The Ghost in the Machine

  Memory cannot be a store of lantern-slides and tape-recordings, nor of S-R building-blocks; so much is evident. But the alternative hypothesis which I have suggested -- that memory is 'dissectible' into hierarchies with different criteria of relevance -- is, frankly, speculative. However, some modest evidence for it can be found in a series of experiments which James Jenkins and I carried out in the psychological laboratory at Stanford University .*

  * The results were published in a technical paper [1]; the gist of the experiment was to show to each subject for a fraction of a second only (by means of an apparatus called a tachistoscope) a number of eight or nine digits, and then let him try to repeat the sequence. The results of several hundred experiments show that a highly significant number of errors (approximately fifty per cent) consisted in the subject correctly identifying all numbers in the sequence, but inverting the order of two or three neighbouring digits. This seems to confirm that the identification of individual digits, and the determination of their sequential order, are carried out by separate branches of the perceptual hierarchy.

  Two Types of Memory

  The 'colour-printing' hypothesis goes some way towards explaining the puzzling phenomena of recall, but it is based solely on the abstractive type of memory, which alone cannot account for the extreme vividness of the 'vivid fragments' or 'picture-strips' mentioned at the beginning of this chapter. After some forty years, I can still hear the voice of the great Austrian actor, Alexander Moissi, whispering the last words of a dying man: 'Give me the sun.' I have forgotten what the play was about, even its author it may have been Strindberg, Ibsen or Tolstoy -- except for the hallucinatory clarity of that one fragment, torn from its context. Such fragments that have survived the decay of the whole to which they once belonged -- like the single lock of hair on the mummy of an Egyptian princess have an uncanny evocative power. They may be auditory -- a line from an otherwise forgotten poem, or a chance remark by a stranger overheard on a bus; or visual a gesture of a child, a mole on a schoolmaster's face; or even refer to taste and smell, like Proust's celebrated madeleine (a French pastry, not a girl). 'There exists a method of retention which seems to be the opposite of memory-formation in abstractive hierarchies. It is characterised by the preservation of vivid details, which, from a purely logical point of view, are often irrelevant; and yet these quasi cinematographic details, picture-strips or "close-ups," which seem to contradict the demands of parsimony, are both enduring and strikingly sharp, and add texture and flavour to memory.' [2]

  But if these fragments are so irrelevant, why have they been preserved? The obvious answer is that while irrelevant from the point of view of logic, they must have some special emotive significance -- which may be conscious or not. Indeed, such 'vivid fragments' are usually described as 'striking', 'evocative', 'nostalgic', 'frightening', or 'moving' -- in a word, they are always emotionally coloured. Thus among the criteria of relevance which decide whether an experience is worth preserving, we have also to include emotional relevance. The reason why a particular experience should have this kind of relevance may be unknown to the subject himself; it may be symbolic or oblique.

  Nobody -- not even a computer theorist -- thinks all the time in terms of abstractive hierarchies; emotion colours all our perceptions, and there is abundant evidence to show that emotional reactions also involve a hierarchy of levels, including some ancient structures in the brain which are phylogenetically much older than the modern structures concerned with abstract conceptualisations (see Chapter XVI). One might speculate that in the formation of picture-strip memories these older, primitive levels in the hierarchy play a dominant part. There are some further considerations in favour of such a hypothesis. Abstractive memory generalises and schematises, while the picture-strip particularises and concretises -- which is a much more primitive method of storing information.*

  * The term 'information' in modern communication-theory is used in a more general sense than in common parlance. It means any input which 'informs' the organism, i.e., reduces its uncertainty. Thus information includes anything from the colour and taste of an apple to the Ninth Symphony of Beethoven. Irrelevant inputs -- i.e., those which do not reduce uncertainty -- convey no information and are called 'noise' -- on the analogy of a noisy telephone line.

  Abstractive memory may be compared with insightful learning, the picture-strip with conditioning. It may also be related to so-called eidetic images. It has been experimentally established [3] that a considerable percentage of children have this faculty. The child is told to fixate his eyes on a picture for about fifteen seconds, and is afterwards able to see it 'projected' on an empty screen, to point out the exact location of each detail, its colour, etc.

  Eidetic images occupy an intermediary position between retinal after-images and what we commonly call 'memory images'; Kluever speaks of these three types or levels of visual memory, and seems to imply that they are hierarchically ordered. Unlike after-images, eidetic images can be produced at will, and after long intervals (even years). They are like hallucinations, except that the child knows that the picture he sees is not 'real'.

  But though quite common in children, eidetic memory fades with the onset of puberty and is rare among adults. Children live in a world of vivid imagery: the eidetic child's way of 'imprinting' pictures on the mind may represent a phylogenetically and ontogenetically earlier form of memory-formation -- which is lost when abstractive, conceptual thinking becomes dominant.

