The Cybernetic Brain

by Andrew Pickering

  41. Besides his journal and the published literature, Ashby also raised the topic of possible links between cybernetics and psychiatry at the Ratio Club. The Bates papers at the Wellcome Archive in London include a three-page typescript by Ashby entitled "Cybernetics and Insanity" (Contemporary Medical Archives Centre, GC/179/B.2a), cataloged as "possibly prepared for Ratio Club meeting 1952." It develops the idea mentioned in the following note that the brain consists of a hierarchy of homeostatic regulators and that mental illness might be identified with malfunctions at the highest-but-one level. There are also two lists of possible topics for discussion at the Ratio Club sent earlier by Ashby to Bates (GC/179/B5), which include "3. To what extent can the abnormalities of brains and machines be reduced to common terms?" (26 April 1950) and "26. The diagnosis and treatment of insanity in machines" (18 February 1950).

  42. The paper was read to a Royal Medico-Psychological Association meeting in Gloucester (presumably at Barnwood House) in July 1953. For the sake of completeness I can mention two more connections that Ashby made from cybernetics to psychiatry. First, the essay suggests a quite novel general understanding of mental illness. Ashby supposes that there exist in the brain homeostatic regulators ancillary to the cortex itself, and that mental pathologies might be due to faults in the ancillary systems. He argues that if that were the case, "we would see that the details of the patient's behaviour were essentially normal, for the details were determined by an essentially normal cortex; but we would find that the general tenor was essentially abnormal, a caricature of some recognisable temperament. . . . Thus we might see the healthy person's ability to think along new and original lines exaggerated to the incomprehensible bizarreness of the schizophrenic" and likewise for the "maniac" and the "melancholic" (123). Second, Ashby returns to the question of the organic basis for mental illness, but from a distinctively cybernetic angle. This hinged, in the first instance, on the question of the brain's essential variables. Ashby had already suggested that biological limits on essential variables were set by heredity, and here he supposed that sometimes heredity would go wrong. He mentioned a child born unable to feel pain, who thus "injures himself seriously and incessantly," and imagined that "the mental defective who is selfmutilating . . . may be abnormal in the same way" (121).

  43. One thinks here of the "sensitive dependence on initial conditions" later thematized in complex systems theory.

  44. Ashby discontinued his specifically psychiatric research after leaving England in 1960, but he continued to contribute to the psychiatric literature after moving to Urbana. See, for example, Ashby (1968c). These later works uniformly seek to educate psychiatrists about cybernetics and, especially, information theory, and the interest in ECT disappears completely. Ashby was happy to talk about psychiatric therapy, as in "The theory of machines . . . may well provide the possibility of a fully scientific basis for the very high-level interactions between patient and psychotherapist" (1968c, 1492). On the other hand, one can still detect understated connections between the degree of connectivity within the brain and lobotomy. Having discussed his student Gardner's discovery of a threshold in "connectance," he remarks that "it is obvious that a system as complex and dynamic as the brain may provide aspects at which this 'mesa' phenomenon may appear, both in aetiology and therapy. There is scope for further investigation into this matter, both in its theory and its application. . . . All the studies of the last twenty years . . . show that systems should be only moderately connected internally, for in all cases too rich internal connection leads to excessive complexity and instability. The psychiatrist knows well enough that no one can produce associations so quickly or so wide-ranging as the acute maniac; yet his behaviour is inferior, for knowing what associations to avoid, how to stick to the point, is an essential feature for effective behaviour" (1968c, 1494, 1496). As it happens, the very last entry Ashby made in his journal was about schizophrenia: he noted that in an article in the British Journal of Psychiatry,vol. 120 (1972), "Schizophrenics were classified by handedness . . . dominant eye . . . and six [sex]" (8 March 1972, pp. 7188–89).

  45. There are many more discussions of texts in Ashby's journal than records of conversations, though see below on Mrs. Bassett.

