The Act of Creation

by Arthur Koestler

  He told it all, and then became quiet. When the sun rose he fell down. Something black came out of his mouth. His face became contorted. The people jumped up and cried. He was dead. [10a]

  Twenty subjects were tested. Their written recollection of the story shrank with the passage of time to a few lines which, in most cases, distorted its content almost 'beyond recognition'. But, with one exception, all the spaced-out versions of all subjects, even after several years, contained the lines: 'Something black came out (or "issued" or "rushed" or "jumped out") of his mouth.' The one exception wrote (after four months): 'My remembrance was in visual terms . . . of breath somehow materializing into a ghost.'

  Fiction and autobiography abound with examples of such 'vivid-fragment-memories': the mole on Granny's chin, the fly crawling over the lump of sugar at the moment of the dramatic climax. Let us call this the 'picture strip' type of memory -- although, of course, the 'vivid fragment' may be auditory or olfactory, or even a whole 'cinematographic sequence' -- like the detailed, auditory-visual sequences which Penfield evoked in his patients through electric stimulation of their exposed temporal lobes (see below).

  Obviously the formation of such 'picture-strip traces' must also be preceded by some filtering process; but in this case the criteria of relevance and parsimony are different, and often directly opposed to those of the abstractive hierarchies. The vivid detail is usually described as 'striking', 'attractive', 'evocative', 'nostalgic', 'frightening'; it always has some emotional significance. It is mostly on a pre-verbal level; but even verbal fragments -- of a poem, of a chill warning or whispered endearment -- are retained because of their affective quality. The filter-matrices that operate in these hierarchies must be emotionally 'attuned' -- let us say to some hypothalamic controls. We may further assume that such picture-strip memories are formed on lower (pre-conceptual) levels than memories of the abstractive kind; they may be symbolic, but their symbolism too is often pre-verbal, and perhaps related to the symbolism of the dream; they may even obey their own special brand of parsimony. We have seen that occasional regression to lower levels of the hierarchy is a conditio sine qua non of creativeness; the combination of the abstractive and picture-strip type of memory may serve the same purpose -- vide Coleridge and Kekulé.

  We are thus led to assume the existence of various hierarchies in perception and cognition, whose criteria of relevance are determined by the attitudes, drives, emotions, which they serve; they interlace with each other on every level, and thereby provide the multi-dimensionality -- or multi-colouration -- of experience; at the same time they also compensate for the impoverishment of experience in the process of memory formation. The outstanding memory which some geniuses are said to have possessed may possibly be due to their many-dimensional ways of analysing and storing experiences.

  Learning to See

  Let us turn to vision.

  Innate perceptual organization provides no more than the primitive foundation on which learning can build. The long controversy between Behaviourists and the nativistically inclined Gestalt school whether perceptual organization is innate or learned has finally been superseded by the more realistic question how much is innate and how much acquired by early learning. The consensus seems to be that colour and brightness constancy, and the recognition of line, angle, and texture are innate in rats as well as men. [11] So is 'primitive unity' (Hebb) -- the segregation of simple figures as coherent entities from the background (for instance, a black splash on a white card). But when it comes to more complex figures where the contrast is less marked, even the figure-ground relation is strongly influenced by learning and expectancy. I have quoted examples from the neglected field of auditory perception where the relativity of figural unity is obvious. The radiologist whom experience has taught to see a peptic ulcer or a lesion of the lung, treats the much sharper contours of the ribs as 'background'. When you hunt for a collar-stud in a drawerful of miscellaneous objects, that small, insipid form, poor in Gestalt Prägnanz, will 'stick out a mile'; the remaining contents of the drawer are 'background'.

  Thus even figural coherence is influenced by past experience and present attitude. When it comes to the identification and recognition of visual Gestalten, it has been shown that even the simplest of them -- triangles, squares -- require an element of learning. The behaviour of chimpanzees reared without pattern vision, of human beings with congenital cataract who had to 'learn to see' after they were operated on, and the cumulative evidence from other experiments indicate, in Hebb's words, 'that the normal human infant goes through the same process, and that we are able to see a square as such in a single glance only as the result of complex learning'. [12] The learning process seems to depend mainly on visual exploration: a two-year-old child will recognize a triangle that has been rotated by 120 degrees only after rotating its own head; [13] but even in the adult, perception is bound up with exploratory motions of the eyes, from conscious movements to the involuntary, minute motions ('drift', 'flick', 'tremor') which move the image across the fovea when the eye seems 'fixed' on a stationary object. Identification of a triangle or square seems to depend on serial scanning of its contours. [14] With practice, the scanning motions may become summary or subliminal; but when scanning is artificially eliminated by a mechanical device, the image disintegrates into fragments. 'Wir tasten mit unserem Blick das Sehfeld ab', Exner wrote in 1891 [15] (We finger over the visual field with our gaze).

