Metaphors Of Memory
By Steven Rose©

We take our technological world, and our memories within it, very much for granted. We leave messages on answer phones or computers for absent friends, we consult our diaries as to free dates and write notes to colleagues arranging dinner, a theatre or a meeting; we check the fridge and write ourselves a shopping list. Each of these is an act of individual memory - but an act in which we have manipulated technologies external to ourselves in order to aid, or supplement, or replace, our internal brain memory system. It was not always thus; individual our memories may be, but they are structured, their very brain mechanisms affected, by the collective, social nature of the way we as humans live. For each of us as individuals and for all of us as a society, technologies, some as old as the act of writing, some as modern as the electronic personal organizer, transform the way we conceive of and the way we use memory. To understand memory we need also to understand the nature and dynamics of this process of transformation.

The greater part of the history of humanity not only pre-dates modern technologies; it even pre-dates writing. For such early human societies, records, individual life histories, just as much as histories of family and tribe, were oral. What failed to survive in an individual's memory, or in the spoken transmitted culture, died for ever. People's memories, internal records of their own experience, must have been their most treasured - but also fragile -possessions. In such oral cultures, memories needed to be preserved, trained, constantly renewed. Special people, the elderly, the bards, became the keepers of the common culture, capable of retelling the epic tales which enshrined each society's origins. Then, each time a tale was told it was unique, the product of a particular interaction of the teller, his or her memories of past stories told, and the present audience. Walter Ong describes how in modern Zaire a bard, asked to narrate all the stories of a local hero, Mwindo, was amazed; no-one had ever performed them all in sequence before. Pressed to do so, he eventually narrated all the stories, partly in prose, partly in verse with occasional choral accompaniment. It took him twelve exhausting days whilst three scribes took down his words. But once written, Mwindo had become transformed. He no longer existed as a continued, remembered recreation of past stories. Instead he had become fixed in the linear memory form demanded by modern cultures. ¹

Although we still retain a concept of memory in this deep, collective sense, the new technologies change the nature of the memorial processes. A video or audiotape, a written record, do more than just reinforce memory; they freeze it, and in imposing a fixed, linear sequence upon it, they simultaneously preserve it and prevent it from evolving and transforming itself with time, just as much as the rigid exoskeleton of an insect or crustacean at the same time defends and constrains its owner. For instance, when, in 1990, world Jewish leaders convened at Wannsee, the lakeside villa where Hitler, Heydrich and others, nearly fifty years earlier, had drawn up the plans for their 'final solution' to the 'Jewish problem', the Nobel Peace Prize winner Elie Wiesel wrote that the intention was to demonstrate that 'memory is stronger than its enemies ... most German men and women in the past refused to speak, refused to remember.' An act of group memory on the one part, confronting an act of social amnesia on the other. Yet an act of memory reinforced not merely by oral tradition but by the written texts, by audio and above all by the visual images of photography and the cinema, terrible images which become fixed in the minds and memories even of those who were distant from the events. The contrast with the old oral cultures could not be greater.

And of course similar strongly reinforced collective memories -and amnesias - underlie many present-day national and ethnic conflicts. As a youngster I was brought up never to forget 'next year in Jerusalem'. When, in 1982, after the massacres of Palestinians in the camps of Sabra and Chatilla I visited the Lebanon, I met many young Palestinians who had certainly never been there but who 'remembered Jaffa' - or even Jerusalem - and their expropriated familial homes in what was now Israel at least as strongly and with as much feeling as those who convened at Wannsee in their act of collective memory. Think of the ways in which the significance of Kosovo for Serbs and Albanians, or of the temple/mosque at Alyodha for Hindus/Muslims becomes reinforced by the collective sharing of technologically preserved images.

The psychoanalyst Jung based his theory of mind in part on the claim that such collective memories were racial and had become deeply inscribed in our biological as well as cultural inheritance. I of course mean nothing of the sort here; rather I am talking about mechanisms of retention and transmission - the sharing and collectivization of memories. In the Soviet Union, the society set up to commemorate the victims of the Stalin era is known as Memorial. But equally, the political organization of the extreme right, of Russian nationalism and anti-semitism, is called Pamyat -Memory. As will become apparent, in the understanding of both the social and the biological functions of memory, the elucidation of forgetting, of individual and social amnesia, provides as powerful a clue as does remembering.

Collective and individual memories, the changing technological forms which enrich and constrain our memories and provide the analogies by which we endeavour to explain them: these are the themes of the present chapter. And, as will become clear in Chapter 5, the technological evolution by which the fluid memory of oral cultures becomes fixed and disciplined in a computer-driven industrial society, finds strange echoes also within individual human development.

The Ancient Arts of Memory

The ancient philosophers were distinctly dubious about the merits of a written culture. Thus Plato describes Socrates as claiming that writing is inhuman, in that it pretends to establish outside the mind what in reality can only be in the mind. As Ong points out, writing reifies, it turns mental processes into manufactured things. Writing destroys memory; those who use it, Plato has Socrates argue, will become forgetful, relying on an external source for what they lack in internal resources. Writing weakens the mind. Instead, memories should be trained, as the Zairean bard's memory must have been. This training is the discipline known as mnemotechnics — a discipline which must have been invented separately at many times and in many cultures. Within western culture, there is a clear history of this mnemotechnic tradition, running back to Greek times, though the written record of the method is not Greek but Roman, and first appears in De Oratore, a famous text on the art of rhetoric - that is, of argument and debate - by the Roman politician and writer Cicero. In it, Cicero attributes the discovery of the rules of memory to a poet, Simonides, who seems to have been active around 477BCE.

The Simonides story appears and reappears throughout Roman, medieval and Renaissance texts. In its basic form it tells how, at a banquet given by a Thessalonian nobleman, Scopas, Simonides was commissioned to chant a lyric poem in honour of his host. When he performed it, however, he also included praise of the twin gods Castor and Pollux. Scopas told the poet he would only pay him half the sum agreed for the performance and that he should claim the rest from the gods. A little later Simonides received a message that two young men were waiting outside to see him. During his absence the roof of the banqueting hall fell in, crushing Scopas and his guests and so mangling the corpses that their relatives could not identify them for burial. The two young men were the gods Castor and Pollux, and they had thus rewarded Simonides by saving his life, and Scopas apparently got his comeuppance for meanness. But - and this is the crucial bit of the story - by remembering the sequence of the places at which they had been sitting at the table, Simonides was able to identify the bodies at the banquet for the relatives. This experience, as Cicero tells the story, suggested to Simonides the principles of the art of memory of which he was said to be the inventor, for he noted that it was through remembering the places at which the guests had been sitting that he had been able to identify the bodies. The key to a good memory is thus the orderly arrangement of the objects to be remembered.

He inferred that persons desiring to train this faculty must select places and form mental images of the things they wish to remember and store those images in the places, so that the order of the places will preserve the order of the things, and the images of the things will denote the things themselves, and we shall employ the places and images respectively as a wax writing-tablet and the letters written on it. ²

Such rules are designed to relate a collection of items which do not have a particular relational logic, but are contingent, like the guests at the banquet, to some structure whose logic is apparent or at least can readily be remembered because of its striking features. Memories, in such mnemotechnic systems, could thus be stored by remembering some familiar environment, commonly a house with a series of rooms, or a public space with prominent monuments and buildings, and 'placing' the items to be remembered in an appropriate sequence within the environment. One could then recall them, for instance during the course of a speech or recitation, by mentally walking through the environment, visiting each location in turn. The ancient texts refer to this method of memorizing as 'artificial' memory, by contrast with given or natural, untrained memory; the artifice of memory seems a project as dear to the ancients as it is to present-day computer enthusiasts. A further Latin text, of unknown authorship, known simply as Ad Herennium, essays a definition of memory, as firm retention / comprehension in the mind of the matter, words and arrangement of objects. The text dwells on how to choose the images which above all give guidance as to the arrangement, which is seen as the key feature of an effective memory:

We ought then to set up images of a kind that can adhere longest in the memory, and we shall do so if we establish likenesses as striking as possible ... if we assign to them exceptional beauty or singular ugliness, if we dress some of them with crowns or purple cloaks, for example, so that the likeness may be more distinct to us, or we somehow disfigure them, as by introducing one stained with blood or soiled with mud or smeared with red paint ... this too will ensure our remembering them more readily. ³

These methods were no mere personal idiosyncrasy or device of a great orator like Cicero, who could apparently speak in the senate for days on end without using notes. Similar descriptions crop up, as Frances Yates describes in her well-known book The Art of Memory, ⁴ in other classical texts. Certain Roman generals were supposed to have used them to recall the names of their soldiers; thus Publius Scipio was said to be able to recognize and name all of his entire army of 35,000 men. Such mnemotechnics were the forerunners of a tradition which ran through the medieval and Renaissance periods and is still alive today.

