0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)
67 Ansichten38 Seiten
This paper gathers together images and written material from a variety of published and unpublished sources to set both the man and his work in a richer context than has been possible for many decades. Grey Walter emerged as a fascinating and far-seeing -gure whose work has not dated, perhaps because it was explicitly based on similar principles to those used today in biologically inspired robotics.
This paper gathers together images and written material from a variety of published and unpublished sources to set both the man and his work in a richer context than has been possible for many decades. Grey Walter emerged as a fascinating and far-seeing -gure whose work has not dated, perhaps because it was explicitly based on similar principles to those used today in biologically inspired robotics.
This paper gathers together images and written material from a variety of published and unpublished sources to set both the man and his work in a richer context than has been possible for many decades. Grey Walter emerged as a fascinating and far-seeing -gure whose work has not dated, perhaps because it was explicitly based on similar principles to those used today in biologically inspired robotics.
1260 , 2085-2121 361 2003 Phil. Trans. R. Soc. Lond. A
Grey Walter Exploration and high adventure: the legacy of
Email alerting service
here article or click - sign up in the box at the top right-hand corner of the Receive free email alerts when new articles cite this article http://rsta.royalsocietypublishing.org/subscriptions go to: Phil. Trans. R. Soc. Lond. A To subscribe to This journal is 2003 The Royal Society on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 10.1098/rsta.2003.1260 Exploration and high adventure: the legacy of Grey Walter By Owen Holland Department of Computer Science, University of Essex, Wivenhoe Park, Colchester CO4 3SQ, UK (owen@essex.ac.uk) Published online 20 August 2003 Although many people feel themselves to be familiar with Grey Walters ideas, and with the design principles of his tortoises, most of the published sources are lacking in detail. This paper gathers together images and written material from a variety of published and unpublished sources to set both the man and his work in a richer con- text than has been possible for many decades. Grey Walter emerges as a fascinating and far-seeing gure whose work has not dated, perhaps because it was explicitly based on similar principles to those used today in biologically inspired robotics. Keywords: cybernetics; biological inspiration; robotics 1. Introduction There can be no doubt that some of the ideas pioneered by W. Grey Walter (1910{ 1977) have inuenced the eld of biologically inspired robotics down to the present day. The other papers in this issue show how the power of the approach is revealed by both the present state of the art, and the promise of future developments. This paper looks backwards rather than forwards; it is an examination of Grey Walters original work in the context of his life and times, and uses a selection of contemporary images and documents to place his achievements against a richer background than has previously been possible. Even a few years ago, little was known of his robotics work beyond what had been published in two papers in Scientic American (Walter 1950a, 1951) and in his book The living brain (Walter 1953a). However, the recent revival of interest in his work and in the early years of the cybernetic movement in Britain has led to the discovery of some fascinating and illuminating material, and it is probably only in the last few years that we have been in a position to make an informed assessment of his true contribution. Some of the material presented in this paper has already been published elsewhere; it is included here in order to give a full and rounded view of Walters work in the context of the theme of this collection. Perhaps the best approach is to begin with a brief biography, illustrated with some of the recently uncovered material. (For a more detailed account of his life and career, see Hayward (2001b).) William Grey Walter, always known as `Grey, was born in 1910, in Kansas City, the only child of Karl Walter (a British journalist, then editor of the Kansas City Star) and Margaret Hardy, an American journalist. The family One contribution of 16 to a Theme `Biologically inspired robotics. Phil. Trans. R. Soc. Lond. A (2003) 361, 2085{2121 2085 c 2003 The Royal Society on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2086 O. Holland moved to Britain during World War I. At Westminster School, Grey Walter was interested in classics, and later in science; he went on to Kings College, Cambridge, obtaining a rst in physiology in the Second Part of the Natural Sciences Tripos in 1931. He stayed at Kings for a further four years, carrying out research into nerve physiology and conditioned reexes; several informative vignettes from this period appear in The living brain. He was also very active socially, being elected to the Conversazione Society (the Cambridge `Apostles) in 1933. His elder son, the late Nicolas Walter, later wrote: [He] remained an active member for the rest of his life. He was associated with several of those later exposed or suspected as Russian spies, and was accordingly investigated by the security service during the 1960s. He was indeed a Communist fellow-traveller before and during the war, and he maintained many contacts in the Soviet Union, but after the war he actually became more sympathetic to anarchism. Walter (1990) Unfortunately, Grey Walters postgraduate dissertation, on `Conduction in nerve and muscle, did not win him the fellowship at Kings that he had hoped for. Accord- ing to Nicolas Walter (personal communication) this was an enormous disappoint- ment to him, and one he felt keenly for the rest of his life, in spite of his later successes. He left Cambridge on taking his MA in 1935, and was invited by Fred- erick Golla to join his research team at the Central Pathological Laboratory of the Maudsley Hospital in London, as described in The living brain. This immediately put him at the centre of UK research into the EEG (electroencephalogram), the eld in which he was soon to become a world leader. In 1939 he moved with Golla to the newly established Burden Neurological Institute (BNI), where he held the position of Director of Physiology, and where he worked for the rest of his career (Cooper & Bird 1989). Walters discoveries in the techniques and clinical applications of electroen- cephalography were numerous and important, and have been summarized in many publications. As his colleague and obituarist, Harold Shipton, noted, `Few scientic disciplines owe as much to one man as electroencephalography and clinical neuro- physiology owe to Grey (Shipton 1977). However, an apparently unpublished type- script from the archives of the BNI gives a unique insight into how he himself viewed his achievements. The document, dated 1967, is entitled `William Grey Walter, his Trumpet Voluntary. (Solo passages underlined.). There is not the space to reproduce the whole of it, but the excerpts below give a good distillation of the EEG-related ndings he reported in over 150 papers, as well as giving us some idea of his technical and conceptual versatility: It was Golla at the Maudsley who encouraged me to look through the apparatus at real problems and from 1935 to 1939 as a Rockefeller fellow I found myself almost forced to gather the ripe fruit that fell from the trees as soon as I had developed the right way of shaking them. . . . [It] was a very messy business to start with, but I did nally recognise the slow rhythms around tumours as something distinctive and the name I gave them|delta rhythms|was my rst enduring contribution to neuro- physiology. . . . During this period also I discovered similar abnormalities Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2087 in epileptics between seizures. As well as these studies of brain disease I tried to make some sense of the variations in alpha rhythms and in order to quantify frequency measurements I invented and built the rst on-line frequency analyser. . . . The operational proof of this instrument is that with it I discovered the complexity of the ne structure of alpha rhythms and demonstrated their relation to individual dierences in mentality and imagery. I also showed that (it was wrong to speak of ) `driving the alpha rhythms with icker|the alpha components remained in the spectrum and the icker-evoked ones added to them. This was an important discov- ery. . . it demonstrated the widespread eects of visual stimulation, from which emerged the discovery of photic activation in epilepsy and the con- cept of certain epileptic seizures as due to coupling or synchronisation of previously uncoupled systems. The identication of quasi-harmonic com- ponents in the inter-seizure EEG of epileptics (by frequency analysis) was the clue to this concept. At this period also (1943{6) I discovered and named theta rhythms. . . and this term also has been accepted, although sometimes I think mis- used. . . . [In] some of the records (of experiments on conditioned reexes) I noticed a slow potential change in certain conditions when the subject was severely penalised for a wrong decision. . . . I began to look for this slow change more carefully. . . . It took a long time, several hundred long polygraphic records and innumerable control studies to establish beyond any doubt that in all normal human brains the frontal cortex develops a potential change of about 20 microvolts following a signal (sensory or internal) which is signicantly related to the expectancy of a subsequent event with which the person is in some way engaged. This is the Contin- gent Negative Variation (CNV) or Expectancy Wave which is now being studied intensively in many laboratories all over the world. . . I have not referred to the various technical innovations and adaptations that I have been involved in during the last 20 years. These were original and productive but were joint eorts with my friends and colleagues. . . . I have been very fortunate in being at the right place at the right time with the right tools and the right people. I do claim, however, to have helped to organise these coincidences. I founded the EEG Society in 1943, organised the rst EEG Congress in 1947, and. . . the International Federation of EEG Societies (of which I was President from 1953{57) and the EEG Journal in that year. . . I claim also that I have never gone seriously wrong in my descriptions, interpretations, inventions or conjec- tures. There are thousands of people working now as I was alone 30 years ago, and I believe that in 10 years they will be working along the lines that I am exploring now. Walter (1967) The work that mainly concerns us here|the design of the tortoises, and the inter- pretation of their behaviour|was carried out in 1948 and 1949. Although it looms large in our view, it is important to realize that it formed a very small part of his intellectual output: only a single chapter in The living brain, and a handful of papers Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2088 O. Holland amongst the 174 recorded under his authorship in the BNI bibliography. This does not mean that he spent only a little time on the tortoises, or considered them unim- portant; the fact is that he was an extremely active person in very many ways, and had a high work rate and an unusually wide range of interests. Nicolas Walter described some aspects of his lifestyle and character as follows: He was. . . involved in the peace movement before and after the war, being a member of the Peace Pledge Union in the 1930s and of the Bristol Com- mittee of 100 in the 1960s; but he was never a pacist, and during the war he was an enthusiastic ocer in the Bristol Home Guard. He was a convinced trade unionist, and was a leading member of the Bristol branch of the Association of Scientic Workers. He was a rm atheist, and frequently addressed humanist meetings. . . . He was interested. . . in the paranormal, and he did some research on the neurological aspects of altered states of consciousness and participated in conferences of the Parapsychology Foundation. He was a uent speaker and writer in sev- eral languages, on general as