The Act of Creation
Page 68
While behaviourists denied the rat the capability to acquire a mental map of a maze, the ethologists have shown that this is precisely what insects do. Their work merits consideration in some detail -- which is done best by textual quotation. Tinbergen and de Kruyt have trained wasps to find their way to the nest by a configuration of certain landmarks (such as fir-cones and twigs). When these training marks were moved, and the wasp might have been expected to show complete disorientation, it 'will suddenly utilize new landmarks completely unrelated to the previous orientation marks on which it had apparently been trained -- a result which has interesting and suggestive similarity to Krechevsky's work on hypotheses in maze-learning. Many such examples lead us imperceptibly to what we may consider as insight-learning. Ammophila hunts caterpillars which are too heavy to be brought back on the wing, and which may thus have to be dragged for a hundred yards or more across and through every imaginable natural obstacle. Here the original learning of the territory has probably been in the main affected by observation from the air, and yet the return has to be made on foot. Although the insect may from time to time leave her prey and take short survey flights, this is by no means invariable . . . and quite often the wasp seems able to maintain orientation while on the ground as a result of earlier aerial reconnaissance.' [23] When large obstacles (metal screens 50 by 120 centimetres) were placed in its path, 'the insect diverged just enough to carry it round the obstacle on a perfectly smooth, even course of maximum economy of effort. On one particular occasion the experiment was immediately repeated twice with the same results, but the third time the insect walked straight at the screen, climbed with perfect ease to the top and, without ever letting go of its caterpillar, flew down to the ground on the other side and continued its journey. On subsequent occasions this insect would adopt now one type of behaviour, now another, but in no instance did she ever show trial and error. The solution to the problem was always smooth, unhesitating, and economical.
This and other individuals which behaved in the same way were then caught and transported in a dark box to a new site, a process taking less than a minute. On release the insect might have been expected to be at least momentarily disorientated. On the contrary, it appeared quite unperturbed and without any orientation flight set out at once on its new course. It was again given the detour test three times on its new course, but it reacted as efficiently as before and within a few minutes had arrived exactly on its nest. Although the insect too, on occasion, shows some evidence of disorientation, nevertheless the overwhelming impression given (as also recorded by Baerends (1941) and other workers) was one of almost uncanny knowledge of the details of the terrain.' [24] Thorpe concludes: 'In some instances it is possible that the homing faculty depends on no more elaborate sensory mechanism than that involved in simple taxes or light-compass reactions. Nevertheless, it is certain that in a large number of examples this is inadequate and that we have a true place memory. [25]
In the case of at least the honey-bee, the memory is communicable. The orientation dance of the bee is certainly as striking an example as one could wish for, of place-learning encoded and re-coded into a symbolic motor-pattern. Yet again the question arises: if, following Thorpe, we call the behaviour of these insects 'insightful', have we not stretched the word into a kind of rubber concept? And once more the only description which implies neither too much nor too little seems to be that the organism has contrived to build a coded model of the invariant and significant aspects of the territorial environment into its nervous system.
The Controversial Rat
Munn's Handbook of Psychological Research on the Rat, published in 1950, contains a bibliographical list of over two thousand five hundred titles; its rate of growth since then is anybody's guess. A considerable portion of this research was devoted to maze studies. Not even Newton, as Bertrand Russell remarked [26], could learn a maze by any method other than trial and error; yet what the rat learns is not a chain of responses, but the pattern of the maze as a whole -- as shown by the experiments with mutilated rats, and in others where the rat takes prompt advantage of short-cuts when a wall is removed, and avoids newly created cul-de-sacs. The evidence is equally conclusive that the rat is capable of latent learning and of forming 'hypotheses' by 'provisional tries'. Yet the Great Rat Controversy was kept going -- partly because S.-R. theorists kept coming up with ingenious alternative interpretations of the evidence, and partly because of the weighted character of much of the experimental procedure itself. A good illustration for this was the 'continuity versus non-continuity' dispute, where Spence and others represented the behaviourist view against Krech, Lashley, etc. The results could be interpreted either or neither way; but -- as Osgood wrote, in summing up the controversy: 'It is significant with respect to methodology in psychological experimentation that, almost without exception, the studies supporting the Lashley view have used the jumping stand, while those supporting the continuity view have used a Yerkes-type discrimination box.' [27]
Not only the experimental conditions, but the experimenters' subjective attitudes seemed to exert their influence on the data obtained. In this respect Rosenthal's 'experiment on experimenters' must have come as a shock to students who had taken at least the 'hard and fast' quantitative data (if not the interpretations) of nearly half a century of rat experiments for granted. Rosenthal gave one group of his research workers rats which, he explained, were 'geniuses' specially bred from a stock with exceptionally good maze-learning records. To a second group of researchers he gave what he explained were 'stupid rats'. In fact, all rats were of the same common-or-garden breed; yet the score-sheets of the 'genius rats' showed unmistakably that they learned to run the maze much faster than the 'stupid rats'. [28] The only explanation Rosenthal could offer was that the bias in the research-workers' minds had somehow been transmitted to the rats -- just how this was done he confessed not to know. These and other experiments by Rosenthal caused one science editor to comment: 'The results throw a pall over the entire range of psychological tests as reported by the psychologists over the last fifty years.' [29]
Thus I shall have little more to say about the bar-pressing and maze-running experiments with rats. In spite of the impressive mathematical apparatus, and the painstaking measurements of 'rates of response', 'habit-strength', 'fractional anticipatory goal-responses', and the rest, rarely in the history of science has a more ambitious theory been built on shakier foundations.