----

Hell as artificial memory.

Paradise as artificial memory.

Fig 4.1 From Cosmas Rossellius Thesaurus Artificiosae Memoriae, Venice, 1579.

----

During much of the Middle Ages they became debased into mere crude devices for remembering numbers and letters. Easily visualized picture sequences or inscribed wheels were supposed to enable those who learned them to recall the ordering of devotional exercises or catalogues of virtues and vices, rather as children's spelling books today offer the pictogram sequences of 'A is for Apple; B is for Ball ...'

But increasingly, and especially from the fourteenth century on, mnemotechnics became more daring. The locus for the memory-images came to be described as a theatre, a memory theatre, in which symbolic statues, rather like those one might have expected to find in a Roman forum, were placed, and at the base of each statue could be stored the item to be memorized. During the early Renaissance period these imaginary theatres became increasingly complex, with gangways, tiers of seats, and classical statues representing virtues, vices and other key figures. But where in the past the exponents of the mnemotechnic art might have envisaged themselves as spectators at such a theatre, looking inward to the stage as an elaborate set full of memory cues, in the Renaissance memory theatres the mnemotechnician was supposed to look outward from the stage, the actor facing an audience whose location in their ordered ranks of seats provided the sequence clues.

----

Fig 4.2 Grammar as memory image

Visual alphabets used for the inscriptions on grammar. From Johannes Romberch Congestorium Artificiose Memorie, edition of Venice, 1533.

----

The theatres even became agents of religious propaganda. In 1596 the Jesuit missionary Matteo Ricci offered a 'memory palace' to the Chinese whom he was bent on converting to his faith. He told them the size of the palace would depend on how much they wanted to remember: the most ambitious construction would consist of several hundred buildings of all shapes and sizes, the more the better, although more modest palaces, temple compounds, government offices, merchants' meeting lodges or even simple pavilions might suffice. Ricci found images likely to be familiar to his Chinese hosts to place in the imaginary rooms and pavilions of the equally imaginary palace to act as the memory loci for the storage of concepts and ideas, though I have to say that I find the relationship of the entire elaborate system to Christian theology somewhat elusive. ⁵

In the hands of Galileo's contemporary but far more dangerous heretic, Giordano Bruno, who unlike Galileo would not recant and was burned by the Inquisition, the theatres also became a key feature of occult, hermetic philosophy. For Bruno they became a way of classifying and hence penetrating to the mysterious core of the universe. Memory gave power over nature. Memory theatres became the very models for heaven and hell (Dante's systematic descriptions of the circles of both in his Divina Commedia have been claimed to be derived from such mnemotechnic systems). A debased version of Bruno's vision and philosophy is still around today. Turn to the classified section of a Sunday newspaper and you will find ads along the lines of 'Loss of memory? A well-known publisher can teach you how to improve it' or 'It may be news to you but the Egyptians knew it long ago ...' On closer inspection many such ads turn out to be placed by an obscure sect calling themselves Rosicrucians, whose ancestry may not be quite as antique as they claim but certainly stretches back to Bruno's day, and retains many Brunian elements. Take up such offers to train your memory and you will even now very likely be given a version of the memory theatre.

By the time of the Renaissance, the memory theatre was turned from a symbolic device, a piece of mental furniture, into an actual construct. In the sixteenth century, and to the disapproval of more rationalist philosophers such as Erasmus, the Venetian Giulio Camillo actually built a wooden theatre crowded with statues which he offered to kings and potentates as a marvellous, almost magical, device for memorizing. Frances Yates even goes on to speculate, daringly, that the lost Globe Theatre of Shakespeare was actually built to the design of a real memory theatre. 'Why', she asks (pp. 173-4), does such a

theatre seem to connect so mysteriously with many aspects of the Renaissance? It is, I would suggest, because it represents a new Renaissance plan of the psyche, a change which has happened within memory, whence outward changes derived their impetus. Medieval man was allowed to use his low faculty of imagination to form corporeal similitudes to help his memory; it was a concession to his weakness. Renaissance Hermetic man believes that he has divine powers; he can form a magic memory through which he grasps the world ... The magic of celestial proportion flows from his world memory into the magical worlds of his oratory and poetry, into the perfect proportions of his art and architecture. Something has happened within the psyche, releasing new powers ...

----

Fig 4.3 A memory theatre (From Robert Fludd's Ars Memoriae.)

The technological metaphor

Real theatre or imaginary, at this point we are already far from Cicero's original intent, and have long transcended Plato's concern that writing might be deleterious to the mind; a technological imperative to harness memory is beginning to emerge. But harnessing memory requires also some sort of effort to understand or explain it, and it is here that the peculiarly double relationship of technology to biology in general and the biology of mind in particular achieves special significance.

Explanation in science proceeds by metaphor. We endeavour to understand how something we don't know works by comparing it to something we do know - or something we can at least imagine we know. Think of one of the most basic of divisions of the known world, the division between animate and inanimate. Within science, the former became the province of biology, the latter of physics. In the pre-technological era within western societies, and in many other cultural traditions, explanations run transitively, in both directions between biology and physics. The irregularities of wind and ram, just as much as the regularities of rivers, the sea and the earth, the stars, sun and moon, are explained animistically, as reflecting the minds or whims of local or global gods, themselves motivated by similar concerns to those that motivate humans. But, equally, animate, biological phenomena are given metaphorical explanations in physical - and increasingly in technological - language. Because biological systems are so complex, they are above all analogized to the most complex, the highest, forms of current technology. Each period, each culture, has such a form - one that David Bolter ⁶ has called its defining technology. Indeed, we periodize ages in humanity's prehistory by such defining technologies - stone age, bronze age, iron age. ^*

* Not just in prehistory; writing about the growth of science as an institution, Hilary Rose and I spoke of 1914-18 as the chemists' war, 1939-45 as the physicist's war. Since then we have entered the age of computer technologists' and even biologists' wars. ⁷

For early cultures one of the most subtle technological forms was that of the potter, who with clay and wheel, glaze and fire, could create shape and pattern. No wonder that for such cultures -and it is a creation myth that turns up again and again in the origin stones of both New and Old Worlds - it is a deity with a potter's wheel who shapes humans and then breathes life into them. Other myths evoke spinning and weaving, as in the loom of life held by the Fates. Memory was and is no stranger to such metaphor; for the ancients, images become inscribed within it - as Cicero puts it in De Oratore - like 'a wax writing-tablet and the letters written on it'. This metaphor resonates down the ages, becoming the basis for philosophical debate in the eighteenth and scientific / ideological debate in the nineteenth and twentieth centuries as to whether humans are born with innate predispositions or as tabula rasa - clean slates on which experience inscribes individual memory.

Today the word memory occurs in a multitude of scientific discourses. Quite apart from the sort of memory that neurobiologists, psychologists and even novelists talk about and with which I am concerned, mathematics and physics, chemistry, molecular biology, genetics, immunology and evolutionary biology, not to mention computer science, all use the term. Why these many uses of the word memory? Do we have here to deal simply with puns - the use of words originating in one context in another, different one - or does the fact that many different scientific discourses use the term cast some light on the mechanisms and processes that may be involved?

Fig 4.4 Sketch of the Globe Theatre based on Robert Fludd (From Frances Yates The Art of Memory)

Fig 4.5 The De Witt sketch of the Swan Theatre (Library of the University of Utrecht)

Can these varied metaphors reveal something about the nature of the processes involved in any phenomenon - even illuminate otherwise unexpected similarities as to process or mechanism between seemingly widely different phenomena — or are they merely figures of speech? In what sense is memory to be taken to be like a wax tablet - or, for that matter, a computer?

One can usefully distinguish between three types of metaphor in science. ⁸ The first is poetic - for example Rutherford's description, early this century, of electrons in orbit around the atomic nucleus as if they were planets revolving around the sun. In using this analogy he surely did not mean that the nucleus and electrons were like the sun and planets or that the forces which related them were gravitational; all that the analogy provides is a useful visual image. The ancient metaphor of the potter's wheel clearly comes into this category.

The second metaphoric mode is evocative, in which a principle from one sphere is transferred to another. Thus until the Middle Ages and the Newtonian revolution, everything that moved seemed to be pushed or pulled by something else. Hence to explain the movement of the sun around the earth, metaphor spoke of horse-drawn fiery chariots.

Finally, one has the metaphor as a statement of structural or organizational identity. Thus when, in the seventeenth century, William Harvey discovered the circulation of the blood and described the heart as a pump, his metaphor had a precise meaning which distinguishes it from the previous two categories. Within the circulatory system, the heart indeed functions as a pump, and organizationally its structure, with valves and emptying and filling phases, resembles at least those pumps which were being mechanically contrived at the time of Harvey's discovery. Treating the heart as a pump enables mathematical models of its action to be made which accurately describe many of its properties.