  46. This observation reflects back on Ashby's understanding of the prototypical four-homeostat setup. While that setup can help us imagine a fully symmetric performative relation between entities and their world, the present discussion suggests that Ashby's basic model was one in which the brain homeostat was uniselector controlled and adaptive but the other three homeostats had their properties fixed: a brain that adapts to the world but not vice versa. He mentions somewhere that it is a good thing that cars and roads do not change their properties when we are learning to drive, otherwise the process would be even more fraught, and it is hard to disagree with that. On the other hand, it is integral to the history of science that the material world can perform in new and surprising ways when subjected to novel trials in the laboratory, and this seems to me to be modelled better by a symmetric field of adaptive homeostats, each capable of taking on new states in response to novel stimuli. More generally, biological species, including the human race, do not typically take the material world as given; they transform it instead, and, again, such transformations may elicit new worldly performances—think of global warming.

  47. We might think again about the standard origin story, of cybernetics from warfare. Against that, we could observe that Ashby's first cybernetic publication appeared in 1940, eleven years before he got Clausewitz out of the public library. From another angle, Ashby's discussion of wrestling and the throatgripping hands strike me as wonderful quick analysis of why our brave leaders prefer to bomb their latest enemies from high altitude (at night, with visionenhancing technologies, after destroying their radar) rather than fighting them in the street.

  48. While most of Ashby's discussions of state-determined machines can be seen as integral to the development of his cybernetics and psychiatry, this letter to Naturewas part of a deliberate strategy to draw attention to his work, recorded in his journal in early June 1944 (p. 1666): "My plan is to write articles on political & economic organisation to try and make a stir there, knowing that then I can say that it is all based on my neuro-physiology. Another line is to watch for some dogmatic statement which I can flatly contradict in public, the bigger the authority who makes it the better." Of course, that he even thought about applying his mathematical analysis to economics is enough to illustrate the instability of its referent.

  49. An Introduction to Cybernetics displays no interest at all in psychiatry, apart from a remarkable paragraph at the end of the preface. After thanking the governors of Barnwood House and Fleming, its director, for their support, Ashby continues: "Though the book covers many topics, these are but means; the end has been throughout to make clear what principles must be followed when one attempts to restore normal function to a sick organism that is, as a human patient, of fearful complexity. It is my faith that the new understanding may lead to new and effective treatments, for the need is great" (1956, vii).

  50. Thanks to the efforts of Ashby's grandson, Michael Ashby, An Introduction to Cyberneticsis now online as a PDF file at the Principia Cyberneticawebsite at http://pcp.vub.ac.be/IntroCyb.pdf. The scanning process has left its marks in some easily spotted typographical errors.

  51. The book is indeed written as a textbook, including many practice questions, with answers at the back. The question for which I have a particular fondness is example 2 in section 2.9 (1956, 15): "In cricket, the runs made during an over transform the side's score from one value to another. . . . What is the cricketer's name for the identical transformation?" The Englishness of it all. I wonder how many readers of the Bulgarian translation, say, or the Japanese, would have known that the answer is "a maiden over" (1956, 274), and how many promising careers in cybernetics ended abruptly with that question.

  52. On 15 October 1951, Ashby made some notes on extracts
from a symposium on information theory held in London in September 1950. "A most important (but obscure) contribution by Shannon. . . . I cannot, at present, follow it. I must find out more about this sort of thing" (p. 3510).

  53. Recall how readily hyperastronomical numbers were liable to appear in the kinds of combinatorics that cybernetics led into, as in Ashby's estimates of how long a multielement homeostat setup would take to reach equilibrium. On Bremermann's limit see, for example, Ashby (1969). Ashby mentions there his work on deriving a measure of "cylindrance," a number that characterizes the extent to which systems of interconnected variables can be decomposed into independent sets, and thus how likely our thinking on such systems is to confront Bremermann's limit. The problems of DAMS were clearly still on Ashby's mind in the mid-1960s; Ashby says of cylindrance that "it treats not only the fairly obvious case in which the relation consists of kwholly independent subrelations but also the much more interesting case in which the whole relation has something of the simplicity of a k-fold division while being in fact still connected. (An elementary example is given by a country's telephone communications, in that although all subscribers are joined potentially to all, the actual communications are almost all by pairs)" (Ashby 1968a, 74).