  Thus with the exception of brightness and colour constancy, and a few other primitive 'innate skills', visual perception is inextricably bound up with learning, i.e. with memory. What we perceive in audition is not the linear pulse of pressure-variations arriving at the eardrum, but an 'inferential construct' of individual voices, instruments, musical or verbal phrases; and what we perceive in vision is not the camera-image on the retina but the 'inferential construct' of people and objects which preserve their constant shape and size, regardless of angle and distance. The eye may be a camera, but immediately behind its lens there is a series of compensating, correcting, and retouching devices -- the perceptual matrices of skilled vision.

  Knowing and Seeing

  The best-known among these are the matrices responsible for the visual constancies and illusions which are found in every elementary textbook. Less attention has been paid to the modification of automatically functioning perceptual matrices by verbal leaning and verbal suggestions, by attitude and expectation. [16] Mitscherlich, an outstanding observer, had denied that there existed any structural difference between the two types of tartaric acid of opposite optical activity. Pasteur, using the same apparatus, saw at once the asymmetric facets on the tartar crystals because his hypothesis on molecular structure demanded that they should be there. [17]

  Perceptual matrices function not only autonomously, but display considerable 'self-assertion'. This is shown in a simple but drastic manner by the difficulty of breaking the arrow illusion:

  a = b

  Another example of 'self-assertion' is the tiresome insistence with which a tune will go round and round in your head; or the infuriating messages -- 'I told you so' -- 'I told you so' rapped out at the rate of once per second by the wheels of your railway-carriage. A more advanced but equally typical illustration for 'knowing is seeing' is this quote from Babbage: 'I will prepare the apparatus, and put you in such a position that [Fraunhofer's dark lines] shall be visible, and yet you shall look for them and not find them: after which, while you remain in the same position, I will instruct you how to see them, and you shall see them, and not merely wonder you did not see them before, but you shall find it impossible to look at the spectrum without seeing them.' [18]

  A pretty illustration of perception impregnated by previous knowledge is in the drawing opposite.

  The bear climbing on the other side of the tree is purely inferential. Yet you see the semi-circles plus four strokes as his paws.

  Levels of Memory

  Perception and memory cannot be un-scrambled. L
et us consider briefly a few types of 'mnemic' processes which intervene on various levels of the hierarchy.

  On the lowest, peripheral level we find automatisms designed to reduce redundancy and to 'compress' the input. It would be uneconomical if each receptor in the retina would signal to report stimulation from a uniformly illuminated area; hence lateral inhibition between neighbouring receptor units, combined with light-adaptation, will filter down the input to signals which report only the relevant spatial differences in illumination -- i.e. the contours of the illuminated area. [19] Likewise, the eye adapts to uniform motion -- as seen in the illusion of reverse movement when the real movement stops (the 'waterfall-illusion'). These automatisms could be called memory processes confined to the psychological present (in the broad sense of 'memory = modification of responsiveness by experience').

  More lasting are after-images -- once regarded as the prototypes of 'photographic' memory. In fact, however, the after-image 'improves' on the original by achieving greater regularity and simplicity. Goethe was the first to observe that the after-image of a square will gradually become transformed into it circle -- a figure of greater symmetry. Rothschild's [20] experiments showed that only regular figures with 'good' contour produce stable after-images; that figures with gaps in their contours appear 'closed' in the after-image; that small irregularities disappear, and elements which 'do not belong' to a figure, such as a squiggle or tail attached to a square, become detached from it and come and go as independent units. Thus the reproduction is far from mechanical, and is controlled both by intrinsic codes and from higher centres. [21]

  There are some significant parallels between after-images and images which have been artificially stabilized on the retina. The latter are obtained by a tiny projector, mounted on a contact-lens worn by the subject. The illuminated target-figure is fixed at the other end of the projector; it moves with the involuntary scanning motions of the eyeball, and its projection remains thus fixed on the same spot of the retina -- subtending a visual angle of two degrees in a patch of light of five degrees, with darkness all round. Under these conditions the perceived image will vanish and reappear much as after-images do -- either as a whole, or it will disintegrate into parts. If the latter is the case, the fragments will be meaningful in one way or the other: a human face will break up into its specific features or groups of features -- profile or top of the head; and a composite monogram will break up into the individual letters and numbers which were hidden in it. Conversely, the elements in a meaningless pattern of curlicews will at first fade and reappear in varions combinations; but after a while they will organize themselves into stable sub-wholes. Central processes fraught with past experience exert an obvious influence on these phenomena; and so does the subject's attitude.

  The next step leads to eidetic images. These occupy an intermediate place in the memory hierarchy between after-images and the 'picture-strip' type of recall. In their direct sensory impact they are comparable to hypnagogic images and close to hallucinations. The experimenter directs the subject to inspect a picture for about thirty seconds without staring (to eliminate after-images), then to look at a grey screen. The average person sees nothing; the eidetic 'projects' the image onto the screen and behaves as if the picture were actually there. He can focus on a detail, point out its exact position on the screen, count the buttons on a coat, the number of spokes on a wheel, and 'read' the letters in a foreign-language text forward or backward. [22] It seems, therefore, that eidetic images 'are seen in the literal sense of the word'. [23] Eidetic recall may be limited to a short interval after inspecting an object or picture, or extend to 'minutes, days, years'. [24] Analogous phenomena seem to exist in other sense-modalities: Beethoven, Mozart, Wagner, and Elgar were supposed to be able to 'hear at will the full texture of an orchestra'. [25]

  Eidetic memory, though rare in adults, seems to be quite common in children before puberty; 'in certain regions,' according to Kluever, 'eighty to a hundred per cent of the children are reported eideletic'. [26] This is a striking confirmation of the commonplace that the child lives in a world of images of great vividness, whereas the average adult's images are grey shadows. The eidetic type of memory seems to be irretrievably lost, in all but exceptional cases, with the transition from the perceptual and affective to the conceptual and symbolic mentality. Pictorial memory, as we saw, belongs to a phylogenetically and ontogenetically earlier level of the mnemic hierarchy.

