the time and the positions being the things which are directly perceived and the relative speed being something which is inferred. On the other hand we may have learned to attribute speed to visual events in a somewhat different manner. As before an object which is seen in a given position is identified with an object seen subsequently in another position and, given an optimum objective interval, the perception is that of a single object moving from one place to another. But the whole of human experience, and not visual events only, falls into a pattern of succession, and so perhaps we set up a kind of model, or scheme, of before-after standards which can be used in a comparative manner in this special case. The primary perceptions would be those of identity, position, and time, and these, all applied together in our special case, appear as "speed of movement."

There is plenty of experimental evidence to show that what is called "perception" often contains inferential elements. If the first kind of explanation is adopted speed perception would be an inference from different positions which are reached in identical time. If the second kind of explanation is preferred speed percep­tion becomes the identification of a movement as com­pleted "before" or "after" something else—anything else, visual or not, which can have its place in a general scheme or model of time standards.

In both kinds of explanation, speed becomes a quality or attribute of events the appreciation of which involves always some psychological activity of comparison either between two, or more, closely concurrent concrete directional changes, or between one such change and an already established scheme of time standards. Fundamentally, in either way, the same speed may be treated as "quicker" or "slower," but eventually, by the normal process of psychological abstraction, we do arrive at a notion of " absolute speeds," and then speed may appear to have all the character of an immedi­ate perception. In fact there is hardly anything in the way of the human interpretation of evidence which has not, at some time or another, been put down to immedi­ate perception or immediate experience.

All the same I am bound to agree that there; may perhaps be some special human mechanism which is directly concerned with the response to and the esti­mation of speed. If there is, nobody can yet say what it is. I agree with Dr. Uttley. that the evidence is against its being based directly upon eye movement or the control of eye movement. But I think also that it is very, very unlikely indeed to be associated with trailing images. Even if there should be a human mechanism,, visual or other, that can act as a direct speed indicator, I cannot myself see how it alone could give us anything more than just " speed," or how, unless other conditions and mechanisms were also coming into play, it could enable an operator to deal with speeds in and of machines anything like as well as we know he is able to do.

A hearty vote of thanks to Sir Frederic Bartlett was moved in appropriate terms by Dr. J. C. Evans, Chairman of the Control Section, which had undertaken the responsibility of arranging the lecture for that meeting. The Section had been fortunate in persuading Sir Frederic Bartlett to give this most interesting, stimulating and authoritative lecture on a subject of great importance to this country, indeed, to the whole of civilisation. The vote of thanks was carried with acclamation, and the meeting terminated.

FIGURE DESCRIPTIONS (the images were not clear enough to display here)

Fig.   1.   AN EXPERIMENT ON SENSORY ANTICIPATION

In this display five neon lights can be illuminated in random sequence at any desired speed. The operator must move his stylus from the middle circle on the "control" board to whatever outer circle is indicated by the stimulus light, and back through the middle to the next indicated circle. In one setting, the next successive stimulus light appears only on return to the centre position after a given control movement is complete; in another - the anticipation setting - the next signal light in succession appears as soon as the appropriate outer circle for the preceding stimulus has been reached.

Fig. 2. ANTICIPATION AND IMMEDIATE MEMORY

In this form of the experiment the operator is attempting a continuous tracking task. The extent of the course displayed, if any, ahead of the point of action reached can be con ­trolled. There are many other forms of the same basic. experiment.

Fig.  3. A "FORESIGHT" EXPERIMENT

The experimenter - left - is following the instructions of the operator-right-in order to bring about a timed series of moves of a number of units which can represent the mobile items in a traffic control situation. All moves are recorded and scored in appropriate ways. The black circle indicates to the operator that at a certain point, a known number of moves ahead, an important new mobile unit will have to become active.