well as technical subjects. . . . He relished making broadcasts and giving talks; he was a frequent guest on the BBC television Brains Trust during the 1950s and 1960s. . . . He also wrote many articles; but he found it dicult to produce more sustained work, and both of his two books were actually written by his father from his notes and conversations.y He always looked much younger than his age, and he kept t at tennis, gliding, and skin-diving. He grew a beard in 1948, when this was still very unfashionable, and he was fond of behaving unconventionally and of wearing either very casual or very formal clothes. Walter (1990) His disregard of conventionality extended into his private life, which was often scandalous by the ordinary standards of the day, though no more so than that of many in his circle. Did his lifestyle and character aect his career in any way? There are certainly grounds for suspecting that they did, although he always received due recognition for his achievements from the EEG community. Hayward (2001b) exam- ined Walters career in the context of the history of science, and noted that, in con- trast to `the reserved and stately gures of Adrian and Sherrington, who dominated neurophysiology, `Walter cultivated a more swashbuckling image as an emotional adventurer (Hayward 2001b, p. 616). After listing some of Walters personal and professional characteristics, he remarked, Such eclectic and amboyant combinations. . . certainly estranged many of his professional colleagues. . . . Despite his leading role in the clinical and commercial development of the EEG during World War II, Walters maverick reputation ensured his continuing alienation from the scientic establishment. Hayward (2001b, p. 616) y The 1961 Penguin edition of his rst book, The living brain, carries the dedication: `To my father, with whom this book was happily written. His second book (Walter 1956a) was a science ction novel, published in the UK as Further outlook and in the US as The curve of the snowake. Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2089 Figure 1. One of the earliest photographs of Elsie, probably dating from late 1949. The notice reads `Please do not feed these machines. Some of his inuential contemporaries may have regarded him as `unsound; others may have objected to his `showmanship, and to what Shipton called his `. . . immense talent for persuasive oratory. . . (Shipton 1977), which on occasion may have carried him beyond the facts. It is possible that the continuing ow of his scientic con- tributions might have brought some eventual ocial acknowledgement, but it was cut short by a tragic accident: as he travelled home on his motor scooter one day in 1970, he collided with a runaway horse, and received serious head injuries. After weeks in a coma, he regained consciousness; he eventually made a good physical recovery, but it soon became clear that he was no longer able to work eectively, and that his research career was at an end. Characteristically, he had never saved for his retirement, but a small pension was arranged for him, and he nally retired in 1975. He died following a heart attack in 1977. 2. The tortoises and the public It is now impossible to establish exactly when Walter began work on the tortoises, but it was probably at some time in 1948. In a letter, Nicolas Walter recalled `. . . I stayed with Grey and Viviany for the last time during the school holidays in spring 1948, and although he talked a lot about his work he said nothing about making models, so y Vivian Dovey was Grey Walters second wife. A colleague for many years, she was co-author with him of eight papers on electroencephalography between 1944 and 1957. Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2090 O. Holland he must have started after that (Walter 1995). However, by late 1949, Grey Walter was demonstrating Elmer and Elsie, the rst two tortoises, to the press, with all the showmanship that some held against him. Figure 1, one of the earliest photographs from this period, is an excellent example of his style. Fortunately, the rst major press story, by Chapman Pincher in the Daily Express of 13 December 1949, is a model of restrained and factual popular reporting, giving reasonably accurate technical details of the robots construction and behavioural repertoire. It seems likely that the background information in the article is also correct: Elmer, the `prototype, is reported to have been built `more than a year ago, indicating a date of 1948; the robots were built in the `backroom laboratory of his house by Grey Walter, `helped by his wife Vivian. Other newspaper accounts were less accurate and more sensational: Toys which feed themselves, sleep, think, walk, and do tricks like a domes- tic animal may go into Tommys Christmas stocking in 1950, said brain specialist Dr. Grey Walter in Bristol last night. For two years he has been experimenting with toys containing an electric brain. . . . Dr. Walter said the toys possess the senses of sight, hunger, touch, and memory. They can walk about the room avoiding obstacles, stroll round the garden, climb stairs, and feed themselves by automatically recharging six-volt accumulators from the light in the room. And they can dance a jig, go to sleep when tired, and give an electric shock if disturbed when they are not playful. . . . Dr. Walter said: `There is no other machine like it in the world. This hobby of making toys with brains is proving of great value in the study of the human brain. The toy has only two cells. The human brain has ten thousand million. `But, said Dr. Walter, `most people get along with using as few as possible. Daily Mail, 17 November 1949 Walter had excellent contacts in the BBC, and in 1950 the tortoises were the subject of a BBC television newsreel lm `Bristols robot tortoises have minds of their own. The commentary (transcribed) was not calculated to appeal to his scientic peers: In a modest villa on the outskirts of Bristol lives Dr. Grey Walter, a neu- rologist, who makes robots as a hobby. They are small, and he doesnt dress them up to look like men|he calls them tortoises. And so cun- ningly have their insides been designed that they respond to the stimuli of light and touch in a completely life-like manner. This model is named Elsie, and she sees out of a photo-electric cell which rotates about her body. When light strikes the cell, driving and steering mechanisms send her hurrying towards it. If she brushes against any objects in her path, contacts are operated which turn the steering away, and so automatically she takes avoiding action. Mrs. Walters pet is Elmer, Elsies brother, in the darker vest. He works in exactly the same way. Dr. Walter says that his electronic toys work exactly as though they have a simple two- cell nervous system, and that, with more cells, they would be able to do many more tricks. Already Elsie has one up on Elmer: when her batteries Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2091 begin to fail, she automatically runs home to her kennel for charging up, and in consequence can lead a much gayer life. The newsreel, which survives, is probably the earliest (and possibly the only) existing lmed record of Elmer and Elsie. It is valuable in many ways. Shot in Walters lounge, it shows the tortoises operating in the environment in which they were developed, and gives many clues about their speed and sensory abilities. The opening scene shows Walter in his workshop|a small study with a workbench and a rack of tools. Elsie, her cover removed, is on the bench, and he is ostensibly making some adjustments. There is a good close-up of the operation of the worm drive to the steering apparatus. He replaces Elsies shell, not without some diculty, and carries her through to the lounge, where he puts her down on what appears to be a polished wooden oor. Behind her is the hutch, or kennel, containing an immobile Elmer. Walter places a small waste bin in front of Elsie, and switches her on, with the clear intention that the avoidance behaviour should be triggered when she hits the waste bin. She begins to move in the typical cycloidal path, with her photocell rotating anticlockwise and scanning the environment, but is very soon attracted towards a source of light to the left of the camera. (The position of the light can be deduced not just from Elsies behaviour but from the reections and shadows in the room.) She hits the waste bin, pushing it aside, but close examination of the lm shows no sign of any behavioural change; the obstacle may well have been too light to actuate the switch mechanism. Walter moves the bin in front of Elsie again, but no behavioural change can be seen when the second collision occurs. Over a shot of a smiling Vivian Walter, the commentary introduces Elmer as `Mrs Walters pet, and Elmer and Elsie are seen moving closer together and eventually colliding several times. In this part of the lm, Elsie is moving twice as fast as in the previous part|the lm speed has obviously been doubled, presumably to sustain the viewers interest. Elmer moves at about the same speed as Elsie, but his photocell scans in the opposite direction, a fact unrecorded in any of Walters writings. After the fourth collision between the robots, Elsie stops moving for several seconds; this must be a malfunction, as the electronic circuitry should not support any such behavioural state. The lm then cuts to another sequence in which both robots are moving freely without colliding. The nal shot shows Elsie moving towards the hutch in a straight line, with the commentary implying that this is how she `runs home to her kennel for charging up. The angle of the shot gives a good view of the charging system at the back of the hutch. The lm ends with Elsie apparently coming to rest in the hutch, but it is not clear whether she has really stopped, or whether the last few frames of lm have been repeated. What is very clear, however, is that the light that is supposed to attract Elsie into the hutch is not lit, and so we are forced to the conclusion that the shot must have been staged, with the steering having been centred and the steering motor disabled so that Elsie would steer a straight course after being lined up with the hutch. Although these observations to some extent devalue the lm as a record of the tortoises performance, they give a useful insight into the unreliability and variability of the machines, and perhaps give some support to the concerns of Walters detractors. Helped by high levels of exposure through more popular post-war media such as illustrated magazines (e.g. Morley 1950; Walter 1950b), the tortoises rapidly became famous, not just in the UK, but worldwide, and Walter received many requests for Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2092 O. Holland Figure 2. Grey Walters `personal tortoise from the 1951 batch, now on display in the Science Museum, London. demonstrations. Their presence was also demanded at the 1951 Festival of Britain, the futuristic celebration of Britains recovery from the war years. Both activities presented problems: Elmer and Elsie had been cobbled together from wartime surplus components and scrap materials|for example, their gearwheels were salvaged from old gas meters and alarm clocks|and were in consequence chronically unreliable. The situation was saved when six more tortoises of an improved electrical and mechanical design were built for Grey Walter in early 1951 by Mr W. J. `Bunny Warren, one of the talented BNI engineers recruited by Walter after the war. Elmer and Elsie appear to have been scrapped at about this time. Of the six 1951 tortoises, two are known to have survived. One, the tortoise used by Walter for most of his demonstrations after 1951, went on display in the London Science Museum in 2000 (see gure 2), and the other has been on show in the MIT Museum since 2001. Although Walter soon moved on to build other models, the tortoises managed to retain a high public prole for several years. Some of this was undoubtedly due to Walters continuing showmanship|for example, in 1955 he released them amongst the audience at a British Association meeting (George 1956)|but much was due to the success in 1953 of The living brain, which went into paperback in 1961, and was reprinted in 1963 and again as late as 1968. A rather dierent view of the tortoises was provided by Pierre de Latils La pensee articielle (de Latil 1953) later published in English as Thinking by machine (de Latil 1956). It is an account, an analysis, and even an attempt at a synthesis of the cybernetic movement, particularly as it had developed in England and France. Grey Walters tortoises are central to it. It contains an extended report of an interview with Grey Walter that can be dated by its context to late 1950. Some of the technical information is not found in any other published record, and is clear and accurate, but some of de Latils purported eye-witness descriptions of the behaviour of the tortoises do not ring true, and were probably glossed from information provided by Walter.y Nevertheless, it provided y One particularly puzzling aspect is that de Latil appears to think that it is possible for the tortoises to nd a static equilibrium at some intermediate light level, and to rest there without moving, whereas their electronic and mechanical design should always produce movement of one sort or another. He notes Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2093 an independent and quasi-scholarly endorsement of the scientic signicance of the tortoises, along with more popular material: a photograph of the Walters watching their nine-month-old child Timothy playing with a tortoise is captioned, `In their country home near Bristol, these parents have two children: one is electronic. Vivian Dovey and Grey Walter have two ospring: Timothy, a human baby and Elsie, the tortoise, of coils and electronic valves (de Latil 1956). 