  54. I cannot resist one last observation on An Introduction to Cybernetics,which is that in it Ashby talked himself yet again into the problem of accumulating adaptations, which we discussed earlier. At the end of the penultimate chapter (Ashby 1956, 260–62), he turned to the question of the "duration of selection"—how long would take it take to pick the right element out of an enormous array of possibilities? This was an immediate translation of the question in Design for a Brain of how long it would take a multihomeostat setup to find its equilibrium state. Again in An Introduction to Cyberneticshe concluded that a direct search would take a hyperastronomical time and remarked that, in contrast, the search would be quick if the system were "reducible," the equivalent of letting each homeostat come to equilibrium independently of the others. Of course, the problem of getting around this assumption of absolute independence was the rock on which DAMS was already, by 1956, foundering and which eventually the gating mechanism of the second edition of Design for a Brainmagically conjured away. But An Introduction to Cyberneticswas an introductory text, so after a simple discussion of reducibility Ashby felt at liberty to remark: "The subject must be left now, but what was said in Design for a Brainon 'iterated systems,' and in the chapters that followed, expands the thesis" (1956, 262). The circle is squared; the unfortunate reader is sent off to wander endlessly in a hall of mirrors.

  55. This was earlier argued by Kenneth Craik (1943). Having read his book, Craik was, as mentioned earlier, one of the first people Ashby wrote to about his nascent cybernetics.

  56. A model is a representation of its object, so one wonders what sort of a mechanism Ashby conceived for representation. If one thinks of the homeostat and DAMS, this is a puzzle, but Ashby's thinking here was distinctly undramatic— he was happy to conceive representation in terms of programs running on digital computers. "The Big Computer—how much difference will be made by its advent may well be a matter of opinion. I rather lean to that of Martin Shubik who suggested that its impact may eventually be as great as that of the telescope, opening up entirely new worlds of fact and idea (after we have learned to use it appropriately)" (Ashby 1966, 90; see also Ashby 1970). Even so, Ashby had a distinctive understanding of how computers might be used. Ashby (1968c) mentions Illiac IV as a parallel processing machine which will need a new style of programming, and continues: "Little, however, is being done in the direction of exploring the 'computer' that is brainlike in the sense of using nearly all its parts nearly all the time" (1493). The discussion that follows includes habituation, which points to DAMS as a referent for the latter style of machine.

  57. Ashby's An Introduction to Cybernetics culminates with the epistemology of selection: " 'Problem solving' is largely, perhaps entirely, a matter of appropriate selection. . . . It is, in fact, difficult to think of a problem, either playful or serious, that does not ultimately require an appropriate selection as necessary and sufficient for its solution" (1956, 272) The book ends within two paragraphs, in a discussion of intelligence amplification, which Ashby discusses elsewhere in terms of the use of a random source to generate a list of possibilities which can then be checked automatically against some well-defined goal.

  58. To put some flesh on this critique, I can note that in his 1970 essay on models, discussed above, Ashby pondered the question of finding the right model for some system in much the same terms as he had with Walker, invoking many of the same historical examples: Newton, Gauss, Mozart. His pièce de résistance, however, was the nineteenth-century mathematician Sir William Rowan Hamilton (Ashby 1970, 109), about whose construction of the "quaternion" system I have, as it happens, written at length (Pickering 1995, chap. 4). My argument there was that Hamilton visibly shifted the "goal" of his research in midstream, starting with the intention of constructing a three-place analogue of complex numbers but ending up with a four-place one instead, and made upall sorts of mathematical possibilities (including ones like noncommuting variables, explicitly disallowed in the mathematics of the day) and tried them out to see what worked. Ashby and Walker would have to regard the former as blatant cheating and lack any language to conceptualize the latter.