In which of these senses, then, are wax tablets - or computers -metaphors for brain memory: poetic, evocative or structural? Or are they indeed none of these, but instead merely mischievous?

The Cartesian rupture

With the birth of modern science in Europe in the seventeenth century, the symmetry of analogizing physical forces to animate ones and biological phenomena to technological models was broken. It is important to realize that this is indeed a particularly western phenomenon, and is to be explained by the fact that science was not born as a singleton but as a twin; it emerged and grew to maturity along with particular forms of bourgeois, capitalist social organization and the two shared many philosophical and ideological premises in their understandings of, and approaches to, the natural and social worlds. ^{7 9} The broken symmetry of western science was long resisted in other cultures with alternative indigenous scientific traditions, most notably of course in China, ¹⁰ where the animate/inanimate division of nature, along with other forms of dualism which the western cultural tradition has naturalized, was never drawn so rigidly.

The science that developed in Europe, however, was epitomized by Galileo, Newton and above all Descartes, who between them debiologized the physical world, turning it into 'mere' mechanism. For them, the defining technology was the clock and its associated systems of gears, cogs and hydraulic transmission, which together could generate precisions of mathematically describable motion hitherto unimaginable. Clockwork redefined time, trapped and reduced the hitherto seamless universe into units which could be separately controlled and costed." ^* Hydraulics were a source of power and controlled movement within this mechanical universe. The new physics generated not merely new explanations of the universe but also new technologies, new production systems and new relationships of production between those engaged in them. Europe became set on a course of industrial and imperial expansion which has far from run its course today, and mathematical physics became the defining model of scientific explanation against which all others should be judged. If the very motions of planets, moon and sun could be reduced to simple mathematics - were nothing more than the ineluctable workings out of equations - why not mere biology?

* With the advent of digital as opposed to analogue watches, time becomes divided up and budgeted even more precisely, increasingly divorced from the world time given by the cycles of day and night, the months, seasons and years. In today's world, wearing an analogue as opposed to a digital watch becomes a small act of resistance - a point first made by the radical physicist Maurice Bazin, one of the finest teachers of popular science.

It might of course have been different. Biology, as an organized science, might have developed before physics, and those less mechanical, more goal-directed (teleonomic), functional and evolutionary modes of explanation of the animate world which biologists favour might have become also the model towards which physicists aspired. Reductionism, with its insistence that in 'the last analysis' the world can be explained in terms of atomic/ quantum properties and a few universal equations, would then seem no more than a ludicrous inversion of proper scientific explanation, ¹¹ and biologists would no longer suffer from a sense of physics-envy and an unease about their subject being a 'soft' rather than a 'hard' science. But it was not to be; the technological rather than the biological metaphor dominated, and in the hands of Descartes, living organisms themselves became clockwork, their internal processes powered by complex systems of hydraulics, tubes and valves.

For humans, as is well known, Descartes made a crucial exception. Although everything about their day-to-day functioning was as mechanical as that of any other animal, humans could also think and above all had a soul, whereas, for Descartes, animals were capable only of fixed responses to their environments. Thought and soul were incorporeal entities, but could interact with the mechanism of the body by way of a particular gland, the pineal, located deep in the brain. Descartes chose the pineal for this localization on two grounds. First, whereas other brain structures are all duplicate, in that the brain consists of two more or less symmetrical hemispheres, the pineal is singular, unduplicated; and mental phenomena must of course be unified. And, second, the pineal is a structure found uniquely in humans and not present in other animals. Descartes was of course wrong on both grounds; there are many other non-duplicated structures in the brain, and other vertebrates also possess pineal glands, but the theory-driven logic of his argument remains appealing for those who want to argue, as he did, for the uniqueness of humans: 'It is morally impossible that there should be sufficient diversity in any machine to allow it to act in all the events of life in the same way as our reason causes us to act.' ¹²

The Cartesian split, between mind and body, a dualism which has clouded western scientific and philosophical thinking with its obsessive and misguided worries about the 'mind-brain problem' for the subsequent three centuries, begins here.

----

Fig 4.6 The Cartesian metaphor

An interpretation of his explanation of the automatism that determines the withdrawal of a hand from something that burns it. From L'Homme de Rene Descartes, Paris, 1664.

----

But it is the metaphors of Cartesian clockwork and hydraulics rather than Cartesian dualism which concern me at present. The present-day animal-rights movement has made much of the way in which such thinking enabled Descartes to dismiss the cries of pain of animals when vivisected as no more than the squeaks of poorly oiled machines. The Cartesian view was taken most seriously perhaps by the French tradition of physiology in the nineteenth century - particularly Claude Bernard - in its indifference to animal suffering. ¹³ The modern repudiation of Descartes is of course right, but I would argue that the clockwork metaphor is as damaging in its partitioning and reduction of humans as it is of non-human animals. Descartes may have saved the soul/mind for Catholicism in its Sunday-best garb, twiddling the knobs of mechanism via the pineal, but he left a clockwork human for the remaining six days of the week, debiologized as well as desacralized and open to treatment as a mere bete machine within the developing industrial revolution of the eighteenth and nineteenth centuries. It would only be a matter of time before technology would challenge the Cartesian 'moral impossibility'.

Against this grave philosophical and ideological disservice there are, to be sure, Cartesian achievements. The localization of function to the brain, even in machine-metaphorical form, was no trivial event. The brain as the seat of the mind and soul is not an automatically self-evident proposition, however natural it seems to us today; for Aristotle that function was reserved to the heart, for the ancient Hebrews to the kidneys and bowels. The Galenic medical tradition had demonstrated that nerves originate in the brain and that motor and sensory functions are abolished by brain injuries. But hydraulic thinking centred not on the fatty and unpromising tissue of which the brain was composed but instead on its fluid-filled core, the ventricles, lovingly drawn by early anatomists, none more strikingly than Leonardo.

As a consequence, early hydraulic memory models had memories stored in the ventricles, and animated by a flowing spirit, controlled by a valve between front and rear portions of the brain. In Descartes's version this crucial task was naturally ascribed to the pineal:

Thus when the soul wants to remember something ... volition makes the gland lean first to one side and then to another, thus driving the spirits towards different regions of the brain until they come upon the one containing traces left by the object we want to remember. These traces consist simply of the fact that the pores of the brain through which the spirits previously made their way, owing to the presence of this object, have thereby become more apt than others to be opened in the same way when the spirits again flow towards them. And so the spirits enter into these pores more easily when they come upon them, thereby producing in the gland that special movement which represents the same object to the soul and makes it recognize the object as the one it wishes to remember. ¹⁴

This ingenious description contains the forerunners of many modern ideas about the mechanisms of memory with which the present book deals - and of the capacity of philosophers to see biological problems as straightforward. In this context I am especially fond of Descartes's use of the term 'simply'. If only it were so ...

How are we to understand these Cartesian metaphors of memory? Descartes may have meant his metaphor to be precise, as structurally accurate a descriptor of the brain and its processes as Harvey's of the heart as a pump, but I suggest we can take it as no more and no less than poetic, a way of thinking about a complex human phenomenon which places it not sui generis but as merely one amongst other types of matter in motion.

Through the eighteenth and nineteenth centuries the metaphors of mind and memory steadily shift. With the discovery by Galvani of 'animal electricity' - that frogs' legs twitched when connected by metal wires - the nervous system ceased to be hydraulic and became instead an electrical maze. And within it, the brain became first a telegraphic signaling system and later, at the start of the present century, a telephone exchange, one of the several metaphors favoured by the great neurophysiologist Sherrington. (Another, unforgettable but clearly poetic, Sherringtonian image saw the brain as an 'enchanted loom' weaving patterns in electricity.) Even more than pipes and valves, telegraphs and telephones were surely systems that were like the brain in more than a poetic sense. The telegraph, for instance, converted sense data into symbols - in the hands of Morse and his successors into specific codes for given individual letters - which could be passed over large distances and be decoded at the other end. The telephone was even more promising, for speech was here converted into patterns of electrical flow across a wire. In the telephone exchange metaphor the brain processes messages coming in and going out; signals from eyes connected to muscle contractions in the leg and so forth.