  Yet if we expect the eidetic image to be a true photographic record, we shall again be disappointed. All reports agree that the development of the image on the screen depends on the child's interest in the picture as a whole, and in its details. Exciting details come out sharply, while adjacent parts may remain blank, blurred, or even appear in complementary colours. Pictures which have no meaning for the child do not appear on the screen. [27] The objects in the image can be moved about, and their colour and size can be changed at will, or in response to verbal suggestions. Synaesthetic phenomena also enter: if the subject's arms are pulled while he is inspecting a horizontal line, the eidetic image of the line will be lengthened; images of the Mueller-Lyer illusion may be lengthened 'by as much as two yards'. [28]

  Image and Meaning

  A further step upward in the hierarchy leads to what I have called the 'picture-strip' kind of memory with emotive significance; and lastly we arrive at the phenomena referred to as 'memory images' in ordinary parlance.

  An image is defined in Drever's Dictionary as 'a revived sense experience, in the absence of the sensory stimulation'. But since most of the sensory stimulation has been irretrievably lost in the filtering-processes of memory formation, only some exceptionally sharp, vivid details are perhaps capable of being 'revived' or 'reproduced'; the remainder of the experience must be 'reconstructed'. It has been known for a long time that introspective reports on 'visual memory images' are largely based on self-deception. Visual recall -- as Semon once wrote -- 'renders only the strongest lights and shadows'; [29] but strictly speaking, even shadows are absent from visual images -- as they are from Chinese paintings; and so are, as a rule, all but the crudest shades of colour. The normal adult's memory-images are much vaguer, sketchier in outline than he is wont to believe; in most cases when he believes that he possesses a visual image of a thing, he is really referring to aggregates of simplified perceptual schemata, held together by conceptual links.

  This has been amply demonstrated (cf. Book One, pp. 346 seq.) In the Binet-Mueller test [30] the subject is directed to memorize a letter square (comprising sixteen or twenty-five letters in random distribution) until he thinks that he has formed a visual image of it, and can 'see' it in his mind's eye. But when he is asked to read the letters in his image in backward order or diagonally, he will take up to ten times longer than when reciting them in their proper serial order from left to right.

  Another classic test is the drawing of elephants by patients suffering from a form of aphasia which impairs symbolic thought but leaves perceptual faculties intact -- a test first used by Pierre Marie, and later by Henry Head: [31]

  Case No. 8 ('semantic' asphasia): Asked to draw an elephant, he moved his pencil about aimlessly, saying 'I can't get the idea'. Then he suddenly drew the outline of the head, back and belly, adding the four legs and an eye; the tusks were indicated but he omitted the trunk. I asked 'What is the characteristic of an elephant?' To this he replied, 'Its trunk; I see I have omitted to put in his trunk'. Case No. 11 . . . made an incomplete drawing of an elephant to order. Asked if he had left out anything, he replied 'His ear', and made a mark on the side of the head. When I inquired 'Has an elephant got anything else?', he answered 'Trunk, eye, tail, toes', marking in each object in turn as he named it; but he forgot the tusks. Case No. 21 (a woman of high intelligence with some semantic disorder): When I requested her to draw an elephant she produced a picture distinctly resembling this animal, except that she gave it a bushy tail and forgot the tusks. After she had finished she exclaimed 'I have
n't put the tusks in. I can't remember where they come. They come from just below the eye, I think; but I don't know. I believe they are teeth and should come out of the top of the jaw really.'

  The quotations show that the visual image of the elephant was not in fact a 'perceptual whole' but a mélange of perceptual and conceptual entities; the glue which held the visual parts together was meaning. Thus in a number of drawings the tusks at first appeared on top of the elephant's head as if they had been horns; and only when their function was remembered were they put in their correct place.

  We have now reached the boundary between the perceptual and symbolic hierarchies -- the highest level of perceptual integration, where symbolic coding must take over if further progress in learning is to be achieved. The schematized visual forms of trunk, legs, tusks, seem to be the upper limit of the patient's capability of forming true perceptual Gestalt-traces. When it comes to reproducing the 'image' of the complete elephant, the visual pattern of the tusk is manipulated as a symbolic unit, labelled a 'tusk'. It is, once more, a double-faced entity: one side is a complex and flexible perceptual whole, the other is a semantic unit which signals the word 'tusk' and is activated by the same signal.