3. Science and the tortoises The rst scientic publication dealing with the tortoises is the Scientic American paper of May 1950, entitled `An imitation of life (Walter 1950a). Although the paper contains eight stylized sketches illustrating the tortoises behaviour (drawn by the artist Bernarda Bryson Shahn), it gives only the briefest textual description of the constructional and electronic details of the robots, and it is not possible to work out exactly how each type of behaviour is generated. A follow-up paper in Scientic American in August 1951, entitled `A machine that learns (Walter 1951), discusses the addition to a tortoise of a hardware learning mechanism (the conditioned reex analogue (CORA)). It includes a time exposure photograph of a tortoise with a torch mounted on its shell nding and entering the brightly lit hutch; the streak of light left by the torch enables the robots trajectory to be seen. However, the major contemporary published source of information on the tortoises is The living brain (Walter 1953a). Until recently, these three publications were all that were generally available by Walter, but the discovery and release of other contemporary materials and artefacts has lled in many gaps left by these original sources (e.g. Holland 1996, 2003). There appear to have been three distinct sources of scientic inspiration for the tortoises: the presence in post-war Britain of a well-developed cybernetic movement largely independent of the better-known American strand associated with Wiener; Grey Walters principled interest in building physical working models to test hypothe- ses; and his theories about brain function. Each of these will be examined in turn. (a) The British cybernetic movement It is not generally realized that the development of cybernetics around Wiener in the US was paralleled by an independent stream of similar ideas in the UK (de Latil 1956; Cordeschi 1987; Hayward 2001a; Pickering 2002). The growth of the Wiener school was essentially centred around his 1948 book Cybernetics (Wiener 1948). In the UK, things were rather dierent; the driving force was a London-based dining club, the Ratio Club (Hayward 2001a). It was founded by the neurologist John Bates in 1949, and met regularly for several years, nally being dissolved in 1958. Membership was by invitation only, and Batess original idea, as set out in a letter to Grey Walter (Bates 1949), was that the club should consist of `about fteen people who had Wieners ideas before Wieners book appeared. In fact they found 20 such of Elmer: `. . . his electronic system found its equilibrium not for a precisely dened light intensity, but for quite a wide range. Thus, he was perfectly happy quietly ruminating under an arm-chair (de Latil 1956, pp. 213, 214). Even more mysteriously, he then quotes Walter on Elmer: `His reexes are really lacking, he is quite lifeless. For days on end he doesnt stir from under the furniture; I must liven him up a bit and make him more intelligent. Because, you see, if an individual is intelligent he has to pay the price of a certain degree of accompanying irritability (de Latil 1956, p. 214). Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2094 O. Holland people, including Ross Ashby, Horace Barlow, Donald Mackay, Philip Woodward, Alan Turing, Jack Good and A. M. Uttley. The groups attitude towards Wiener is nicely summed up by Bates in the draft of a paper on cybernetics intended for the British Medical Journal : Those who have been inuenced by these ideas so far, would not acknow- ledge any particular indebtedness to Wiener, for although he was the rst to collect them together under one cover, they had been common know- ledge to many workers in biology who had contacts with various types of engineering during the war. Bates (1952) Grey Walters own account of the genesis of the tortoises in The living brain empha- sizes the intellectual independence of the British tradition: The rst notion of constructing a free goal-seeking mechanism goes back to a wartime talk with the psychologist, Kenneth Craik, whose untimely death was one of the greatest losses Cambridge has suered in years. When he was engaged on a war job for the Government, he came to get the help of our automatic analyser with some very complicated curves he had obtained, curves relating to the aiming errors of air gunners. Goal-seeking missiles were literally much in the air in those days; so, in our minds, were scanning mechanisms. Long before the home study was turned into a workshop, the two ideas, goal-seeking and scanning, had combined as the essential mechanical conception of a working model that would behave like a very simple animal. Walter (1953a, p. 125) Walters own view of Wiener can be seen in a letter to Professor Adrian in 1947: We had a visit yesterday from a Professor Wiener, from Boston. I met him over there last winter and nd his views somewhat dicult to absorb, but he represents quite a large group in the States, including McCulloch and Rosenblueth. These people are thinking on very much the same lines as Kenneth Craik did, but with much less sparkle and humour. Walter (1947) Other members of the Ratio Club also acknowledged their indebtedness to Craik, and in fact when the name of the club was under discussion in the early meetings, one proposal was that it should be called the Craik Club. (b) Working models Although the production and use of physical working models of biological systems is now commonplace, it is rare to read a justication of the procedure. However, when the rst eorts at what Craik called the `synthetic method (Craik, cited in Cordeschi (2002)) appeared in the rst half of the twentieth century, each attempt usually included a discussion of its legitimacy and usefulness. (For an excellent recent Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2095 investigation of these issues, see Cordeschi (2002).) Grey Walters views on the sub- ject are hidden away in one of the appendixes to The living brain; they are typically forceful and well argued, and are worth reproducing: In general, it is legitimate to study a model of a mysterious process if three conditions are fullled: 1. Several features of the mystery must be known. 2. The model must contain the absolute minimum of working parts to reproduce the known features. 3. The model must reproduce other features, either as predictions, or as unexpected combinations. Walter (1953a, p. 280) He goes on to describe and discuss `An electric model of nerve. The source of the model is not given, but a reference earlier in the text implies that it is his own recent work. He remarks: Altogether the model seems to display 18 or so of the known proper- ties of nervous and synaptic excitation and propagation. Many of these properties were not foreseen as following inevitably from the elementary features originally specied as imitable. . . . The model is simply the ana- logue of one set of familiar mathematical expressions relating to passive networks linked by a nonlinear operator in the form of a discharge tube. It could quite well be formed of chemical or mechanical parts and does not in theory contain more information than do the algebraic equations. Its advantage is that, being a real object, it has constant dimensions; hence its predictions are more explicit and detailed than those of the equations, in which the constants are more arbitrary and independent. Walter (1953a, pp. 284{286) This passage is particularly valuable because both his theory and his example are separate from his work on the tortoises; this shows that the tortoises grew out of a well-considered method of investigation, and were not merely ad hoc novelties unconnected with the rest of his work. His view of the role of physical models in the broader progress of cybernetics is revealed in the summary of a lecture given in 1957: From these experiments with models and their counterparts, elaborate investigations of real people planned on the basis of the model studies, we are beginning to feel for a cybernetic specication of living mechanisms, in order to form crystallized hypotheses which are clear enough to see through, pure enough to swallow and brittle enough to break cleanly when they are proved inadequate, as of course they must be. Walter (1957) (c) Theories of brain function An idea that had interested Grey Walter for some time, and that he probably rst heard from Craik, was that a control system operating in a complex world would have to be able to represent that complexity internally. (Within cybernetics, and Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2096 O. Holland expressed in a revised and rened form, this became Ashbys well-known law of req- uisite variety.y) It was clear that the brain itself, with its vast number of neurons, was enormously complex, and so it was easy to assume that it was capable of repre- senting the external world adequately because of its sheer size. From the beginning, Grey Walter recognized the impossibility of building a device with anything like the number of functional units that the brain appeared to possess, and so he began to wonder if there was any way in which internal complexity could be achieved using a relatively small number of components. By 1948 he thought he had found a suitable method, and he had begun to wonder if in fact the brain used this method, rather than relying on its immense number of components. In a radio talk on the brain in that year, he spoke as follows: I am going to develop the hypothesis that the functional interconnection between brain cells bears some relation to the processes of thought and consciousness, and my rst premise is that the variety of permutation in these connections is at least as great as the diversity and complexity which we are subjectively aware of in our own minds and objectively assume in the minds of others. The general idea behind this hypothesis is that the function of the brain is to make a working model of external reality, and it is clearly of the rst importance to establish at the outset that there are as many working parts in the model as there are in the full scale pattern. Walter (1948) In `Totems, toys, and tools, ch. 5 of The living brain, he brought the practical and theoretical strands of this idea together: An entirely dierent approach seemed necessary to make it a practical problem, if we were to learn about life by imitation as well as observation of living things. . . . It meant asking whether the elaboration of cerebral functions may possibly derive not so much from the number of its units, as from the richness of their interconnexion. . . . As a hypothesis, this speculation had the great advantage that its validity could be tested experimentally. An imitation of two or three interconnected elements, including reexes to demonstrate their behaviour, should be a simple matter for a laboratory that had produced the EEG analyser. . . . Walter (1953a, p. 118) What lay behind this plan was that he had realized that the number of dierent ways of connecting up a number of elements was very much larger than the number of elements, and that it might be possible to devise a system to exploit this charac- teristic. He began by supposing that each dierent pattern of interconnection could be made to produce a dierent behavioural response, and worked out that a system of 1000 elements could be interconnected in about 10 300 000 dierent ways|a number y The law of requisite variety is usually expressed in the form: `Only variety (in the regulator) can destroy variety (in the system being regulated). (In Ashbys writings, `variety referred to the number of distinct actual or possible states in a system or subsystem.) For the original formulation of the law, see Ashby (1956). Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2097 of responses he thought sucient for any conceivable individual entity `. . . Even were many millions of permutations excluded as being lethal or ineective. . . (Walter 1953a, p. 120). But how might dierent patterns of interconnection give rise to dif- ferent observable behaviours? He gained some insight into this problem by reasoning as follows: . . . how many ways of behaviour would be possible for a creature with a brain having only two cells? Behaviour would depend on the activity of one or both of these cells|call them A and B. If (1) neither is active, there would be no action to be observed; if (2) A is active, behaviour of type a would be observed; if (3) B is active, behaviour of type b; if (4) A and B are both active, but independently, there would be behaviour of both types a and b, mixed together; if (5) A is `driving B, type b would be observed, but subordinate to A; if (6) B is `driving A, type a would be subordinate to B; if (7) A and B are `driving each other, behaviour will alternate between type a and type b. The internal states of such a system in these seven modes may be represented symbolically as 0; A; B; A + B; A !B; A B; A $ B with behaviour types, 0; a; b; a + b; b(fA); a(fB); ababab : : : : From the above it will be seen that the rst four ways of behaviour would be identiable by simple observation, without interfering with the system, whereas the last three could only be identied by operating on the system|by, as it were, dissecting out the arrows. Walter (1953a, p. 119) (He rst presented a shorter version of these arguments, using only six modes of behaviour, in the 1950 Scientic American paper.) The tortoises can therefore be seen on one level as Walters attempt to see whether a simple working model with a few behaviour-producing elements, capable of being interconnected in many ways, could be made to produce a relatively large number of behaviour patterns. However, he was not just concerned with behaviour in general, but with a set of behaviours that could be said to be characteristic and denitive of living beings, and in particular of animals. It was this audacity, and his apparent success in achieving his aims, that lifted the tortoises from being mere testbeds for the verication of a neurophysiological speculation to becoming icons of technological progress in the post-war British revival. The two key sources for tracing Walters thoughts about the imitation of life are the 1950 Scientic American article, and ch. 5 of The living brain, from 1953. They dier in a very interesting way. In 1950 he presented the tortoise enterprise as being an investigation of what could be achieved using a small number of behaviour-generating elements capable of being connected in a relatively large number of dierent permuta- tions. He merely noted some of the ways in which the tortoises behaviour resembled that of animals, and remarked: Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2098 O. Holland Figure 3. A labelled diagram of Elsie with her shell removed. These machines are perhaps the simplest that can be said to resemble animals. Crude though they are, they give an eerie impression of pur- posefulness, independence, and spontaneity. Walter (1950a) By 1953 he had a long list of the qualities required to pass the test of having imitated life: Not in looks, but in action, the model must resemble an animal. There- fore, it must have these or some measure of these attributes: explo- ration, curiosity, free-will in the sense of unpredictability, goal-seeking, self-regulation, avoidance of dilemmas, foresight, memory, learning, for- getting, association of ideas, form recognition, and the elements of social accommodation. Such is life. Walter (1953a, pp. 120, 121) Of the thirteen attributes in the new list, six (from foresight to form recognition) are those derived from the learning machine CORA, and so are not really properties of the tortoises alone. It is dicult to avoid the conclusion that Walter drew up his 1953 list on the basis of his interpretations of the observed behaviour of the tortoises, rather than having designed the tortoises to display every characteristic in the list. The test of the successful imitation of life applied in the 1950 paper is much more relaxed and intuitive than the test implied in The living brain, and it is reasonable to suppose that his judgement of his success in 1950 was independent of the 1953 list. The individual examples of tortoise behaviour are of course the same in both sources. 4. Tortoise mechanisms and functions In order to understand how the tortoises worked, it is best to start with the mechan- ical design. Figure 3, from the archives of the BNI, is a conveniently labelled view Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2099 Figure 4. The original circuit diagram for the 1951 batch of tortoises. It is clearer than the version in The living brain, and is particularly useful because it also shows the component types and values. of Elsie with her shell removed. On the left, a single structure extends vertically downwards from the photoelectric cell to the driving wheel; this structure is capable of being rotated in one direction only about its vertical axis by the steering motor acting via the steering gear. The driving wheel also rotates in one direction only when driven by the driving motor. The two rear wheels, seen on the right, are not driven. The photoelectric cell is tted with a shroud which blocks the entry of light from all directions except the front of the cell; the shroud and cell are aligned with the driving wheel so that the direction in which light is sensed is always the direction toward which the driving wheel is moving. When the shell is tted, it is hung on the rubber touch contact mount. If the tortoise is on the level, and the shell is not touching anything, the contact is open circuit, but if the shell is deected by gravity (when the tortoise is on a slope) or touch, the contact is closed. The headlamp is a torch bulb connected in series with the steering motor, so that it is lit only when the steering motor is turned on. A hole in the shell allows the headlamp to be seen from the front when the shell is on. The circuit diagram shown in The living brain is unfortunately not very easy to understand. Figure 4 shows a rather clearer version from the BNI archives which describes the almost identical circuit used in the 1951 batch. The photoelectric cell (PEC) is on the extreme left; just above it is the touch contact. The two miniature thermionic valves, or vacuum tubes, are represented by the two circular symbols in the upper half of the diagram. Current ows through the valves from the upper at electrodes (the anodes, or plates) to the lower looped electrodes (the cathodes). The dotted lines within the symbols represent the control electrodes, which control the current ow; the upper control electrode is known as the screen, and the lower as the grid. The valves are arranged as a two-stage amplier. The input voltage to the rst stage (the valve on the left) is determined by the amount of light falling on the photoelectric cell; this voltage controls the current through the valve. The voltage produced by this current is fed to the second stage, and controls the current through Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2100 O. Holland the valve on the right. Above each valve is a relay coil (marked RL1 and RL2); when sucient current passes through a relay coil, it changes over the contacts on the corresponding relay (marked RL1 and RL2 below each valve), switching from the contact on the left (`o) to the contact on the right (`on) as the relay coil current increases. The contacts of RL1 can deliver only about half the current of RL2, as RL1 is connected to the power line via the headlamp and a resistor, whereas RL2 is connected directly. How do the mechanical construction and the circuit interact with environmental conditions to produce behaviour? The standard sources (Walter 1950a, 1953a) give little more than summary descriptions of the behaviours, with very shallow technical content. However, a remarkable document found in the BNI archives (Walter 1960) enables us both to understand in detail how the tortoises worked, or were meant to work, and to appreciate how Grey Walter himself viewed their operation. The typescript, headed `Machina speculatrix|notes on operation, appears to date from around 1960. The main text, which is probably unnished, since it ends rather sud- denly and without any conclusion, is a clear and insightful technical explanation of how the tortoise produces lifelike behaviour. It was published for the rst time in Holland (1996) in a slightly shortened version; the full text can be found in Holland (2003). It is too long to reproduce here; what follows is a precis of it, with brief extracts from the critical passages. When the circuit is switched on in the absence of light, the current through the rst stage is at a maximum, and that through the second stage is at a minimum. RL1 is therefore switched on, and RL2 is o. This produces what Walter calls behaviour pattern E (for exploration): 1. The driving motor is on at half speed, the headlamp being in series with it. This propels the model slowly in the direction of the driving wheel. But this is itself rotated by| 2. The steering scanning motor (which) is on at full speed. This turns the driving wheel continuously so that the direction of motion is con- tinually changing. The same spindle supports the photoelectric cell or `eye so this rotates too; it is always `looking in the direction in which the model is moving. This is called scanning. The combination of lin- ear motion (driving) with circular rotation (steering-scanning) gives the model a cycloidal trajectory, rather like a point on the wheel of a moving vehicle. This cycloidal exploration continues indenitely in the dark or when there is no light on the horizon bright enough to aect the `eye. (The mask on the eye provides both blinkers that give it a direction of gaze and a visor that stops it seeing ordinary room lights above it.) Walter (1960) When the photoelectric cell is aligned with a moderately bright source of light, the current through the rst stage decreases slightly|not enough to aect the relay RL1|but the current through the second stage increases markedly, switching RL2 on. The turning motor now receives no current, but the drive motor, fed by RL1 and RL2, receives maximum current. This produces behaviour pattern P (for positive phototropic response): Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2101 . . . the driving wheel is xed at whatever angle it was when the light was seen, and the scanning of the horizon by the eye also stops of course. At the same time. . . the driving motor is turned up to full speed. The model stops looking slowly round and hurries toward the light. However, unless the light was seen when the eye happened to be facing straight ahead, the angle at which the steering came to rest at the moment of sighting will deect the model gradually away from the light. When the deection is so great that the activation level of the photo-cell falls below threshold, the Relay 2 opens again, the scanner starts up, the drive is reduced to half speed and the model is re-positioned, this time so that the light is more directly ahead. This process of progressive orientation is an important part of the behaviour mechanism. It is cumulative|every time the model steers itself