When, during the 1920s, it was discovered that the brain was indeed in a state of ceaseless electrical flux, that electrodes placed on the scalp could detect regular bursts and rhythmic waves of electrical activity changing with thought and rest, sleep and wakefulness, this was instantly assimilated to the telephone exchange model, with subscribers dialing in and being connected to their required addresses by a central operator. But the telephone exchange is merely the paradigm form of the electrical office of the first half of this century. Here, for instance, is the metaphor at its most primitive, from a children's encyclopedia of the period:

Imagine your brain as the executive branch of a big business ... Seated at the big desk in the headquarters office is the General Manager - your conscious self - with telephone lines running to all departments ... Suppose you are walking absent-mindedly in the street and meet your friend Johnny Jones. He calls your name, you stop, say 'Hullo!' and shake hands. It all seems very simple, but let's see what happened during that time in your brain. The instant Johnny Jones called your name, your Hearing Manager reported the sound, and your Camera Man flashed a picture of him to the camera room. 'Watch out!' came the signal to your desk, and at the same instant both messages were laid m front of you. As quick as lightning your little office boy, Memory, ran to his filing case and pulled out a card. The card told you that that voice and that face belonged to a person named Johnny Jones and that he was your friend. Instantly you began issuing orders ... ¹⁵

The computer and artefactual intelligence

From wax tablet to little office boy with a filing case in about 2,000 years doesn't seem like too rapid a rate of metaphoric progress, and to call this even a poetic metaphor would seem to debase the term. But the real challenge to Descartes's 'moral impossibility' came with the defining technology of the second half of the twentieth century, the computer. The immediate antecedents of today's machines, like those of so many other major technologies, are military. They include the logical games of the Cambridge mathematician Alan Turing, pressed into practical use during the code-breaking exercises of British Intelligence at Bletchley Park (no further than a long stone's throw from my own laboratory today) during the 1939-45 war. They were given electronic form by a different set of military requirements - the need to develop effective servo-mechanical devices to calculate elevation and direction so as to fire anti-aircraft guns against rapidly moving targets - techniques developed by the US mathematician Norbert Wiener, who gave the new science a new name, cybernetics, by which it became fashionable in the decades after 1945.

Wiener and fellow mathematician John Von Neumann, together, of course, with US (and some, small-scale, British) industry were responsible, in those years, for giving the new science and the technology it generated electronic form and theory. The military interest - and its impetus for new developments - has never faded in the intervening half-century, and reached a crescendo during the 1980s, a decade of seemingly unbridled spending under the auspices of the Reagan administration's Star Wars program, which demanded computing power on a scale of unparalleled extravagance. Computers which worked like or could replace brains became not merely a science fiction but a serious military goal. 'Intelligent systems' which could replace or supplement skilled, highly trained and expensive humans in flying planes and firing weapons seem an attractive option. Indeed it has become hard these days to attend a scientific conference on themes associated with learning, memory and computer models thereof without finding a strongly hovering US military presence, whether navy, air-force or the somewhat sinisterly acronymed DARPA - the Defense Advanced Research Projects Agency.

That the computer was something qualitatively new was obvious from the start. Certainly, electromechanical calculating machines and their relatives already existed. But the general-purpose computer was far more than just a very fast calculator and storer of data; it could manipulate, compare and transform information in ways that have made possible wholly novel technologies, instrumentation and even the scientific questions one can conceive of asking of the universe. Slowly but with increasing acceleration over the past two decades computer technology has transformed our ways of understanding and operating upon the world. Small wonder that its ideological resonances have been so profound. From the very start, the relationship between computers and minds/brains was at the forefront of the thinking of its inventors and was apparent in their language. For instance Von Neumann's digital computer consists of a central processing unit which carries out arithmetical and electrical operations, and a storage unit, immediately christened by its designers a memory.

A computer memory consists of chips (silicon wafers with transistors engraved on them) which store data in the form of a binary code - that is, each unit can exist in one of two states (O, 1). Implicit in this design of course is that anything that the computer stores and manipulates must first be converted into a form in which it can be represented in this numerical, binary mode as a number of bits (binary units) of information. Information in this sense has a technical - even technological - rather than an everyday definition, which will need further analysis later. Also requiring further note is that implicit in the name given to the information store - the computer memory - is the claim that in some way what the computer is doing in holding and processing binary units of information is analogous to what we as humans do with our own memory.

At first sight, this might be seen as encouraging. Does this language system not describe a physical, inanimate mechanism by analogy to a biological system? And was the failure to do this not what I was regretting about the seventeenth-century Cartesian transition? Sadly, no. As will become apparent, the practical and ideological power of the technology surpasses that of the biology, so that the metaphor reverses itself. Instead of biologizing the computer, we find ourselves challenged by the insistence that human memory is merely an inferior version of computer memory, and that if we want to understand how the human brain works we had better concentrate on studying and building computers.

Nor is this the aberration of a few macho science fiction enthusiasts; it has been central to the agenda of computer designers and their philosopher fellow-travellers from the earliest days. Turing himself began it in 1950 - not long before his suicide - with one of his many logical games. Suppose you were in communication, via a teletype, with a second teletype in an adjoining room. This second teletype could be controlled either by another human or a machine. How could you determine whether your fellow-communicant was human or machine? Clearly the machine would have to be clever enough to imitate human fallibility rather than machine perfection for those tasks for which machines were better than humans (e.g. speed and accuracy of calculating), but equally the machine would have to do as well as a human at things humans do supremely - or else find a plausible enough lie for failing to do so. This is the nub of the so-called Turing test, and he believed that 'within fifty years' a computer could be programmed to have a strong chance of passing this test. ¹⁶

To create a machine that could pass the Turing test has become the holy grail for the generations since 1950 of those committed to the pursuit of what they call, modestly enough, artificial intelligence. But how to go about it? From the beginning, there were two contrasting approaches, which we may characterize, crudely, as reductionist and holistic. Looking back over the period with the benefit of hindsight, one of the pioneers and prophets of the holistic approach offered this fairy-tale account:

Once upon a time two daughter sciences were born to the new science of cybernetics. One sister was natural, with features inherited from the study of the brain, from the way nature does things. The other was artificial, related from the beginning to the use of computers. Each of the sister sciences tried to build models of intelligence, but from very different materials. The natural sister built models (called neural networks) out of mathematically purified neurones. The artificial sister built her models out of computer programs.

In the first bloom of their youth the two were equally successful and equally pursued by suitors from other fields of knowledge. They got on very well together. Their relationship changed in the early sixties when a new monarch appeared, one with the largest coffers ever seen in the kingdom of the sciences: Lord DARPA ... The artificial sister grew jealous and was determined to keep for herself the access to Lord DARPA's research funds. The natural sister would have to be slain.

The bloody work was attempted by two staunch followers of the artificial sister, Marvin Minsky and Seymour Papert, cast in the role of the huntsmen sent to slay Snow White and bring back her heart as proof of the deed. Their weapon was not the dagger but the mightier pen, from which came a book, Perceptrons -purporting to prove that neural nets could never fill their promise of building models of mind: only computer programs could do this. Victory seemed assured ... ¹⁷

Seymour Papert's fairy tale, of course, ends with the holists triumphant, not a view that is very widely shared in the 'Al community' at present. As will become clear, my own view is that Papert's fairy-tale metaphor is as flawed as his memory/ intelligence metaphors. Neither fairy-tale sister is Cinderella - nor even Prince Charming; both modeling approaches are flawed if their intention is to provide structural metaphors for the way real brains work and real memories are stored. But it is worth looking a little more closely at the pretensions of both protagonists.

As Papert rightly describes it, one group of modelers, those I describe as reductionist, argued that the proper approach for Al was to take the brain and to endeavour to simulate some of its known properties using computers. The brain's units of function were assumed to be its nerve cells, neurons; and the brain was supposed to store, process and transform information through the functioning of networks of these neurons. The task was then to make mathematical models of how the neurons might function, assemble them into nets and test how different ways of connecting the cells within the nets might generate varying forms of output, including networks that could change their properties and output functions as a result of experience - that is, could both 'learn' and 'remember'. Such simulations were first performed by Frank Rosenblatt in the mid-1950s with a modeling system called a Perceptron. Perceptrons were triumphs of computing, but it soon became clear that they were very inadequate representations of real brain neurons. Although they could seemingly learn - that is, change their output properties in response to different inputs, so as, for example, to recognize and classify simple patterns - they failed at anything that was much more complex or even remotely resembled real-life problems.

In the 1960s and 1970s the intractable difficulties that the neuron-modeling approach had run into - and the theoretical limitations exposed by Papert and Minsky — led to its virtual abandonment. It was at this time, for example, that a UK government-sponsored study of the future of Al led to the conclusion that its promise had been much overstated, and the British research effort in the area was drastically scaled down. ¹⁸ Interest in its possibilities was dramatically revived, however, in the late 1980s, by a new and innovative approach. Earlier generations of computers were essentially serial processors - that is, they could carry out, albeit incredibly rapidly, only one operation at a time in sequence. Fast though they are, computers that carry out operations in this linear and sequential way face fixed limits to their speed of operation; messages after all cannot travel from one part of a computer to another faster than the speed of light, a limitation that has become known as the Von Neumann bottleneck. Perhaps it was this limitation, as the new generations of supercomputers pushed technology to the edge of the achievable, or perhaps it was the closer liaison between brain scientists and the computer modelers, that helped persuade Al enthusiasts that real brains don't work like this at all, but instead carry out many operations in parallel, and in a distributed manner, many parts of a network of cells being involved in any single function, and no single cell being uniquely involved in any. The speed limitation could be overcome if computers could be designed more like brains - that is, capable of parallel and distributed rather than sequential and linear operations.