  Images and Schemata

  Leaving eidetics and picture-strips aside, when normal adults talk about their memory images and assert that they can literally 'see' a remembered scene or face in their mind's eye, they are usually victims of a subde form of self-deception. One way of showing this is the Binet-Muller test. The subject is asked to concentrate on a letter-square of, say, five rows of five letters each, until he thinks that he has formed a visual image of the square which he can 'see' in his mind's eye. When the square is taken away, he can in fact fluently 'read' out the letters -- or so he thinks. But when asked to 'read' the square back to front or diagonally, he will take up to ten times longer. He honestly believes that he has formed a visual image, whereas in fact he has learned the sequence by rote; if he could really 'see' the square, he could read it in all directions with equal speed and ease.

  This fallacy has been known for a long time. One of the earliest students of the subject, Richard Semon (who coined the word 'mneme' for memory), wrote half a century ago that visual recall 'renders only the strongest lights and shadows'. In fact, even shadows are usually absent from visual memories, and all but the crudest shades of colour. An image is defined as 'a revived sense-experience in the absence of sensory stimulation' [4]; but since most details of the experience were lost in the filtering process of memory-formation, our visual images are much vaguer and sketchier than we are wont to believe. They are skeletonised visual generalisations -- outlines, patterns, schemata -- abstracted from the original output by several interlocking visual hierarchies, much as the melody, the timbre of voice and the words are extracted from the Caruso aria.

  We use various, often confusing, words for these optic schemata -- confusing because visual configurations are not easily translated into verbal terms. Yet the caricaturist can evoke the face of Hitler or Mao by a surprisingly few strokes, which schematise what we call a 'general impression'; adding perhaps a 'vivid detail', by sticking a cigar in Churchill's mouth. When we try to describe a person's face, we use expressions like 'bony', 'humorous', 'brutal', 'sad'. Verbally, each of these attributes is extremely difficult to define; visually, they are generalisations stripped of detail, but each definable by a few strokes of a pencil: they are perceptual holons.

  Recognising a person does not mean matching his retinal image against a lantern-slide in the memory-store containing his photographic likeness; it means subjecting the input to a hierarchy of scanning devices which extract from it certain basic configurations -- the 'R-nesses', so to speak. Several
perceptual hierarchies may collaborate in the task. A face, or a landscape, may have a 'melody', a 'timbre', a 'message' and several other attributes. My attitude to the person or landscape will determine which aspects are to be considered as relevant, to be abstracted and stored, and which to be filtered out. For purposes of recognition, the 'melody' alone may be sufficient. But the recall of the face in its absence will be the more complete the more branches of the perceptual hierarchy have participated in retaining it. The richer the network connecting them, the more effectively it will compensate for the impoverishment of experience in the process of storing it. The outstanding memories which some great men are said to have possessed may be due to this multi-dimensional way of analysing and storing experiences.*

  * In the language of the information theorist: 'When information is put in outline form, it is easy to include information about the relations among the major parts and information about the internal relations of parts in each of the sub-outlines. Detailed information about the relations of sub-parts belonging to different parts has no place in the outline and is likely to be lost. The loss of such information and the preservation mainly of information about hierarchic order is a salient characteristic that distinguishes the drawings of a child or someone untrained in representation from the drawing of a trained artist' (Simon [5]).

  But for the great majority of people, recall is much less of a pictorial nature than they believe -- see the experiment with the letter square. We overestimate the precision of our imagery, as we overestimate the precision of our verbal thinking; quite often we think that we know exactly what we want to say, but ah, when it comes to putting it on paper! We are unaware of the blurs and gaps in our verbal thinking, as we are unaware of the missing detail, the empty spaces between the visual schemata.

  Learning by Rote

  The dullest sort of memory, which I have not mentioned so far, consists of word-sequences which have been learned by rote. But even here we find hicrarchic order. The items memorised are not single elementary bits, but larger holons which tend to form patterns. A poem learned by heart is given coherence by patterns of rhyme, rhythm, syntax and meaning, superimposed on each other on the colour-print principle. The job of memorising is thus reduced to fitting the patterns together and filling in the gaps they leave. The same applies to learning a piano sonata, where the structure of the musical holons -- the architecture of movements, of themes and variations, development and recapitulation, rhythm and harmony -- is equally obvious. Where the data to be stored show no apparent cohesion, as in the case of memorising the dates of battles and reigns, or a string of nonsense syllables, all sorts of mnemonic devices or jingles will be invented to provide some structural pattern.

  Thus even rote-learning is never purely mechanical. A certain amount of 'stamping in' by repetition is often indispensable to provide cohesion. How much 'stamping in' is needed depends on the meaningfulness of the task, and on the subject's capacity for comprehending it. At one extreme there is the dog in the Pavlovian laboratory, who needs days or weeks of monotonously repeated experiences to cotton on to the fact that the figure of an ellipse shown on a cardboard signals food, but a circle does not. No wonder -- for outside the laboratory, food is not signalled by ellipses on cardboards, and the dog's perceptual hierarchies are not attuned to treating them as relevant events. Similar considerations apply to Thorndike's cats in puzzle-boxes and Skinner's pigeons. They are all given tasks to learn for which they lack the native equipment, and which they can only learn by 'stamping in'. To proclaim this procedure to be the paradigm of human learning was one of the grotesque aberrations of flat-earth psychology.*

  * For a more detailed discussion, see The Act of Creation, Book Two, Chapter XII.