The Cat in the Box
The cat in the puzzle box, in Thorndike's classic experiments, is also put into a situation so designed that it can be solved only by trial and error. The box is equipped with contrivances such as rings, loops, turning bolts, pedal boards, etc., and with a door which opens, according to the experiment, when the animal operates one of these contrivances, or several of them in a given order. Thus the cat may have to turn bolt B which, however, becomes loose only after bolt A has already been turned; or it may have to pull a string, or a loop, in order to free itself. When the cat is put into the box, it 'tries to squeeze through any opening; it claws or bites at the bars or wire; it thrusts its paws out through any opening, and claws at everything it reaches; it continues its efforts when it strikes anything loose and shaky; it may claw at things within the box. It does not pay very much attention to the food outside, but seems simply to strive instinctively to escape from confinement. The vigour with which it struggles is extraordinary. For eight or ten minutes it will claw and bite and squeeze incessantly.' [30]
The cat's behaviour is typical of that phase in a 'blocked' situation, where organized behaviour disintegrates and yields to more or less random trials. The solution -- pushing a bolt or pulling a loop or even licking itself -- will be first hit upon by chance, and after a number of repetitions, it will be retained; the learning curve will be more or less continuous or it may show a sudden, sharp drop. The objections against this kind of experimental design are essentially the same as against classical conditioning: they create an artifi
cial universe. 'The solution of Thorndike's problems demanded behaviours that were quite beyond the animal's normal repertoire. Cats do not get out of boxes by pressing buttons or by washing themselves; rather they try to squeeze through narrow openings or scratch at the barriers, and Thorndike's animals were observed attempting just such solutions as part of their early trial and error. In other words, the correct response in a situation like Thorndike's could only be hit upon by sheer, blind chance.' [31]
The conclusion which Thorndike and his followers drew from these experiments -- 'that animals are incapable of higher mental processes such as reasoning and insight -- that they are limited to the stamping-in and stamping-out mode of trial and error' -- is one of the most astonishing examples of question-begging in the history of modern science. One might just as well adapt Thorndike's method of reasoning to human education, teach children nonsense syllables which can only be learned by rote, and then conclude that children are only capable of learning by rote. And yet, to quote Hilgard: 'for nearly half a century one learning theory dominated all others in America, despite numerous attacks upon it and the rise of its many rivals. It is the theory of Edward L. Thorndike.' [32] Tolman, writing forty years after the publication of Thorndike's Animal Intelligence, went even further:
The psychology of animal learning -- not to mention that of child-learning -- has been and still is primarily a matter of agreeing or disagreeing with Thorndike, or trying in minor ways to improve upon him. Gestalt psychologists, conditioned-reflex psychologists, sign-Gestalt psychologists -- all of us here in America seem to have taken Thorndike, overtly or covertly, as our starting point. [32a]
Cat experiments of a type diametrically opposed to Thorndike's were carried out by Adams. It is amusing to compare the description of the behaviour of Thorndike's cats which I have just quoted, with the behaviour of Adams' cats:
A piece of liver is suspended from the top of a wire-cage, so that the liver rests on the floor inside the cage, loosely held by the thread. A hungry cat in the room with the cage, but outside it, sees the liver and walks over to the cage. It hesitates for a time and its head moves up and down as though it is studying the string. Then it jumps on top of the cage, catches the string in its mouth, raises the liver by joint use of mouth and paw, and leaps down with the stick at the end of the string in its mouth. [33]
The behaviour of Adams' cat, first 'thinking out' the solution of the problem, then acting it out in an unhesitating, smooth, purposeful manner, is of the same kind as Köhler's chimpanzees'. The contrast between this type of intelligent problem-solving, and Thorndike's stamping-in process seems to be complete. Yet the cat's learning in the box is by no means the blind, random process which Thorndike and his followers read into it. In the first place we must realize that the cat's handicap lies not only, as in classical conditioning, in the irrelevance of the clues which in its natural environment the cat would ignore; but also in the fact that the clue -- say, a loop hanging from the ceiling -- is hidden or 'drowned' among other equally irrelevant clues -- latches, bolts, etc. The unnaturalness of the tasks set in these experiments is illustrated by the 'lick-yourself-to-escape' type of rule. Yet even in this surrealistic universe, the cat's behaviour testifies to a remarkably high I.Q. After the initial bewilderment, as adjustment to the laboratory situation progresses, the range of the cat's tries will be narrowed down, and loops, bolts, etc., will be paid an increasing mount of attention, as members of the nascent matrix. The cat develops, like Pavlov's dogs, an attitude of expectancy, of 'Means-End Readiness' (Tolman); it begins to form 'hypotheses'. [34] Thus, when the cat has learned to get out of the box by clawing at a loop, and the loop is then displaced from the front to the rear wall of the box, it will learn to free itself much quicker than before it had abstracted the loop-Gestalt from other clues. If, however, the loop is replaced by a small wooden platform hung in the same place which the loop occupied before, the animal will free itself after a short while by striking at the platform; in this case, the location is the clue. Thus the 'loop hypothesis' can exist side by side with the 'place hypothesis' -- just as Krechevsky's rats, who had to guess whether the food was hidden by a door of a given colour, or of a given location, formed first a colour hypothesis, then a place hypothesis. [35]
At this stage, the cat's behaviour can be described as a series of 'provisional tries'. [36] These tries, far from being governed by chance, show great plasticity: if the cat has learned to escape by pu||ing a string with its foot, it may on the next occasion free itself by pulling the string with its teeth, which requires an entirely different sequence of motions. Even where the 'correct' response was the perverse action of licking its own fur, the act is reduced in the final trials 'to a mere symbolic vestige.' [37]
Whichever way we look at it, the cat's behaviour is most fittingly -- if somewhat metaphorically -- described as learning the rules of Thorndike's game by a process of elimination and empirical induction. The learning curve is a function of several variables: it will show gradual or sudden progress, continuity and discontinuity, according to the experimental conditions, individual learning capacity, fatigue, and chance.