The result has been an explosion of interest in new computer designs based on what are described as parallel distributed processing (PDP) principles, promising new generations of equipment which fascinate the military, industry and Al modelers in equal measure - though it is of course the first two who call the financial shots. As but one measure of the scale of this interest, in the late 1980s the European Community Research Directorate, claiming that Europe was lagging behind the US and Japan in the exploitation of these new systems, allocated 50 million ECU (about the equivalent of $50 million) for PDP-based research into neural modeling. When, in 1986, David Rumelhart, James McClelland and their colleagues at MIT published a large, two-volume book of papers on the potential of PDP for brain modeling, it is said to have sold 6,000 copies the day it appeared on the market. ¹⁹

The principles of the new modeling approach are known as connectionism. Like the earlier one, they are based on the idea that the brain is composed of ensembles of neurons with multiple connections between them. Appropriately connected sets of such cells can be made to show learning- and memory-like behaviour in that they will sort and classify inputs and slowly change their output properties in response to novel input patterns. But, unlike the earlier Perceptron type of model, the 'memory' does not reside in any one single cell or pair of connected cells within the network; rather it is a property of the network as a whole. Further, whereas in Perceptron-type models single units were called upon to receive direct inputs from the external world and modify their output properties accordingly, in the new connectionist models the networks are more complex, and include what the modelers call 'hidden layers' - arrays of 'cells' located between inputs and outputs. The difference in power that this change effects is dramatic.

The earlier generations of Al models were wired up almost as if the brain were a simple telephone switching system, with direct links between sense organs, like eyes and ears, and output organs, like muscles. They virtually ignored the fact that the vast majority of nerve cells within complex brains are not in direct communication with the outside world either by way of sensory input or motor output, but connect internally, receiving messages from and replying to other neurons; that is, there is a vast amount of internal processing of any messages that arrive in the brain, and also a great deal of private traffic between these cells (called interneurons), before any external responses are made. The hidden layers of PDF models are intended to serve almost as such interneurons, and massively increase the power of the networks to learn, to generalize, to predict.

Connectionist models are attractive to industry and the military because they promise to leap over earlier limitations on computing power. But they have also attracted a surge of enthusiasm amongst neurobiologists, many of whom believe that here at last is a model which comes close to what brains - or at least parts of brains -might actually be like. The last three years have seen a flurry of new research journals offering neural network models claiming to explain many aspects of brain processes; and the commanders and ideologues of this New Model Army seem in almost continual circuit of the globe, from one high-powered conference and seminar to another, scarcely having time to touch down in their own offices and labs to collect the latest simulation before getting airborne again.

Even philosophers have begun to listen; one of the more influential recent books amongst neurobiologists, who are not given to reading philosophy, has been the California-based Patricia Churchland's Neurophilosophy. ²⁰ Churchland's strategy in the book is to review long-standing problems in the philosophy of mind, match these with a competent review of contemporary neurobiological findings, and to conclude that reductionism rules, OK. Salvation, for her, comes from connectionism, and she followed the book with a couple of papers in the major journal Science, written jointly with San Diego neuroscientist Terrence Sejnowski, offering a prospectus for what they call a computational neuroscience, ²¹ a phrase which itself now forms the title of other books and journals. The fact that philosophers, modelers and neurobiologists are actually listening to one another, and that computer people have at last begun to show some respect for biological as well as artefactual brains, clearly makes their analyses an advance over the earlier ones, in which Al enthusiasts tended to run away with preconceived notions of what nerve cells did, and soon cut off all meaningful contact with the biological phenomena which the neurobiologists were studying. The neurobiologists' enthusiasm for Churchland stems, however, I believe, from the fact that, rather than challenging our assumptions, she shows us a rather uncritical respect. Her book thus serves, in a way unusual for a philosophy text, as a rather flattering reflecting mirror held up for us to shine in. ^* It is not merely that our warts are hidden in the reflection. Our very posture, in all its reductionist discomfort, has been given the Hollywood treatment. Yet the limitations of connectionist neurobiology and the philosophy it spawns are very clear, and I believe in the long run will fatally flaw them, for reasons which will become apparent as my account proceeds.

----

^* Nonetheless, a significant number of neurobiologists are uneasy with Churchland's reductionism. The issue came to a head a couple of years ago at one of those Swiss ski-meetings beloved of neuroscientists - meetings at which early morning and late afternoon scientific sessions are wrapped around the real business of the day, which is to get out onto the slopes for as long as daylight and stamina persist - part of the phenomenon perhaps best embraced under the well-known thesis known as the 'leisure of the theoried classes'. The theme of the meeting was 'the relationship between neuroanatomy and psychology', and Churchland was to give the opening talk. She set up her reductionist stall, arguing for the ultimate collapse of psychology into neuroanatomy, perhaps expecting an easy ride from a group of neurobiologists, and found to her surprise that it was strongly opposed by most present - especially the neuroanatomists!

----

Meanwhile, what of the second of Papert's two sisters, the 'artificial', holistic one? In this approach, no attempt was made to model brains; instead the emphasis was on modeling minds. That is, the modelers would try to identify what they believed to be the functions of mind processes, such as 'belief, hearing, vision, touch, inquiry, explaining, demanding, requesting ...' (I take this seemingly eclectic, but quite characteristic, set from Minsky's latest book The Society of Mind). ²² They would then endeavour to model the logic of these processes, irrespective of whether the models they produced in any way resembled real brains. What mattered was only that the models performed; that is, their outputs matched the expectations of the modelers about what human outputs might be if they were carrying out the functions they thought they were modeling.

To see the difference between these two approaches, take, for example, a person at a rifle stand in a fairground, endeavouring to shoot at a moving line of metal ducks at the back of the stand. The reductionist PDF approach would ask how the neural system might be connected so that the moving spatial image of the ducks might be conveyed via the retina to appropriate regions of the brain (hidden layers) and how these might then change their properties so as to learn to generate appropriate motor outputs. The holistic approach would instead ask: how can we construct a servo-mechanism which receives inputs as to the position and motion of the ducks and then directs output accordingly? Do the outputs of this mechanism resemble those in a real human carrying out the same task? If not, why not?

For the bulk of artificial intelligence's forty-year history, it is this latter approach that has been the most powerful. But, in developing it, its protagonists have moved away from how biological brains and psychological minds might work and instead concentrated on solving problems embedded in the silicon of computer chips and in mathematical logic - an approach which may produce bigger and better machines, but has become entirely indifferent to their relationship with the biological systems they were once attempting to model. Their rallying cry has been bluntly expressed by their most enthusiastic spokeswoman, Sussex University philosopher Margaret Boden: 'You don't need brains to be brainy'. ²³

The flawed metaphor

For all the sloganizing of top-down and bottom-up modelers, if the task is to understand how real brains and minds - or even memory - may work, I believe both approaches to be fundamentally flawed. Hence the failure of all the previous predictions about just when the Al people would come up with a really mind-like computer, regarded by the Wiener-enthusiasts of the early fifties as certain to arrive by the 1960s, then postponed to the seventies, eighties or even the bimillennium as time went by and models and programs from Perceptrons onwards came and vanished.

There have been three major recent critiques of the methodology and pretensions of Al, from, respectively, a philosopher, a mathematician and an immunologist. Let me describe them briefly before turning to my own problems with the computing metaphor.

First, the philosopher, John Searle, whose argument is based on standing the Turing test on its head. Imagine a non-Chinese speaker in a closed room, who receives questions written in Chinese through a machine. Available in the room is a code which enables the Chinese symbols to be matched against a second set which constitute replies to the questions being asked. The replies can be fed out through the machine to the outside world. For the observers outside, it is clear that questions in Chinese are being answered appropriately in Chinese; the person inside the room has passed a type of Turing test. But in no way could one infer that the person in the room understood, is conscious of, is intelligently responding to, the content of the messages being passed in and out; what is happening is no more than a purely automatic operation. This, says Searle, is what computers are doing, and why they cannot be regarded as intelligent or conscious. ²⁴

Second, the Oxford mathematician, Roger Penrose, whose recent book The Emperor's New Mind ²⁵ is presented as a sustained critique of connectionism. In essence, Penrose's point is quite straightforward: connectionism, if it is to work, depends on a relatively fixed and stable relationship of cells within a neural network, modified only in response to specific inputs and then responding in a deterministic way to this modified response. Penrose invokes both physical and mathematical theory against this claim: physical theory in the form of quantum mechanisms, which, he argues, result in a built-in indeterminacy in neural responses; and mathematics in the form of the highly fashionable chaos theory, which shows how indeterminate systems can nonetheless produce lawfully ordered outputs - just as, for instance, the random and indeterminate motions of gas molecules m a jar nonetheless together produce the precise and predictable relations between temperature, pressure and volume that are given in Boyle's simple gas law.