  Gestalt theorists, on the other hand, are inclined to equally extreme views of the opposite kind. They would maintain that true insightful learning excludes all trial and error and is based on a total understanding of the 'total situation'. In the present theory, insight and understanding are regarded as matters of degree, and not, as the Gestalt school holds, an all-or-nothing affair. Insight depends on the multi-dimensional analysis of the input in its various aspects, on extracting relevant messages from irrelevant noise, identifying patterns in the mosaic until it has become saturated, as it were, with meaning.

  To sum up: we must assume the existence of multiple, interlocking hierarchies of perception which provide the multi-dimensionality or multi-colouration of experience. In the process of storing memories each hierarchy strips down the input to bare essentials, according to its own criteria of significance.

  Recalling the experience requires dressing it up again. This is made possible, up to a point, by the co-operation of the hierarchies concerned, each of which contributes those factors which it has deemed worth preserving. The process is comparable to the superimposition of colour-plates in printing -- or of the wallpaper-maker's several stencils. Added to this are touches of 'vivid detail', perhaps fragments of eidetic imagery, which carry a strong emotive charge -- and the result is a kind of collage, with glass eyes and a strand of genuine hair stuck onto the hazy schematised figure.

  It may also happen that fragments of different origin are mistakenly incorporated into the collage included in the recall of experiences to which they do not belong. For memory is a vast archive of abstracts and curios, which are all the time being rearranged and revalued by the archivist; the past is constantly being re-made by the present. But most of the making and remaking is not consciously experienced. The canons of perception and memory operate instantaneously and unconsciously; we are always playing games without awareness of the rules.

  VII

  THE HELMSMAN

  The human being is the highest self-regulating system. Ivan Petrovich Pavlov

  I have used the terms 'interlocking' or 'interlacing' hierarchies. Of course hierarchies do not operate in a vacuum. The liver is part of the digestive, the heart of the circulatory, system; but the heart is dependent on the glucose which the liver provides, and the liver depends on the correct functioning of the heart. This truism of the inter-dependence of the various processes in an organism is probably the main cause of the confusion which has obscured from view its hierarchic structure. It is as if the sight of the foliage of entwined branches in a dense forest made us forget that the branches originate in separate trees. The trees are vertical structures. The meeting points of branches from neighbouring trees form horizontal networks at several levels. Without the trees there could be no entwining, and no network. Without the network each tree would be isolated and there would be no integration of functions. Arborisation and reticulation (from reticulum = net) seem to be complementary principles in the architecture of organisms.

  To get a possible misunderstanding out of the way, I must insert here a rather obvious remark. A forest consists of a multitude of trees. A living orgasm is an integrated whole -- a single tree. And yet I have been talking of perceptual and motor hierarchies as if they were separate entities. In fact, of course, they are merely main branches on the same tree, or 'sub-hierarchies'. But to call them that would be unnecessarily pedantic, as each branch of a hierarchy is itself hierarchically structured. Thus it is often convenient to regard the Foreign Office and the War Office as separate hierarchies, although they are branches of Government joined at Cabinet level.

  Sensory-Motor Routines

  The most obvious example of interlocking hierarchies is the sensory-motor system. The sensory hierarchy processes 'information' and transmits it in a steady flow to the conscious ego at the apex; the ego makes decisions which are spelled out by the downward stream of impulses in the motor hierarchy. But the apex is not the only point of contact between the two systems; they are connected by entwining 'networks' on various levels.

  The network on the lowest level consists of so-called local reflexes. They are short-cuts between the ascending and descending flow, like loops connecting the opposite traffic streams on a motor highway: routine reactions to
routine types of stimuli, like the knee-jerk, which do not require the intervention of higher mental processes. The level to which decision-making is referred depends on the complexity of the situation. The knee-jerk, or the blink-reflex, is usually completed before the stimulus has reached awareness.

  One of the fundamental errors of the crude Watsonian brand of Behaviourism was the assumption that complex activities result from the summation of isolated local reflexes. We now know that the opposite is true, that local reflexes are the last to make their appearance in the development of the nervous system in the embryo: 'Behaviour develops in man . . . by the expansion of a total pattern that is integrated as a whole from the beginning, and by individuation of partial patterns (reflexes) within the unitary whole' (Coghill [1]). Moreover, reflexes are influenced by higher levels of the hierarchy: even the knee-reflex goes haywire if the patient knows what the doctor is up to. Human behaviour is not a succession of knee-jerks and eye-blinks, and any attempt to reduce it to these terms leads again to flat-earth psychology.