  59. This biographical information is taken from Alexander's website: www. patternlanguage.com. Besides architecture, Alexander's work has also been very influential in software design, but I will not follow that here—see the "wiki" link at the website just cited. I thank Brian Marick for an introduction to this aspect of Alexander's work.

  60. Alexander's notions on design match Ashby's thoughts on evolutionary design and the blueprint method, but the latter were unpublished, and Alexander arrives at them from a different route.

  61. One can find instances of Ashby talking about arrays of lightbulbs as a way of conjuring up the horrendous numbers one can generate by combinatorics, but all the examples that I have yet found date from after Alexander's book was published. See, for example, Ashby (1970, 99).

  62. As usual, bringing the ontology down to earth involved adding something by way of specification: in this instance, the assumption that the matrix of interconnecting misfits indeed falls into weakly coupled sectors. Alexander thus "solved" the problem of complexity by fiat. DAMS, in contrast, was intended to solve this problem for itself. As analogues to the brain, Alexander's designs were not as adaptable as Ashby had hoped DAMS would be.

  63. I am grateful to Stuart Kauffman for conversations about his life and work when he visited the University of Illinois in March 2004.

  64. Ashby, von Foerster, and Walker (1962) had earlier discussed the instability of a network of idealized neurons. They showed analytically that the network would settle down either to zero or full activity depending upon the intensity of the input stimulation. In a further continuation of this work, Gardner and Ashby (1970) used computer calculations to explore the stability of linear networks as a function of the density of their interconnection and discovered a discontinuity with respect to this variable: below some value (dependent on the total number of elements in the system) the system would be stable; above that it would be unstable. Gardner's research was for a master's thesis at the University of Illinois, cited in Ashby (1968b).

  65. Waldrop (1992, chap. 3) covers the trajectory of Kauffman's research career. He became interested in the problematic of genetic circuits and embryogenesis as a graduate student in 1963 on reading the work of Jacob and Monod on gene-switching, and tried to explore the behavior of small, idealized, genetic circuits in pencil-and-paper computations, before paying for the computer simulations in which the results discussed below first emerged. In 1966, he contacted the doyen of American cybernetics, Warren McCulloch, who was well known for his work on neural networks (dating from McCulloch and Pitts's foundational paper in 1943, and identical to genetic nets
at a binary level of abstraction). At McCulloch's invitation, Kauffman spent the period from September to December of 1967 with McCulloch's group at the MIT Research Laboratory of Electronics (living at McCulloch's house). Kauffman's first publication on genetic networks was written jointly with McCulloch as an internal report of that laboratory: Kauffman and McCulloch (1967). Kauffman's first refereed publications appeared in 1969: Kauffman (1969a, 1969b). In 1969 at the University of Chicago, "Kauffman heard about Ashby's Design for a Brain,and "I got in touch with him as soon as I found out about it" (Waldrop 1992, 121). The only citation of prior work on networks in Kauffman (1969b) is of Walker and Ashby (1966), and the paper thanks Crayton Walker (and others) for encouragement and criticism. In 1971 the American Cybernetics Society awarded Kauffman its Wiener Gold Medal for the work discussed here. As a sidelight on the continuing marginalization of cybernetics,Waldrop manages to tell this whole story without ever using the word "cybernetics" (and there is no entry for it in the index to his book). McCulloch is described as "one of the grand old men of neurophysiology—not to mention computer science, artificial intelligence, and the philosophy of mind" (Waldrop 1992, 113), and Ashby is an "English neurophysiologist" (120).

  66. For earlier and explicitly cybernetic work in this area, including an electrical model of cell differentiation, see Goldacre's presentation at the second Namur conference in 1958 (Goldacre 1960a, also Goldacre and Bean 1960). At the same meeting Goldacre presented the paper "Can aMachine Create aWork of Art?" based on the same electrical model (Goldacre 1960b). Goldacre's affiliation was to the National Cancer Hospital in London. For a continuing strand of self-consciously cybernetic biology, see the journal Biological Cybernetics(originally entitled Kybernetik).On the history of developmental biology in the twentieth century, see Keller (2002).