According to Penrose, then, a reductionist strategy must fail on two related grounds. In the first place, indeterminacy at the level of the neuron and its synaptic interconnections means that one will never be able to understand the mind or the brain simply by an analysis of its individual components, whose responses are inherently unpredictable. In the second place, however, this indeterminacy at the level of the component gives way to predictability at the level of the system. Consciousness, intelligence, memory thus emerge as properties of the brain as a system rather than those of individual components within that system.

The third critique of Al and its information-processing methodology is that of the Nobel Prizewinning Rockefeller University immunologist and theoretician Gerald Edelman. In a recent trilogy of books, ²⁶ Edelman has attempted the almost impossibly ambitious task of developing general theories of developmental biology, neural organization and consciousness based on a metaphor derived, not from physics or technology, but from evolutionary theory, and specifically from his understanding of Darwinian natural selection. Because in some respects I share Edelman's critique, though not his choice of the metaphor of selection, and a fuller exposition of his position requires more biology than I have yet introduced into this discussion, I will delay my engagement with it until the final chapter, and turn instead to my own unease with the metaphorical world of the computer-asbrain/mind/memory.

The analogy, while of potent fascination to many, has always been suspect amongst biologically grounded neuroscientists, on both structural and organizational grounds. Structurally, the properties of chips, AND/OR gates, logic circuits, or whatever, do not at all resemble those of neurons, if indeed it is neurons that are to be regarded as the relevant units of exchange within the nervous system. The units of which the computer is composed are determinate, with a small number of inputs and outputs, and the processes that they carry out with such impressive regularity are linear and error-free. They can store and transform information according to set rules. One consequence of this, for computer brain-modelers, has been the persistent tendency to attempt to reify certain types of process in which minds/brains participate. For instance the very concept of 'artificial intelligence' implies that intelligence is simply the property of the machine itself (I would argue that such a reification is equally inappropriate either for brains or for computers).

The brain/computer metaphor fails because the neuronal systems that comprise the brain, unlike a computer, are radically indeterminate. This critique of determmacy goes further than Penrose's, because I want to emphasize that brains and the organisms they inhabit, above all human brains and human beings, are not closed systems, like the molecules of a gas inside a sealed jar. Instead they are open systems, formed by their own past history and continually in interaction with the natural and social worlds outside, both changing them and being changed in their turn. This openness provides a further level of indeterminacy to the functioning of both brain and behaviour.

Unlike computers, brains are not error-free machines and they do not work in a linear mode - or even a mode simply reducible to a small number of hidden layers. Central nervous system neurons each have many thousands of inputs (synapses) of varying weights and origins (perhaps 10 14—10 15 in the human brain; that is getting on for a hundred thousand more connections in an individual brain than there are people alive on the earth today!). The brain shows a great deal of plasticity - that is, capacity to modify its structure, chemistry, physiology and output - in response to contingencies of development and experience, yet also manifests redundancy and an extraordinary resilience of functionally appropriate output despite injury and insult. Brains carry out linear computations relatively slowly yet can exercise judgemental functions with an extreme ease that baffles the modelers.

Consider a simple experiment in which a person is briefly shown a list of four digits, and asked to remember and recite them back. Most people can perform the task with ease. However, if the number of digits is extended to seven or eight, then most of us begin to fail at the task - especially if the length of time between seeing and recalling the numbers is stretched from a few minutes to an hour or more. On this basis, the maximum human memory for digit spans of this sort can be simply calculated in bits, as follows for an eight-digit string:

First, the bits for the numbers themselves

8 x log ₂ 10 = 8 X 3.32 = 26.56

Then, because the numbers have to be in the proper order, a calculation for order information is needed, and this is given as

log, 8! = 15.30

Giving a total of just 41.86 bits - and yet it would appear that we don't have the memory capacity to do it! Now compare this with the number of bits available in the memory of a simple pocket calculator - about 1,000. And the 10-centimetre double-density floppy disk currently inserted into my beloved Apple Mac, onto which I am typing these words, has a memory storage capacity of more than 10 million bits.

What does a calculation of bits like this actually mean? The Cambridge mathematician John Griffith once estimated ²⁷ that if humans were presented with, and stored, information at the rate of 1 bit a second for every moment of a seventy-year lifespan, at the end of their life they would have stored some 10¹⁴ bits - or about the equivalent of the information content of the Encyclopedia Britannica, that is 10,000 floppy disks and not so far off what the hard disk on my computer can manage. Impressive, for a micro, but it would seem surprisingly small for a functioning brain.

Can it really be that humans have no more memory capacity than a microcomputer? Clearly something is wrong somewhere with such calculations. What can it be? Here's a clue: although I cannot remember more than eight digits flashed onto a screen in front of me, I once demonstrated the power of human memory to an audience by flashing up a much longer, 48-digit list, turning my back to the screen and calling it off perfectly:

524719382793633521255440908653225141355600362629

How can I do this when failing the eight-digit test? Simply, the 48-digit list is not random, but contains a sequence of birthdays, telephone numbers and other codes that I have a regular need to use and therefore remember. But I don't remember them as numerical information in the error-free way that a computer would, and I don't store them in a unimodal number sequence. By contrast with computers, brain memory is error-full, and uses multiple different modalities. For me, unlike a computer or the Chinese translator in Searle's closed room, this particular unique number sequence has a meaning which is also unique for me. And it is on the basis of meaning, not sequence, that I am recalling them. Further, I would want to insist that meaning is not synonymous with information. Meaning implies a dynamic of interaction between myself and the digits; meaning is a process which is not reducible to a number of bits of information.

Another example. At dinner last night I was offered a menu with a multiple choice of dishes. I read the menu, chose and ate a meal, and can now tell you that it included broccoli soup and poached salmon. Information in the printed text of the menu was transformed into recollections of earlier tastes, then the spoken order to the waiter, and then the material reality of the food and its actual taste. And now when I tell you I ate broccoli soup and salmon I neither offer you the printed menu nor the food - still less do I expect you to taste it directly - instead I further modify last night's experience by translating it into spoken words. ²⁸ At each point in this sequence there has been more than just a switch in the modality in which the information is expressed: there has been an input of work on that information which has irreversibly transformed it (and this of course says nothing about the work that each listener or reader of this description does in subsequently further transforming, working on and interpreting the data I have just offered).

Thus brains do not work with information in the computer sense, but with meaning. And meaning is a historically and developmentally shaped process, expressed by individuals in interaction with their natural and social environment. Indeed, one of the problems of studying memory is precisely that it is a dialectical phenomenon. Because each time we remember, we in some senses do work on and transform our memories; they are not simply being called up from store and, once consulted, replaced unmodified. Our memories are recreated each time we remember. I will have much more to say about this work of re-creation, of remembering, in the final chapter of this book.

If my critique has so far been addressed to the reductionist, connectionist modelers, related arguments affect the holists. Consider the contrast between the relative ease with which programmers were able to develop chess-playing programs to Grand Master level and the difficulty in developing a robot system which can laboriously pile an orange pyramid onto a blue cube. Compare this with the ease with which an untrained human can toss an apple core into a waste basket five metres away, or, for that matter, play poker. Sure, one can devise a program that can calculate the odds of drawing a flush against three of a kind, but poker involves a type of competitive psychology, of bluff, and it requires the appreciation and assessment of non-cognitive inputs for which, I believe, there can be no effective machine analogy. Humans might derive some pleasure from playing chess against a program, none from betting against a computer in poker. Perhaps a poker test should replace the Turing or Searle test?

The holistic attempt to bypass the problem of the brain entirely, and to concentrate instead on modeling the mind, empties all real biological phenomena from psychology in its attempt to explain behaviour. This arid, almost scholastic tradition argues that, if one can identify appropriate mind properties and processes, then one can model these properties and processes in abstract thought-experiments or sets of mathematical symbols and subsequently incarnate them (or, perhaps better to say, 'inmachinate' them) into silicon components, optical switches or magnetic monopoles just as well as into the complex bits of carbon chemistry out of which evolutionary processes have generated real brains. ²⁹ Hence Boden's ludicrous aphorism 'you don't need brains to be brainy' -by which she means that you can model mind processes using whatever are the latest and most powerful computer systems without paying any attention to the underlying biology. All that is needed is machinery, or mathematical models of such machinery, which will respond to particular inputs with outputs which resemble those that brains might have. Such models work like machine-translation systems, which, given a sentence in French, can convert it to its English equivalent, even though the way a machine translates from one language to another in no way resembles how people may carry out a similar task.

