The Cybernetic Brain

by Andrew Pickering

  It sounded like an ideal critical experiment. I cannot remember what exactly the reinforcement arrangements were, but the cell already had them in place in order to study the rate of adaptation to input changes, and we had created various gaps in the filigree by now.24 And so it was that two very tired young men trailed a microphone down into Baker Street from the upstairs window, and picked up the random noise of dawn traffic in the street. I was leaning out of the window, while Gordon studied the cell. "It's growing an ear," he said solemnly (ipsissima verba).

  A few years later Gordon was to write [Pask 1960b, 261]:

  We have made an ear and we have made a magnetic receptor. The ear can discriminate two frequencies, one of the order of fifty cycles per second and the other of the order of one hundred cycles per second. The "training" procedure takes approximately half a day and once having got the ability to recognize sound at all, the ability to recognize and discriminate two sounds comes more rapidly. . . . The ear, incidentally, looks rather like an ear. It is a gap in the thread structure in which you have fibrils which resonate at the excitation frequency.

  This, then, was the truly remarkable feature of Pask's chemical computers. One way to put it is to note that the "senses" of all the cybernetic machines we have discussed so far were defined in advance. At base they were sensitive to electrical currents, and at one remove to whatever sensors and motors were hooked up to them. Pask's chemical computers, however, acquired new senses which were not designed or built into them at all—hence "growing an ear," but also acquiring a sensitivity to magnetic fields (a quite nonhuman "sense") in other experiments. If the homeostat, say, could adapt to new patterns within a fixed range of input modalities, Pask's chemical computers went decisively beyond the homeostat in searching over an open-ended range of possible modalities. If we think of the chemical computers as model brains, these brains were, at least in this one sense, superior to human brains, which have not developed any new senses in a very long time.

  One possibly confusing point here is that Pask and Beer trained the computer to acquire the faculty of hearing and responding to sound—as if they somehow inserted the sense of hearing into the computer even if they did not explicitly design it in from the start. But to think that way would be to miss a key point. One should imagine the computer not in the training situation but in use—as hooked up to a factory, say—in which the coupled system was running to equilibrium. In that event, in its trial reconfigurations, a soundsensitive thread structure might come into existence within the chemical computer and find itself reinforced in its interactions with the factory in the absence of any intervention from the experimenter whatsoever. In this scenario, the machine could thus genuinely evolve new senses in its performative interactions with its environment: hearing, a feeling for magnetic fields, or, indeed, an indefinite number of senses for which we have no name. And, as both Beer and Cariani have emphasized, no machine that could do this had been built before—or since, unless very recently: "It could well have been the first device ever to do this [develop a new sense], and no-one has ever mentioned another in my hearing" (S. Beer 2001, 555).

  The Epistemology of Cybernetic Research

  A CYBERNETIC HYPOTHESIS IS SOMETHING WHICH THE . . . CYBERNETICIANS DIG FROM THE OUTLANDISH SOIL OF THEIR ORGANIC ASSEMBLAGES.

  GORDON PASK,"ORGANIC CONTROL AND THE CYBERNETIC METHOD" (1958, 171)

  BUT, MORE IMPORTANT THAN THIS IS THE QUESTION OF WHETHER, IN SOME SENSE, THE NETWORK IS LIKE MY IMAGE OF MYSELF BEING A MANAGER (THIS PART OF THE INTERVIEW IS DIFFICULT, FOR THERE IS NO VERBAL COMMUNICATION . . . ). ON THIS TEST, I SHALL ACCEPT THE NETWORK IF AND ONLY IF IT SOMETIMES LAUGHS OUTRIGHT.

  GORDON PASK,AN APPROACH TO CYBERNETICS (1961, 113)

  Throughout this book I have discussed the performative epistemology that I associate with cybernetics, the idea that representational knowledge is geared into performance, a detour away from and back to performance. It might be useful to come at this topic from a different angle here, via Pask's own epistemological reflections on his work with chemical computers, where he articulated a distinctive understanding of the "cybernetic method." Pask concluded his essay "Organic Control and the Cybernetic Method" (1958) by discussing cybernetics as a distinctive form of practice.He first defines a generic figure of an "observer" as "any person or appropriate mechanism which achieves a well defined relationship with reference to an observed assemblage" (Pask 1958, 172).25 He then makes a distinction between what I call two "stances" that the observer can take up with respect to the object of inquiry (172–73):

  Any observer is limited by a finite rate at which he or it may make decisions. Since the limit exists we shall distinguish a scientific observer who minimises interaction with an observed assemblage and a participant observer who, in general, tries to maximise his interaction with an assemblage. If observers were omniscient there would be no distinction. A scientific observer decides whether or not the evidence of an observation leads him to accept each of a finite set of hypotheses, and may, as a result, determine his next observation. Since he is minimally associated with the assemblage he may determine his next observation precisely. . . . A scientific observer seeks to confirm as many hypotheses as possible.