slightly o-beam the momentary operation of the steering-scanning mechanism brings it back more nearly on course and it ends up with a heading on-beam. The process often looks clumsy, because the eye seems to veer away from the light and then the scanner has to make nearly a whole rotation to bring it back, but inevitably with each such operation the model gets itself into a better position to bear down directly on its goal. The aiming-error is steadily reduced as the goal is approached. Walter (1960) As the tortoise approaches a light, the intensity of light falling on the photoelectric cell will increase, and for a suciently bright light the current through RL1 will become low enough to cause RL1 to switch o. The current through RL2 will increase, but this will have no eect since RL2 is already on. The turning motor will therefore operate at half speed, and the drive motor will operate at full speed. This produces what Walter calls behaviour pattern N, for negative phototropism: The result is that when the model gets `too close to a light it veers smoothly away from it and avoids the fate of a moth in a candle. M. Speculatrix is moderate and restrained|it seeks an optimum light, not a maximum. Walter (1960) We can take issue with Walter here. In fact, the model will veer smoothly away from the light only for a moment, because the shroud on the photoelectric cell makes its response highly directional; as soon as the turning motor has turned the spindle so that the sensor is no longer aligned with the light source, the model will revert to behaviour pattern E. As it scans round, the sensor will once again become aligned with the light, and behaviour pattern N will again be triggered, and so on. The model certainly avoids the moths fate, but it is clear that N is simply E with dierent motor speeds, rather than the opposite of P, and is not really a negative phototropism. If the shell makes contact with an obstacle, or if the gradient is suciently steep, the stick-and-ring touch contact will close. Via a capacitor, this connects the anode of the valve in the second stage of the amplier to the grid of the valve in the rst stage, and this phase-shifted feedback connection converts the circuit into an oscillator, with a period of around a second. This produces behaviour pattern O, for obstacle avoidance: Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2102 O. Holland As arranged in M. Speculatrix, the oscillators [sic] recur about once a second, and their eect is to open and close Relays 1 and 2 alternately as long as the skin is displaced. This makes the model butt, turn and recoil continuously until it is clear of the obstacle. It may edge steadily along until it comes to an edge it can get round, it may shove the obstacle to one side if it is movable, or if it gets into a tight corner it may end by swivelling right round and trying another approach. In any case it is very pertinacious and it is also quite discerning, because as long as it is in trouble it will not respond to a light, however intense and attractive. It cannot, because as long as the skin is displacing the limit switch the ampliers are completely preoccupied with sending signals back and forth to one another and are quite blind to outside information|an oscillator does not act as an amplier. When the model has cleared an obstacle and the skin swings back to its normal position, the input{output circuit is opened and after one more oscillation the ampliers resume their function of transforming light signals into movements of the relays and the whole model. Walter (1960) In some of his writings, he made much of the persistence of the oscillation for one cycle after the obstacle had been cleared, often referring to it as a memory of the obstacle. De Latil was also impressed by this `. . . astonishing \memory" of the colli- sion with the obstacle (de Latil 1956, p. 220) but it is surely an overinterpretation. Behaviour patterns E, P, N and O constitute the entire behavioural repertoire of the tortoise; all the reported complexity of tortoise behaviour is a result of the sequential activation of these behaviour patterns by the interaction between the tortoise and the environment. 5. Tortoise behaviour: the evidence All that we seem to require now is conrmation that the behaviour described in x 4 and in the published sources was actually produced by the tortoises at the time. Again, the BNI archives have provided the crucial evidence, in the form of a set of time exposure photographs in which Elmer and Elsie, with lighted candles mounted on their shells, trace out their trajectories in various environmental arrangements. (It must have been necessary to shield each tortoises photocell from the light of the candle it was carrying; under magnication, one of the photographs shows what looks like a screen mounted just behind the photocell.) The photographs of single tortoise behaviour are all of Elsie, who seems to have been rather more responsive than Elmer, perhaps because Elmer was the prototype (Pincher 1949). De Latil reported that on the day of his visit to Walters house in 1950 . . . Elsie was aicted with a very unstable, very feminine mood; her reg- ulating mechanism was hypersensitive. . . . Elmer, on the other hand, had been given a very stable, very bourgeois character. . . . de Latil (1956, p. 213) Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2103 Figure 5. Elsie and Elmer, released at the top of the picture, interact with each other, and then head for the hutch. The photographs were taken at Grey Walters house, probably in late 1949 or early 1950. They may well have served as the basis for some of the drawings of trajectories in `An imitation of life. As a very valuable bonus, the BNI archives also yielded a typescript entitled `Accomplishments of an artefact, consisting of descriptions of what seem to be several of the photographs, and of others that have unfortunately been lost. The undated document seems to have been written to accompany some sort of display or exhibition of work at the BNI; it mentions The living brain, so it dates from 1953 or later. Although I have attributed the text to him (Walter 1953b), it is impossible to be sure that Grey Walter was the author, but some of the phrasing is very similar to his other work, and it seems likely that the document must at least have been approved by him. In what follows, the photographs will be matched with their apparent descriptions; as will be seen, the combination is much more informative than either source of information taken alone. A somewhat more technical examination of some of these materials can be found in Holland (2003). We begin by describing gure 5 in some detail to show how the basic behaviour patterns can be identied from the traces. Elsie (with the smooth one-piece shell) is at the top left at the beginning of the exposure, and Elmer (with the segmented shell, the original tortoise) is at the top right. After a rather messy interaction with Elmer, Elsie crosses his track. Immediately afterwards, we can see Elsie executing behaviour pattern E: the faint cycloidal trace (about ve cycles) is made by the headlamp, and the heavy zigzag by the candle. Elsie then switches into behaviour pattern P for about a body length, presumably as a result of catching a glimpse of the hutch light, and then reverts to E for a couple of cycles. She then makes a long straight excursion towards the hutch (P again) followed by another ve cycles of what may Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2104 O. Holland Figure 6. Elsie demonstrates `pertinacity. be E, or a mixture of E and N. Elmers behaviour is less clear because his headlamp is too dim to be seen, but the candle trace implies a similar alternation of E with P. `Accomplishments of an artefact contains what appears to be a rather misleading description of this photograph: Social organisation The formation of a cooperative and a competitive society. When the two creatures are released at the same time in the dark, each is attracted by the others headlight but each in being attracted extinguishes the source of attraction to the other. The result is a stately circulating movement of minuet-like character; whenever the creatures touch they become obsta- cles and withdraw but are attracted again in rhythmic fashion. While this evolution was in progress the light in the feeding hutch was turned on; the common goal disrupted the cooperative organization and trans- formed it into a ruthless competition, in which both creatures jostled for entrance to the source of nourishment. Walter (1953b) Our dissatisfaction with this is triggered by the description in the same document of the behaviour of Elsie in what is almost certainly gure 6: Pertinacity Catching sight of a faraway candle the creature loses itself behind an opaque and polished re-screen, behind which it sidles. On the way it catches sight of the reection of its candle in the re-screen and spends some time chasing its tail, but later catches another glimpse of the distant candle and homes into an orbit round its original goal. Walter (1953b) This passage, reasonably satisfactory in itself, shows that the candles used to produce the trajectory traces are bright enough to trigger behaviour pattern P, and so it is probable that the interaction between the two tortoises in gure 5 is mediated wholly or partly by the candles, rather than by the headlamps. Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2105 Figure 7. Elsie performs in front of a mirror, but is probably responding to the candlelight rather than to her pilot light. The same criticism can be applied to gure 7, which shows Elsie in front of a mirror. This may be the photograph described in `Accomplishments of an artefact as follows: Recognition of self A pilot light is included in the scanning circuit in such a way that the headlamp is extinguished whenever another source of light is encoun- tered. If, however, this other source happens to be a reection of the headlamp itself in a mirror, the light is extinguished as soon as it is per- ceived and being no longer perceived, the light is again illuminated, and so forth. This situation sets up a feedback circuit of which the environ- ment is a part, and in consequence the creature performs a characteristic dance which, since it appears always and only in this situation, may be regarded formally as being diagnostic of self-recognition. This suggests the hypothesis that recognition of self may depend upon perception of ones eect upon the environment. Walter (1953b) The drawing of the famous `mirror dance in `An imitation of life is nothing like the regular alternation between approach and avoidance shown in the photograph, being an altogether more irregular and complex trajectory. There may well have been a mirror dance that could have been argued to be a form of self-recognition, but unfortunately this photograph cannot be said to be a record of it. The brightest light visible to the camera, and presumably to the photocell, is the candle on the tortoises back and its reection in the mirror. The trace is far more likely to reect the alternation of behaviour pattern P (approach to the reected candlelight) with behaviour pattern O (obstacle avoidance on contact with the mirror). We can be sure that Walter used this image as an example of the mirror dance because it appears in the form of a diagram in the transcript of a talk he gave in 1954 (Walter 1956b); the text matches closely the account given in `Accomplishments of an artefact. Interestingly, the description of the mirror dance in de Latils book also matches this photograph rather than Grey Walters original description and Bernarda Bryson Shahns sketch: Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2106 O. Holland Figure 8. `Discernment. Elsie at rst approaches the lamp, but on encountering the obstacle she `ignores the lamp until she has escaped from the obstacle. Yet another trap, a mirror, was placed in front of Elsie. What would she do? As if attracted by her own image, she approached the mirror, where the light from her breast was reected. But she hit herself against the glass. She then waltzed around the mirror in zigzag movements, to and fro, as if admiring her own reection. de Latil (1956) In The living brain, Grey Walter characterizes discernment as `Distinction between eective and ineective behaviour and describes