This separation of mind from its actual material base is in some ways a reversion to the old Cartesian programme of brain/mind dualism. But, equally, the insistence on treating the brain as a sort of black box whose internal biological mechanisms and processes are irrelevant, and all that matters is to match inputs to outputs, is reminiscent of the behaviourist programme in psychology, about which there will be much more to say in Chapter 6. For the behaviourists, both mind and the brain were unnecessary concepts. Behaviour, of humans or other animals, was to be understood in its own terms, as simply a series of stimuli and responses linked in chains and adopted by the organism in response to rewards for approved and punishments for undesirable behaviour. If Cartesian dualism is thesis, and behaviourism antithesis, the programme of the holistic modelers epitomized by Boden's slogan is a sort of unholy Hegelian synthesis.

To unpack the implications of Boden's phrase, let us replace it for a moment with its equivalent 'You don't need two legs in order to locomote'. Well, surely this is true: one can move along the ground with four legs, like a horse, many legs, like a millipede, or none, like a snake or snail; or with wheels on a track like a railway train, or with freely moving wheels like a car, or with caterpillar tracks like a tank. One could travel on a cushion of air like a hovercraft or by magnetic levitation. One could, conceivably, fly at speed on a magic carpet. All these modes of locomotion, in terms of a function defined as getting from A to B, may be equivalent. But the principles they employ show radical - in some cases fundamental - differences. If we should be interested in the specific question of human walking - how it is achieved, how it is acquired, and how it may be impaired temporarily by over consumption of alcohol or lastingly by brain damage - studying fairy tales about flying carpets or the principles of the internal combustion engine, or even the wave forms of snake tracks, will be of only limited assistance. And it is just these specifics, however boringly trivial they seem to philosophers, that most brain scientists are interested in.

Lest this should seem an excessive polemic, consider the lists of mind processes which Minsky's book, to which I referred earlier, identifies as to be modelled. Minsky sees 'mind' as a 'society' of arbitrarily defined and hierarchically arranged 'agents', of 'memory', 'anger', 'sleep', 'demanding', 'believing', 'more', or what you-will. These labels are then given to a series of 'black boxes' (actually, black-outlined open ellipses and arrows on a page) connected up in seemingly random manner, out of which a theory of mind is supposed to emerge. This exercise seems a classic example of a top-down approach, far removed from the biological world which I - and I suspect most other people - inhabit. How am I to judge which, if any, of the multifarious, jostling and seemingly arbitrary members of Minsky's 'society of mind' have manifestations in identifiable brain processes?

Suppose I were to invent a wholly different listing, including 'spirituality, belief in mutant teenage ninja turtles, scepticism and an inability to distinguish a hamburger from the polystyrene package in which it is encased', how would I be able to decide whether the 'agents' in any human brain were more similar to Minsky's or to mine? The point is surely that it is possible to generate a strictly infinite number of models capable of being drawn as black boxes linked together by arrows which can pass some sort of theory test - for if the output doesn't seem quite right one can always redirect the arrows, add more or turn them from hard into dotted lines. In this make-believe world we can with impunity draw legs on snakes or fix roller skates to human feet for convenience so as to get the 'right' output. But the material, biological world imposes rather more rigorous reality tests upon empirically based science than this. Minsky's holistic models are precisely the type of evocative analogy neurobiologists do not need in our efforts to understand biologically-real brains and behaviour.

I am certainly not arguing that computer modeling and its metaphors are worthless. Making metaphors does seem to be a fundamental part of the scientific endeavour, and without such approaches, neurobiology cannot succeed in its task of understanding real ^* - that is, biological - brains and their functions, whilst for Al itself, progress depends on a proper grounding in such biology rather than the imposition of purely rational, cognitive, top-down models. But Al needs to know its place, which is never to reverse its metaphors by privileging the model over the biology, but instead to show some humility in confronting that marvellous object of its study, the brain.

^* I have lost count of the- number of times I have written the word 'real' in drafting this chapter, only, frequently, to edit it out on the grounds that the realist/social constructionist debate within both philosophy and the sociology of knowledge has become so intense that to use the term without hedging it round with a defensive thicket of qualifications is to run the risk of being charged with intellectual naiveté - or its wicked sibling, natural scientific arrogance towards other forms of understanding of the world. But I really don't want to get into this debate here. For those who want me to lay on the line precisely where I stand, I am a qualified realist in the historical relativist tradition; put briefly, I claim there is a material universe of which we can have- sound knowledge, albeit knowledge which is coloured by our historical and social locations, the current state of our technology and the framework within which we seek that knowledge.' ³⁰ That means that when I speak of 'real' brains I remain unashamedly prepared to defend their existence and my capacity to achieve objective knowledge about them and how they work against sociological or philosophical doubters. But not here, please; I have other fish to fry.

Memory, natural and artificial

At the beginning of this chapter, I described how Greek and Roman philosophers and rhetoricians contrasted two types of memory, natural and artificial. It was artificial memory that could be trained and analogized to writing on wax tablets, which led to the search for technological metaphor. Natural memory by contrast was a given, a human quality that required no explanation, merely recognition. Yet, as I have also argued, such is the powerful interaction of our technology with our very biology that the fact of creating a technology-driven society in which metaphors for memory have become central changes the very nature of our memory itself. The act of writing, as both Plato and Zairean bard recognized, fixes the fluid, dynamic memory of oral cultures into linear form. The establishment of mass-circulation printed texts, as opposed to hand-crafted and variable manuscripts, as Walter Ong points out, further stabilizes and controls memory, standardizing and collectivizing our understandings, creating

a sense of closure not only in literary works but also in analytic, philosophical and scientific works. With print came the catechism and the 'textbook', less discursive and less disputatious than most previous presentations ... Catechisms and textbooks presented 'facts' ... memorizable, flat statements ... the memorable statements of oral cultures tended ... to [present] not 'facts'; but rather reflections ... ³¹

Modern technologies - photography, film, video and audiotape, and above all the computer - restructure consciousness and memory even more profoundly, imposing new orders upon our understanding of and actions upon the world. On the one hand, such technologies freeze memories with all the rigidity of old Victorian sepia family portraits, providing an exoskeleton which prevents them from maturing and transforming themselves as they would do if untrammeled and without constant external cues within our own internal memory systems. On the other, they dissolve the barriers between fact and fiction in quite subversive ways. Think of the TV penchant for docudramas, or the Woody Allen movie in which our hero suddenly inserts himself within a well-known sequence of Hitler addressing his supporters at a Nuremberg rally.

As I key these very words into my computer and they appear on the screen in front of me I am acutely conscious of these contradictions. Once, when I drafted a chapter, it would initially be laboriously by long-hand; I would edit my text and type it, thereby fixing it more or less permanently, if only because the labour of shifting and reordering material was then too great to contemplate for all but the gravest of reasons. Now all is fluid; this chapter, which set out to be written sequentially, has grown, not linearly, but throughout its entire body as draft sentences which once would be fixed in place here suddenly take wing and fly to other sections entirely. My memory for a once-planned sequencing of material is subverted by the liberatory power of the technology and refuses to be confined by the discipline I endeavoured to impose upon it. The Platonic, Ciceronian distinction between natural and artificial can no longer be maintained. If our study of memory must reject the computer model and metaphor in favour of the biological and 'real', we must equally recognize that the very nature and mechanisms of memory are being transformed by the technology whose explanatory power we are rejecting.