NOTES
To p. 559. Hilgard calls Hull's system 'the most influential of the theories between 1930 and 1950, judging from the experimental and theoretical studies engendered by it, whether in its defence, its amendment or its refutation' (Hilgard, 1958, p. 192).
To p. 559. The objection to this is not that Hull postulated a continuous series linking rat to man, but that his 'primary laws' are epitomized by the bar-pressing act of the rat, which he regarded as the atomic unit of behaviour. The fallacy of this reasoning seems to be derived from Hull's implied notion of mental progress from rat to man as a linear gradient. Theories of this kind fail to take into account the hierarchic principle in mental evolution -- reflected in the hierarchy of levels in the nervous system. If instead of linear gradients, we think in terms of levels of increasing complexity, then a difference in degrees does become a difference in kind. Since the basic mechanisms of sexual reproduction are common to all mammalian species, Hull's postulate seems to imply that detailed study of sexual behaviour in the rat would eventually yield the 'primary laws' underlying the Kinsey reports on the sexual behaviour of the American male and female. Homologue principles (such as the part-whole relation, or control by feedback) do operate on all levels, but they are general principles, not specific 'units' or 'atoms' of behaviour.
To p. 565. Coding is an irreversible act, and once the code is established, it will be relatively permanent -- until it decays. If, however, the dog is fooled repeatedly and in quick succession ('massed practice'), i.e. food is withheld after the buzzer has sounded, a negative code will superimpose itself on the previous one. The first few times the response will stop at salivation short of chewing; but soon it will stop short of salivation. After a few hours' rest, however, salivation is restored by 'spontaneous recovery' -- a paradox which has bedevilled learning theory for a long time. Perhaps the explanation may be sought on the following lines. The whole attitude of the dog, as it has become adapted to the laboratory situation, is based on the expectation that all stimuli are events relevant to food; and that the negative code (buzzer -- > no food), if it has been quickly superimposed on the positive one (buzzer -- > food), is of a more temporary and brittle nature than its opponent. If, however, the unrewarded signals are spaced out over a greater length of time, i.e. if the extinction-drill approximates in thoroughness the original drill, extinction becomes final (Cf. e.g. Hebb, 1958, pp. 134-5, 147).
XIII
THE PITFALLS OF GESTALT
More about Chimpanzees
If the S.-R. theorist's method of designing experiments seems to be aimed at printing a wiring circuit into the animal's nervous system, Köhler's method was to provide it with a do-it-yourself kit. The main task of the experimenter, as Köher saw it, was to arrange for his chimpanzees conditions which favour original discovery by placing the necessary paraphernalia in the
ir cage -- solid and hollow sticks, crates, etc.; and to make the task just difficult enough to exceed, by a fraction as it were, the limit of the animal's repertory of skills.
I have mentioned a few typical examples of the chimpanzees' achievements in Book One ( Chapter V). In the use of tools the decisive factor was the discovery that a previously acquired playful technique M1 could be applied as a mediating performance to solve a problem in the blocked matrix M2. Nueva applied her stick -- which previously she had used only for pushing things about in play -- to rake in a banana placed beyond her reach outside the cage. In similar ways, the chimpanzees used sticks as jumping poles to get at fruit hung high from the ceiling; as implements to make holes in the wire-netting of the cage, to dig up roots in the earth or to prise open the lid of a water-tank; they used sticks as traps to capture crowds of succulent ants, and as weapons for stabbing at fowls and killing lizards. Each of these new achievements was based on the combination of two or more already existing skills, and some of them on serial sequences of 'Eureka processes': when a chimp had discovered the use of a stick as a rake, a short stick was then used to rake in a longer stick to rake in the bananas.