the tortoises manifestation of it as follows: When the machine is moving towards an attractive light and meets an obstacle, or nds the way too steep, the induction of internal oscillation does not merely provide a means of escape|it also eliminates the attrac- tiveness of the light, which has no interest for the machine until after the obstacle has been dealt with. There is a brief `memory of the obstacle, so that the search for lights, and the attraction to them when found, is not resumed for a second or so after a material conict. Walter (1953a, p. 128) This `memory is simply the persistence of the oscillation for a few cycles after the switch is opened. Figure 8 appears to correspond to the description of `discernment in `Accomplish- ments of an artefact: Presented with a remote goal (seen at the top of the slide) the creature encounters a solid obstacle which it cannot move, and although it can still Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2107 Figure 9. Elsie appears to achieve an optimum. see the candle it devotes itself to circumventing the obstacle (of which it retains a short memory) before it circles round in an orbit and reaches the objective. Walter (1953b) It is clearly not possible to deduce all of this from the trace of this trajectory. Dealing with the obstacle involves four or ve forward and backward movements of the candle; it is impossible to be sure that the forward movements did not contain any episodes of behaviour pattern P. Similarly, the trajectory contains no unequivocal evidence for the claimed memory of the obstacle. However, it is at least consistent with the description. Figure 9 is probably the photograph described in `Accomplishments of an artefact as `search for an optimum: Attracted at rst by a distant bright light the creature reaches the zone of brilliant illumination where it is repelled by the excessive brilliance of the light and circles round it at a respectful distance, exhibiting a search for optima rather than maxima|the idea of moderation of the classical philosophers. Walter (1953b) The implication here is that the tortoise approaches the light using behaviour pattern P, but when the light becomes too intense, it is repelled by behaviour pattern N. The problem is that, although P is easily recognized by its long straight or curving runs, it is impossible to tell whether a run of P is terminated by E or N. E will occur quite frequently, when the motion due to P causes the photocell to lose its alignment with the light; N will occur only if the light becomes too bright while still in alignment|but the immediate eect of N will be to rotate the photocell, which will soon cause the light to become misaligned and lead immediately to E. It is therefore Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2108 O. Holland Figure 10. Elsie demonstrates `free will: the solution of the dilemma of Buridans ass. impossible to conclude from the photographic evidence that N is denitely involved to any degree in this particular trajectory; it could easily have been produced by P and E acting alone, and would then merely constitute a failure to achieve a maximum, rather than a success in achieving an optimum. For Grey Walter, one of the major achievements of the tortoise was the demonstra- tion of `free will, in the sense of unpredictability; in fact, the 1950 Scientic American article was subtitled `Concerning the authors instructive genus of mechanical tor- toises. Although they possess only two sensory organs and two electronic nerve cells, they exhibit `free will. Figure 10 shows the experimental set-up he used to demon- strate this: two light sources, with the tortoise started equidistantly from them. In `Accomplishments of an artefact, the commentary on what is probably gure 10 runs as follows: Free-will The solution of the dilemma of Buridans ass. The photoelectric cell which functions as the creatures eye scans the horizon continuously until a light signal is picked up; the scanning stops, and the creature is directed towards the goal. This mechanism converts a spatial situation into a tem- poral one and in this process the dilemma of two symmetrical attractions is automatically solved, so that by the scholastic denition the creature appears endowed with `free-will. It approaches and investigates rst one goal and then abandons this to investigate the other one, circling between the two until some other stimulus appears or it perishes for want of nour- ishment. Walter (1953b) In his writings about the tortoises, Grey Walter gave much weight to an attribute he called `internal stability|the claimed ability of the tortoises to maintain their battery charge within limits by recharging themselves when necessary. (In fact, the names Elmer and Elsie were derived from ELectroMEchanical Robots, Light- Sensitive with Internal and External stability (Walter 1950a, p. 43).) A feature of Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2109 Figure 11. Elsie enters the hutch from a starting position on the left. The reection on the oor shows that the hutch light is on, and so behaviour pattern N must have been disabled, either by running down the battery or by adjusting the relay settings. the tortoises circuitry was that, as the batteries became exhausted, the amplier gain decreased, making it increasingly dicult to produce behaviour pattern N (negative phototropism). Walter installed an automatic recharging system inside the tortoises hutch, along with a 20 W lamp. (Part of the charging system can be seen at the back of the hutch in gure 1; the light from the lamp is also visible.) Initially, a tortoise would be repelled by the bright light from the lamp (behaviour pattern N), but once the battery had run down suciently, behaviour pattern P would be produced, and the tortoise would approach the lamp, enter the hutch, and be engaged by the charg- ing system. Any further movement would be prevented until the battery was fully charged. When the tortoise was automatically released from the charger, it would once again be repelled by the lamp (behaviour pattern N) and would leave the hutch. Although the scheme is obviously satisfactory in theory, there are no records show- ing that this cycle of events was ever demonstrated with complete success, though the reduction in gain denitely occurs. Perhaps signicantly, Walter remarks: This arrangement is very far from perfect; there is no doubt that, if left to themselves, a majority of the creatures would perish by the wayside, their supplies of energy exhausted in the search for signicant illumination or in conict with immovable obstacles or insatiable fellow creatures. Walter (1953b, p. 130) Interestingly, when the original hutch was set on re by the lamp and destroyed, the replacement hutch was tted with a lamp, but no charging system. Whatever doubt may exist concerning the success of the implementation of internal stability, there is no doubt that the tortoises could make their way into the hutch under the inuence of the lamp; gures 11 and 12 show the trajectories of two such Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2110 O. Holland Figure 12. Elsie enters the hutch from a starting position on the right. movements. In `Accomplishments of an artefact, these two photographs are described as follows: Simple goal-seeking Started in the dark the creature nds its way into a beam of light and homes on the beam into its feeding hutch. Walter (1953b) The reference to the feeding hutch is now the only relic of the idea of internal stability. Although the analyses of the surviving photographs and of the BBC newsreel have given us some clues to the tortoises abilities, there is still considerable doubt about the exact nature of some of the claimed behaviours. One possible method of resolving the situation would be to construct replicas of the original tortoises, and to study their behaviour in the same way as did Walter. The problem with this approach is that there is simply not enough detailed information available about Elmer and Elsie; the best source, Appendix B of The living brain, is too general to serve as a specication, and every roboticist knows that, in robotics more than in most other areas of engineering, the devil is in the details. However, things are dierent when it comes to the 1951 batch of tortoises: we have excellent documentation, as well as two surviving examples. In 1995, two replica tortoises were constructed at the University of the West of England. They used two 1951 shells obtained from the BNI, and were close to the originals in all important aspects, though modern motors and batteries were used. (For a description of this project, and for the rst publication of some of the archival materials from the BNI, see Holland (1996).) At that time, the newly discovered tortoise now in the Science Museum had been made functional again with the assistance of `Bunny Warren; although testing was limited in extent by the poor condition of the original tortoise, the replicas appeared to behave in very much the same way. This was unfortunate, because the production of behaviour pattern N in a normal environment turned out to be a very dicult and uncertain matter, even for the original tortoise. There is probably no single Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2111 reason for this. The magnetic properties of the relays of the replicas were inferior to those of the original, and so the switching threshold characteristics were dierent; the photoelectric cells used were the original type, but were of necessity several decades old, and known to become less sensitive both with use and with time; and there is documentary evidence that the tortoises of the 1951 batch were much less sensitive to light than Elmer and Elsie, and so the production of N may have been a problem anyway (Holland 1996). The replicas were able to reproduce behaviour sequences involving only behaviour patterns E, P and O, but failed when the involvement of N was required. It would of course be possible to modify the circuitry of the replicas to amplify the output of the photocell, but this has not yet been undertaken. In fact, there are no records of any of the 1951 tortoises having been used to demonstrate all the functions claimed by Walter, so any tests (especially unsuccessful ones) carried out on the replicas would not necessarily enable conclusions to be drawn about the behaviour of Elmer and Elsie. 6. Later work: CORA and IRMA Walter did not conne himself to reproducing the behaviours identied in the original paper on the tortoises, but extended his observations intermittently over the years. Some of his ideas have a very modern ring to them. For example, he noticed how the tortoises behaviour in a modiable environment could produce apparently useful changes in the environment: If there are a number of light low obstacles that can be moved easily over the oor and over which the model can see an attractive light, it will nd its way between them, and in doing so will butt them aside. As it nds its way toward the light and then veers away from it and wanders about it will gradually clear the obstacles away and sometimes seems to arrange them neatly against the wall. This tidy behaviour looks very sensible but is an example of how apparently rened attitudes can develop from the interaction of elementary reex functions. Walter (1960) Figure 13 shows what is clearly the set-up for such an experiment. Unfortunately, no further accounts or images of this work have yet been found, but it is without any doubt a signicant anticipation of the section of modern robotics dealing with the use of stigmergy (Bonabeau et al . 1999). In his obituary of Walter, Shipton noted: The successful design of Machina speculatrix, the famous tortoise, gave Grey enormous satisfaction and led him to consider whole families of cybernetic models of biological systems, only a few of which saw the light of day. Shipton (1977) We have no means of knowing just how many were physically realized, but at least two have survived: CORA and IRMA (innate releasing mechanism analogue). Both are now in the Science Museum. Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2112 O. Holland Figure 13. Walter watches as one of the 1951 tortoises clears some movable blocks out of its path|perhaps the rst observations of stigmergy in robots. (a) CORA In the list of life-like characteristics presented in The living brain, Walter included `foresight, memory, learning, forgetting, association of ideas, form recognition. None of these was present in the tortoises; in fact, they are the claimed attributes of another electromechanical model (CORA; see gure 14) originally developed to illustrate some of his ideas about Pavlovian conditioning. In many ways, CORA was closer to his heart than were the tortoises, as becomes clear on reading The living brain, which contains a whole chapter on the theory underpinning the model, along with diagrams of its outputs, and an appendix giving detailed circuit information. He had worked on conditioning for many years, and had met Pavlov, and worked with Pavlovs students. More importantly, he had studied the eects of the process of conditioning on the EEG, his primary sphere of interest. In Walters view of Pavlovian conditioning, an initially neutral stimulus acquires the ability to produce the specic reex response normally triggered by some spe- cic stimulus; the learning procedure consists of the repeated presentation of the neutral stimulus before the specic stimulus. His analysis had convinced him that in order to reproduce this type of learning in a model `. . . no fewer than seven distinct operations must be performed (Walter 1951). CORA was a straightforward elec- tronic implementation of these operations. As a neutral stimulus, he chose a whistle, coupled to CORA with a microphone. The specic stimulus could be any stimulus input producing one of the tortoises behaviour patterns|he mentions a moderate light (producing P), and contact with an obstacle (producing O). Although he gives detailed treatments of the signal processing within CORA during and after learning, his descriptions of actual experiments with tortoises are very sketchy: In one arrangement. . . the specic stimulus is a moderate light and the neutral one is the sound of a whistle. The whistle is blown just before Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2113 Figure 14. CORA. the light is seen; after this has been repeated 10 or 20 times the model has `learned that the sound means light and will come to the whistle as though it were a light. . . . In another arrangement the specic stimulus is touch, that is, an encounter with an obstacle. In that case the whistle is blown just as the model comes into contact with the obstacle, so that after a while the warning whistle triggers a withdrawal and avoidance reaction. This process may of course be accelerated by formal education: instead of waiting for the creature to hit a natural obstacle the experimenter can blow the whistle and kick the model. After a dozen kicks the model will know that a whistle means trouble, and it can thus be guided away from danger by its master. Walter (1951, p. 62) In spite of his eorts, which included the publication of an article in Scientic American (Walter 1951), CORA had only a small fraction of the impact of the tortoises, and made little lasting impression. This is perhaps unfortunate and unfair, because his ideas on the relationship between learning and statistical prediction were far ahead of their time. Nevertheless, he was undoubtedly the rst person to integrate a meaningful physical model of learning into a robot worthy of the name, although the closeness of that integration is not at all clear. Some of his writings give the impression that the rst version of CORA had been built into a tortoise, but none of Grey Walters colleagues at the time have any clear recollection that this was ever done. Once more, de Latil provides some interesting information: At rst, the learning circuits were incorporated in a `Tortoise. . . but this was too complicated for class demonstrations, so they were put in a dis- play box. de Latil (1956, caption facing p. 51) Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2114 O. Holland Figure 15. Grey Walter posing with a tortoise and CORA on a visit to New York. The main purpose of the nal version of CORA was undoubtedly pedagogic|the superstructure can be unplugged and replaced to face away from the controlling dials, so that the operator does not block the classs view of the progress of `learning. De Latil also sheds some light on Elsies fate: At the end of 1950, Elsie and Elmer had a little sister, unless it was a daughter as it may well have been. In truth it appears to have been a case of parthenogenesis. CORA, the newcomer, was constructed with organs belonging to Elsie, who thus succumbed, poor thing! de Latil (1956, p. 247) (However, an inspection of CORAs internals reveals that the components are more similar to those used in the later batch of tortoises than to those visible in pho- tographs of Elsie.) In Walters 1954 talk, he states . . . the situation gets incredibly complicated when you have a model. . . of learning by association actually moving round the room. . . . So I detached the learning apparatus from the moving model, and what I am going to show you now are preparations rather than complete working models. Walter (1956b, pp. 40, 41) He then goes on to demonstrate CORA, later remarking, `I have done some work with a moving model equipped with one of these learning devices (Walter 1956b, p. 53). The photograph accompanying the published talk is the one used in the second Scientic American article; from the shell, the tortoise appears to be from the 1951 batch, so it is unlikely that it is tted with an internal version of CORA. However, Walter certainly performed experiments with CORA externally connected to a tortoise; gure 15 shows such an arrangement, clearly posed for the camera Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2115 Figure 16. IRMA. rather than set up for a real experiment. The tortoise in the Science Museum, which was Walters `personal tortoise, still retains connections made to its circuitry which could well have formed the interface to CORA, especially since it is in fact the tortoise in gure 15. (b) IRMA Perhaps the last of Walters behavioural models was IRMA. Both the device (g- ure 16) and the circuit diagram survive, along with an explanatory diagram. Almost all of what we know of IRMA comes from a single source|the transcript of his 1954 talk to a small group including the ethologist Konrad Lorenz: When Lorenz was with us in Bristol about a year ago, I said that when we next met I would have a model of imprinting to show him. . . . I have made a model which Lorenz may say is not in fact a model of imprinting at all, but it is a model of something, and it has some interesting properties. Walter (1956b, p. 40) The origins of the name are clear: This model is called I.R.M.A. It consists of two networks which are arranged for convenience in three layers. . . . This might conceivably be helpful in neuro-anatomical analysis because it suggests that for instinc- tive, innate releasing mechanism (I.R.M.) type of behaviour one should look for a minimum of three levels. . . . Walter (1956b, p. 42) His explanations of what it does and how it works take two pages, and lack his usual clarity, but if, after switching the machine on, one of the six push buttons on the top is pressed, all the available power will be diverted to the corresponding circuit (illuminating the associated light); from then on, only that button will produce any more than a minimal response when pressed. This primacy eect is presumably what is supposed to be an analogue of imprinting. However, if the power supply is adjusted (perhaps by rotating the knob marked `AGE on the model), the primacy eect can Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2116 O. Holland give way to a recency eect, in which the system responds to the latest button press regardless of any made previously: `. . . instead of being an imprinting model it becomes a model of fashion (Walter 1956b, p. 43). He does not mention having made any attempt to connect it to a tortoise, and apart from a note of its appearance in a display at the Royal Institution in 1956 it does not seem to be mentioned in print again. 7. Grey Walter and biologically inspired robotics We can approach the assessment of the signicance of Grey Walters work for bio- logically inspired robotics by asking three distinct questions. What was he the rst to do? To what extent did his work inuence the subsequent course of biologically inspired robotics? To what extent has biologically inspired robotics developed in the directions he foresaw? It will be convenient to examine each of these in turn, but we should rst attempt to decide what we mean by biologically inspired robotics. There are two main moti- vations for linking biology and robotics. The rst is to attempt to make good robots by copying or adapting ideas from biology; the second is to use robots as tools for biological investigation. These proceedings furnish examples of each. Straddling both approaches is the area of what has become known as adaptive behaviour, in which both natural and articial systems, including biologically inspired robots, are studied in a context emphasizing such notions as embodiment, autonomous agents, emergence, morphology and agent{environment coupling. A technique, even a phi- losophy, common to much of this work is that of behaviour-based robotics, associated with R. A. Brooks and MIT. However, the scope of behaviour-based robotics is much broader than that of biologically inspired robotics; it was developed as a reaction to the dominant paradigm of knowledge-based robotics, and is now established as one of the main strands within robotics research. (a) What was Grey Walter the rst to do? Walters list of rsts in biologically inspired robotics and the related areas is impressive. The tortoises were designed to test a biological hypothesis about how combinations of relatively few elements might give rise to complexity of behaviour; they were probably the rst biologically inspired robots of any real interest.y The robots were intended to produce behaviour characteristic of animals, and Walter was the rst to emphasize the importance of behavioural completeness: Not in looks, but in action, the model must resemble an animal. There- fore, it must have these or some measure of these attributes: explo- ration, curiosity, free-will in the sense of unpredictability, goal-seeking, self-regulation, avoidance of dilemmas, foresight, memory, learning, for- getting, association of ideas, form recognition, and the elements of social accommodation. Such is life. Walter (1953a, pp. 120, 121) y Some earlier phototropic mobile robots had been given animal-like names (e.g. the `Electric Dog of Hammond and Miessner, dating from 1912 (Cordeschi 2002) and the 1929 `Philidog designed by Piraux (de Latil 1956)); however, as de Latil remarks, such automata `did nothing unexpected (de Latil 1956, p. 240). Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2117 The tortoises had to exist in a normal everyday environment, rather than in some special environment created to take account of their limitations. He was the rst to implement a self-recharging robot. He made the rst observations of emergence in robotics, both in the sense of the designer being pleasantly surprised at the unan- ticipated appearance of some useful side eect of his design, and in the sense that the interaction of two or more behavioural subsystems could produce a distinct and useful additional behaviour. The second sense is clearly demonstrated by several of his remarks in Walter (1960); in fact, they amount to the earliest formulation of the basic idea of behaviour-based robotics (Holland 1996). As noted above, he was the rst to show how a robots actions on an environment could change it in such a way that the robots future behaviour was changed in a useful way, and he was also the rst to carry out experiments in learning on a behaving robot. Because he built more than one robot, he was also the rst in the eld of multiple robotics, showing how the behavioural interactions between two robots of the same type would produce emergent characteristics of interest if not utility. He also made the earliest observations in the eld of what is now known as collective robotics: Simple models of behaviour can act as if they could recognize themselves and one another; furthermore, when there are several together they begin to aggregate in pairs and ocks, particularly if they are crowded into a corral. . . . The process of herding is nonlinear. In a free space they are individuals; as the barriers are brought in and the enclosure diminishes, suddenly|there is a ock. But if the crowding is increased, suddenly again there is a change to an explosive society of scuing strangers. And at any time the aggregation may be turned into a congregation by attraction of all individuals to a common goal. Further studies have shown that in certain conditions one machine will tend to be a `leader. Often this one is the least sensitive of the crowd, sometimes even it is `blind. Walter (1957) (b) To what extent did Grey Walters work inuence the subsequent course of biologically inspired robotics? This question is somewhat embarrassing for the robotics community, because the answer is that it had very little direct inuence. Perhaps the technology of the tor- toises was too inaccessible; without a grounding in the electronics of the post-war world, it can be dicult to understand how their circuits operate. Perhaps the papers and the book are too removed in time, tone and style from modernity; The living brain, in particular, is very old fashioned, but this is perhaps not surprising when one discovers that it was written by his father, who was of course educated in the nineteenth century. Many of what are regarded as the key achievements within bio- logically inspired and behaviour-based robotics have involved techniques and obser- vations with which he was familiar, but which have had to be painstakingly rediscov- ered in modern times. His technical priority does not of course diminish the credit due to the modern investigators, especially since one of his most important texts (Walter 1960) was rst published less than a decade ago. However, one wonders whether robotics in general, and biologically inspired robotics in particular, might Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2118 O. Holland have advanced further and faster if his work had not been allowed to fade away quite so fast. On the other hand, the indirect eects of Grey Walters work may have inuenced modern robotics in a number of ways. The publicity the tortoises received encouraged many technically inclined individuals to try and build similar machines; the electronic hobbyist magazines of the period record many such projects. In particular, Rodney Brooks recalls attempting to build his own version of the tortoise after reading The living brain (Brooks 2002). Later, while Brooks was working with Hans Moravec on the Stanford Cart, the tortoises again came to mind: Despite the serious intent of the project, I could not but help feeling disappointed. Grey Walter had been able to get his tortoises to operate autonomously for hours on end, moving about and interacting with a dynamically changing world and with each other. His robots were con- structed from parts costing a few tens of dollars. Here at the centre of high technology, a robot relying on millions of dollars of equipment did not appear to operate nearly as well. Internally it was doing much more than Grey Walters tortoises had ever done|it was building accurate three- dimensional models of the world and formulating detailed plans within those models. But to an external observer all that internal cogitation was hardly worth it. Brooks (2002, p. 30) Less than a decade later, Brookss own design principles were producing simple reactive robots within the new behaviour-based philosophy. (c) To what extent has biologically inspired robotics developed in the directions Walter foresaw? Answering this question is made easier because in 1968, in his last writings on the tortoises, Walter speculated about how they might be developed in the future: It would only be a matter of patience and ingenuity to endow M. specula- trix with other `senses besides sight and touch, to enable it to respond to audible signals audibly, and so forth; also to provide it with hands. There is no serious diculty about the elaboration of function, once the prin- ciples of mechanical `life have been demonstrated in a working model. If the principles are preserved, no matter how elaborate the functions of the machine, its mimicry of life will be valid and illuminating. Our experiments and observations with Machina speculatrix and its cousins and ospring (of which there are now quite a number) shows us that, as we had suspected, the complexity of the brain and its behaviour can be imitated to an extraordinary degree by relatively simple machin- ery. And a new era has opened out in our experiments. Instead of using the old-fashioned valves with their extravagant heaters and their enor- mous bulk, we can make the same device with transistors. And with micromodular construction in which whole ampliers are built into the space of what only yesterday was one transistor we can include all the Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2119 ampliers needed for quite an elaborate synthetic animal. We are now envisaging the construction of a creature which instead of looking as the original did, like a rather large and clumsy tortoise, resembles more closely a small eager, active and rather intelligent beetle. There seems no limit to which this miniaturisation could go. Already designers are thinking in terms of circuits in which the actual scale of the active elements will not be much larger, perhaps even smaller, than the nerve cells of the living brain itself. This opens a truly fantastic vista of exploration and high adventure. . . Walter (1968) We do not have to look far to see that things have progressed in the directions he foresaw. The behaviour-based principles with which he would surely be in sympathy have been applied to more complex entities such as the humanoid robots COG and Kismet (Brooks 2002); their mimicry of life is certainly `valid and illuminating. The concept of the `small, eager, active and rather intelligent beetle matches several of the products of Brookss Articial Insect Lab at MIT; no doubt we should also include the widely used commercial research robots such as the Kheperay in this category. As for the scale of `the active elements, modern semiconductor manufacturing techniques routinely produce transistors three orders of magnitude smaller than typical neurons, and so we are well placed to undertake the `exploration and high adventure he anticipated, even if there are as yet few signs of it in the present state of robotics. In modern usage, the word `legacy has acquired a new meaning: a legacy system is something left over from a previous technological generation, something obsolete that we would rid ourselves of if only we could. Some might be tempted to dismiss Walters tortoise project as a mere historical quirk|a legacy system with no real contemporary relevance. This is certainly a mistaken attitude. His work constitutes a real legacy in the traditional sense: something built up by a previous generation, and preserved and passed on so that it can benet a new generation. Twenty-ve years after his death, his uniquely personal approach still repays the eort made to understand it, and his thoroughgoing minimalism is a refreshing and perhaps necessary counter to the facile complexity oered by todays technology. The author is grateful to the anonymous referees for their helpful comments. Much of this work could not have been undertaken without the generous help and technical assistance of Mr W. J. Warren and the support of Dr S. Butler, Scientic Director of the BNI. Dr R. Cooper, Mrs C. Go, Dr Rhodri Hayward and Professor Harold Shipton assisted with documentary and pictorial sources. Allan Jones of the Open University provided details of Walters broadcast talks. The author thanks the late Nicolas Walter and the BNI for permission to quote from Walters unpublished notes. References Ashby, W. R. 1956 An introduction to cybernetics. London: Chapman and Hall. Bates, J. 1949 Letter from John Bates to Grey Walter 7/7/1949. In the archives of the Ratio Club, Wellcome Institute Library, London. y See the Cyberbotics s.a.r.l. Khepera website at http://www.cyberbotics.com/products/robots/ khepera.html. Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from 2120 O. Holland Bates, J. 1952 Manuscript for BMJ (the date is provisional). In the archives of the Ratio Club, Wellcome Institute Library, London. Bonabeau, E., Dorigo, M. & Theraulaz, G. 1999 Swarm intelligence: from natural to articial systems. Santa Fe Institute Studies in the Science of Complexity. Oxford University Press. Brooks, R. A. 2002 Robot: the future of esh and machines. London: Penguin. Cooper, R. & Bird, J. 1989 The Burden: fty years of clinical and experimental neuroscience at the Burden Neurological Institute. Bristol, UK: White Tree Books. Cordeschi, R. 1987 Purpose, feedback and homeostasis: dimensions of a controversy in psycho- logical explanation. In Studies in the History of Psychology and the Social Sciences. 4 (ed. S. Bem & H. Rappard). Leiden: Psychologisch Institut van der Rijksuniversiteit. Cordeschi, R. 2002 The discovery of the articial: behaviour, mind and machines before and beyond cybernetics. Dordrecht: Kluwer. de Latil, P. 1953 La pensee articielle. Paris: Gallimard. de Latil, P. 1956 Thinking by machine: a study of cybernetics (trans. Y. M. Golla). London: Sidgwick and Jackson. George, F. 1956 Robots on the march. John Bull, 5 May 1956. Hayward, R. 2001a Our friends electric: mechanical models of the mind in post-war Britain. In Psychology in Britain: essays and personal reminiscences (ed. G. Bunn, A. D. Lovie & G. Richards). London: Science Museum and British Psychological Society. Hayward, R. 2001b The tortoise and the love-machine: Grey Walter and the politics of electro- encephalography. Sci. Context 14, 615{641. Holland, O. E. 1996 Grey Walter: the pioneer of real articial life. In Articial life. V (ed. C. G. Langton & K. Shimohara), pp. 33{41. MIT Press. Holland, O. E. 2003 The rst biologically inspired robots. Robotica 21, 351{363. Morley, D. W. 1950 New light on the brain. Picture Post, 25 March 1950, pp. 18{23. Pickering, A. 2002 Cybernetics and the mangle: Ashby, Beer and Pask. In La reconguration des sciences pour laction dans les annees 1950 (ed. A. Dahan & D. Pestre). Paris: Presses de lEHESS. Pincher, C. 1949 Elmer and Elsie|they see, feel, move and even feed. Daily Express, 13 Decem- ber 1949. Shipton, H. W. 1977 Obituary: W. Grey Walter. Electroencephalogr. Clin. Neurophysiol. 43, iii{iv. Walter, N. 1990 William Grey Walter: personal notes for DNB. Nicolas Walter, April 1990 (typescript). BNI Papers, Science Museum. Walter, N. 1995 Letter to Owen Holland, 22 March 1995. BNI Papers, Science Museum. Walter, W. G. 1947 Letter from Grey Walter to Professor Adrian, 12 June 1947. BNI Papers, Science Museum. Walter, W. G. 1948 The enchanted loom. Typescript of radio broadcast (BBC Third Programme, 23 April 1948). BNI Papers, Science Museum. Walter, W. G. 1950a An imitation of life. Scient. Am. 182, 42{45. Walter, W. G. 1950b Electronische schildpadden met menselijke eigenschappen. (Electronic tur- tles with human characteristics.) De Internationale Echo, vol. 48, pp. 208{211. Holland. Walter, W. G. 1951 A machine that learns. Scient. Am. 185, 60{63. Walter, W. G. 1953a The living brain. London: Duckworth. Walter, W. G. 1953b Accomplishments of an artefact. Typescript (the date and attribution are provisional). BNI Papers, Science Museum. Walter, W. G. 1956a Further outlook. London: Duckworth. Walter, W. G. 1956b Transcript of rst discussion. In Discussions on child development (ed. J. M. Tanner & B. Inhelder), vol. 2, pp. 21{74. London: Tavistock Publications. Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from The legacy of Grey Walter 2121 Walter, W. G. 1957 Summary for the press of a lecture given at the Graduate Center for Psychotherapy, New York. Typescript. BNI Papers, Science Museum. Walter, W. G. 1960 Notes on operation: Machina Speculatrix. Typescript (the date is provi- sional). BNI Papers, Science Museum. Walter, W. G. 1967 William Grey Walter, his Trumpet Voluntary (Solo passages underlined.). Typescript (marked `sent to Dr Pampiglione 2.2.67). BNI Papers, Science Museum. Walter, W. G. 1968 The future of Machina speculatrix. Typescript for an unknown foreign- language publication; text sent to translator ca. 1968. BNI Papers, Science Museum. Wiener, N. 1948 Cybernetics. Wiley. Phil. Trans. R. Soc. Lond. A (2003) on May 24, 2011 rsta.royalsocietypublishing.org Downloaded from