Perhaps this helps explain the extraordinary concern over the nature of memory that characterizes art and literature. Jane Austen well describes this fascination when she has the stoic heroine of Mansfield Park, Fanny Price, say:

If any one faculty of our nature may be called more wonderful than the rest, I do think it is memory. There seems something more speakingly incomprehensible in the powers, the failures, the inequalities of memory, than in any other of our intelligences. The memory is sometimes so retentive, so serviceable, so obedient - at others, so bewildered and so weak - and at others again, so tyrannical, so beyond control! - We are to be sure a miracle every way - but our powers of recollecting and of forgetting, do seem peculiarly past finding out. ³²

Writers have long tried to capture this peculiar power. Yet where the order and linearity of sequential memory characterize the nineteenth-century novel, as the twentieth dawned, temporal order disintegrated. For Marcel Proust, whose fifteen-volume A la Recherche du Temps Perdu is a long-drawn-out struggle to recall and thus to transcend a painful past, the trigger which evokes the entire history is the taste of a madeleine cake. For contemporary writers the issue is even more problematic. In the Canadian writer Margaret Atwood's novel Cat's Eye, a complex, painful and fragmentarily recalled childhood is grasped entire only when, almost at the end of the book, the heroine rediscovers and holds in her hand an emblem of her childhood, a mysteriously coloured cat's-eye marble. Janet Frame's autobiographical novels are described by a reviewer as a 'meditation on the deceptive layers of memory: "where in my earlier years time had been horizontal, progressive, day after day, year after year, with memories being a true personal history"', as time progresses, order and linearity disintegrate. ³³

For the converse, the excess of memory, we can turn to one of the more extraordinary short stones of that haunting Argentinian word-magician Jorge Luis Borges, a tale of a young man, Funes, who, in a manner not dissimilar to Simonides, seemed to be able to remember everything:

We in a glance perceive three wine glasses on the table; Funes saw all the shoots, clusters and grapes on the vine. He remembered the shapes of the clouds in the south at dawn on the 30th of April of 1882, and he could compare them in his recollection with the marbled grain in the design of a leather-bound book which he had seen only once, and with the lines in the spray which an oar raised in the Rio Negro on the eve of the battle of the Quebracho ... These recollections were not simple; each visual image was linked to muscular sensations, thermal sensations, etc. He could reconstruct all his dreams, all his fancies. Two or three times he had reconstructed an entire day. He told me: / have more memories in myself alone than all men have had since the world was a world. And again: my dreams are like your vigils ... my memory, sir, is like a garbage disposal ... A circumference on a blackboard, a rectangular triangle, a rhomb, are forms we can fully intuit; the same held true with [Funes] for the tempestuous mane of stallion, a herd of cattle in a pass, the ever-changing flame or the innumerable ash, the many faces of a dead man during the course of a protracted wake. He could perceive I do not know how many stars in the sky ... ³⁴

It is no accident that, in the story, Funes dies young - of an overdose of memory, so to say.

The problem of the freezing and fixation of artificial memory is even greater when we move from the individual to the collective in memory. Is it possible to create a space in which we can both assimilate into our own experience the meaning of the ever-screaming, ever-napalm-burnt child on our television screens without simply freeze-framing it, fixing it for ever and thus losing the dynamic of real, biological memory? Such a fixation of images produces a peculiar and novel form of artificial memory, of the sort of which Plato was particularly distrustful. What were once individual experiences, to be made and remade in our imagination and memory, like my own recollections of a birthday party, or the recovery of a long-lost Mr. Goss from some recess of the brain, are now public. They are memories which we all share as part of collective experience, even generations not yet born at the time of the Vietnam war or the Nazi death camps. This makes them peculiarly powerful as a way of providing social coherence. They have become part of our shared history. But equally, we can no longer make and remake them in our own minds, assimilate them fully into our lived experience and consciousness, because they are for ever fixed by the video. Further, the power of the camera and the film-maker allows history - that is, collective memory - to be remade at the behest of a Big Brother with a memory hole, a revisionist holocaust historian, or a minister of education with a belief in the importance of not forgetting 1066.

The new technologies offer unrivalled prospects for, on the one hand, artificial memory, and on the other, the production of completely fabricated memories of the Woody Allen type, or even a sort of social amnesia, the public erasure in which the Stalinist airbrushes remove Trotsky from the photographs of the makers of the Bolshevik revolution. As I write these words the Soviet state and communist party are melting away before the very eyes of an astonished world in images equally forever fixed; demonstrators pulling a tank-driver from his turret, Yeltsin's accusing finger pointing to the list of conspirators Gorbachev is compelled to read. As minor communist party officials now scurry to rewrite their own parts in history, we may also expect social amnesia on a massive scale.

Small wonder that social movements find a continued need to rescue their own memories - and that they are as continually traduced - as Margaret Atwood found out when her novel The Handmaid's Tale, 35 an attempt to create a feminist history of the immediate future through the memories of a youngish woman in a masculine, fundamentalist-Christian-dominated world, was turned into a big-budget movie which sold itself by eliminating memory and flaunting just that sexual availability Atwood's novel had criticized.

The collective in memory transcends the methods and models of the neurobiologist just as much as it does the metaphors of the computer, and will inevitably elude my attempts to grasp it here. I can go no further yet, but must return to the relatively safer shores of individual brain memory, and await the final chapter of the book to attempt to approach the collective as well as the individual. My conclusion from this chapter must echo the words of Fanny Price, only rejecting her view that the mechanisms of memory are 'peculiarly past finding out', for it is here that I wish to stake the claim for neuroscience. The matter cannot be left to the novelists any more than to the modelers, though the former certainly make a better fist of it, and are easier to read. Our task is to show that, just as there is more to memory than a wax tablet, a willing office boy or an ensemble of artificial neurons, there is also more than just the taste of a madeleine or the feel of a cat's-eye. As the psychologist Dalbir Bindra put it:

Psychologists have for too long tried to escape the reality of the brain in favour of physical, chemical, literary, linguistic, mathematical and computer metaphors. Now they must face the brain. ³⁶

 

References

1. Ong, W Orality and Literacy, Methuen, 1982, p. 146.

2. Cicero, De Oratore, II, Ixxxvi, 351-4, quoted in Yates, F The Art of Memory, Penguin, 1966, pp. 17-18.

3. Quoted by Patten B M The history of memory arts. Neurology 40,346-52, 1990.

4. Yates, op. cit.

5. Spence, J D The Memory Palace of Matteo Ricci, Viking, 1985.

6. Bolter, D Turing's Man, University of North Carolina Press, 1984.

7. Rose, H A, and Rose, S P R Science and Society, Penguin, 1969.

8. Wall, P D, and Safran, J Artefactual intelligence, in The Limits to Science, ed. Rose, S P R, and Appignanesi, L. Blackwell, 1986.

9. Many of the references to this debate will be found in the two books edited by myself and Hilary Rose: Rose, H A, and Rose, S P R The Political Economy of Science and The Radicalisation of Science, both Macmillan, 1976.

10. Needham, J Science and Civilisation in China, Cambridge University Press (continuing series of volumes).

11. Rose, S P R Molecules and Minds, Open University Press, 1987.

12. Descartes, R Philosophical Works, 1:116.

13. See, e.g., Westcott, E A Century of Vivisection and Anti-vivisection, Daniel, Ashingdon, 1947; Lansbury, C The Old Brown Dog: women workers and vivisection in Edwardian England, University of Wisconsin Press, Madison, 1985; Rose, H A Gendered reflections on the laboratory in medicine, in Hand, Brain and Heart, Polity, Oxford, forthcoming.

14. Descartes, R Les Passions de I'ame, 1649, quoted in Dudai, Y The Neurobiology of Memory, Oxford University Press, 1989.

15. From the reference to Brain in The New Book of Knowledge, Vol II BIB-CHIC. I was brought up on this encyclopedia.

16. Hodges, A Alan Turing: The enigma of intelligence, Counterpoint, 1983.

17. Papert, S One Al or Many, in The Artificial Intelligence Debate, ed. Graubard, S R, MIT Press, 1988, pp. 3-4.

18. Lighthill, J Artificial Intelligence, Science Research Council, 1973.

19. Rumelhart, D E, McClelland, J L, and the PDP Research Group Parallel Distributed Processing, MIT Press, 1986, 2 vols.

20. Churchland, P S Neurophilosophy: Towards a unified science of the mind-brain, MIT Press, 1986.

21. Churchland P S, and Sejnowski, T J Perspectives on computational neuroscience. Science 242, 741-5, 1988.

22. Minsky, M The Society of Mind, Heinemann, 1987.

23. Boden, M in The Limits of Science, ed. Rose, S P R, and Appignanesi, L, Blackwell, 1986.

24. Searle, J R Minds, Brains and Science, Harvard University Press, 1984.

25. Penrose, R The Emperor's New Mind, Oxford University Press, 1990.

26. Edelman, G Neural Darwinism: The theory of neuronal group selection; Topobiology; and The Remembered Present, Basic Books, 1987, 1988,1989.

27. Griffith, J S A View of the Brain, Oxford, 1967.

28. I owe this example to Von Foerster, H What is memory that it may have hindsight and foresight as well? Atorga, Feb-April, Comm 122-4, 1969.

29. All these examples of memory metaphors occur in the current model-builders' lexicons. See for instance Delacour, J, and Levy, J C S (eds) Systems with Learning and Memory Abilities, Elsevier, 1988.

30. See the references to this debate in Chapter 1.

31. Ong, op. cit., p. 134.

32. Austen,] Mansfield Park, 1814, Chapter 22.

33. Frame, J An Autobiography, The Women's Press, 1990.

34. Borges, J L Funes the Memorious, in Fictions, Calder, 1965.

35. Atwood, M The Handmaid's Tale, Virago, 1987.

36. Bindra, D Metaphors, computers and the brain. Annual meeting of the Royal Society of Canada, mimeo, 1980.

 

 

Excerpt from:
The Making of Memory
-From molecules to mind-
By Steven Rose © 1992