  Leaving aside questions of confirmation versus falsification, here we recognize a standard stereotype of the hypothesis-testing scientist. Note that on this stereotype, the scientist's access to matter passes through representations: a hypothesis is a verbal formulation—"If I do X then Y will happen"—itself embedded in all sorts of statements, theoretical and empirical. On the other hand (173; my italics),

  a cybernetician is a participant observer who decides upon a move which will modify the assemblage and, in general, favour his interaction with it. But, in order to achieve interaction he must be able to infer similarity with the assemblage. In the same way cybernetic control mechanisms must be similar to the controlled assemblage. The development of this similarity is the development of a common language.. . . [The cybernetician needs] to adopt new languages,in order to interact with an assemblage. [There is] an ignorance on the observer's part, about the kind of enquiry he should make. A common language is a dynamic idea, and once built up must be used. Thus if a conversation is disturbed it must be restarted, and one of the structured regions we have discussed must continually rebuild itself. . . . A cybernetician tries, by interaction, to bring about a state of a macrosystem which exhibits a consistent pattern of behaviour that may be represented by a logically tractable analogy.

  There is, I think, an original philosophy of science adumbrated in these few sentences, which deserves a brief exegesis. Most important, the emphasis is on performative interaction with the object to be known, modification which might promote further interaction. One can recall here the earlier discussion of chemical computers and of the manager coming to terms with the controller in the same way as the controller comes to terms with the factory—by each interfering performatively with the other until some mutually acceptable, not pregiven, equilibrium is found. What Pask adds is that cyberneticians learn in general about their objects in just the same way: they interfere with them as much as possible in an exploratory fashion to see what they will do, with each observation provoking new, situated interferences. But what, then, of Pask's references to language and analogy? Does this return us to the hypothesistesting model he just associated with "science"? No, because, first, Pask does not regard language as a given and stable medium in which hypotheses can be formulated and judged. The cybernetician does not know the appropriate terms—the language, the relevance conditions—for describing his or her object in advance; they have to be discovered in interaction with that object. Further, we know we have found suitable terms (not a true description) when we use them to construct a model of the object which enables us to understand its behavior when subject to additional interferences. Cybernetic interference produces new languages in which to address and interrogate its object.

  But second, as in his usage
of "conversation," it seems clear that Pask's sense of "language" is not necessarily verbal or representational in the usual sense. The model that grows in the cybernetician's interaction with some object might be nonverbal—as in the "model" of the factory that builds up in the chemical computer as it comes into equilibrium with the factory—and it may be a material object which bears no resemblance at all to the thing modelled—a thread structure does not look like Bill Smith happily making ornamental plaques; the homeostat does not look like a brain. Or it might be a conceptual construct—one of Stafford Beer's models of the Chilean economy, for example. All that matters is that the model facilitates continued commerce with the object itself.

  Where does this leave us? Clearly, Pask's account of the cybernetic method indeed points to a performative epistemology. Second, we can think of his chemical computers as a vivid act of epistemological theater. The thread structures stage for us the idea that knowledge (of the factory, of the manager) need not take a representational form. Third, Pask's contrast between the scientific and the cybernetic methods warrants some brief elaboration. One could sum up the findings of the last twenty years and more of science studies as the discovery that real scientists are more like Paskian cyberneticians than his stereotype of them. They, too, struggle open-endedly with their objects and invent new languages and models to get to grips with them (Pickering 1995). But we need to think here of Pask's two kinds of observer and their different stances with respect to the world. Such open-ended struggles indeed happen in scientific practice, but this is thrust into the background of the modern sciences and forgotten, effaced, in the "hypothesis testing" model of science. And the argument I having been trying to make throughout this book—albeit with an emphasis on ontology rather than epistemology—is that these stances are consequential. Although we are all in the same boat, they make a difference: cybernetics, in its practices and in its products—chemical computers that develop new senses being a striking example of the latter—is different in its specificity from the modern sciences. This is, of course, precisely Pask's argument rehearsed earlier in the context of aesthetics and Musicolour— cybernetics suggests an unfamiliar and productive stance in science, as well the arts, entertainment, and teaching.26

  CAs, Social Science, and F-22s

  Pask discontinued his work on chemical computers in the early 1960s, and we should think about this along the lines already indicated in the previous chapter. Like Beer's biological computers, Pask's chemical ones were a valiant attempt at radical innovation undertaken with no support, more or less as a hobby, typified by "two very tired young men" trailing a microphone out of a window as the sun came up over Baker Street. We could also note that even within the cybernetics community, no one, as far as I know, sought to emulate and elaborate Pask's efforts—this in contrast, for example, to the many emulators of Walter's tortoises. Meanwhile, from the later 1950s onward typing trainers and teaching machines held out more immediate prospects of paying the rent. But one spin-off from Pask's research is interesting to follow briefly.

  In an attempt to understand the dynamics of his threads, Pask undertook a series of numerical simulations of their behavior, which involved a form of idealization which is now very familiar in the sciences of complexity: he represented them schematically as two-dimensional cellular automata (chap. 4). In these simulations the dish of ferrous sulphate was replaced by a two-dimensional space, with "automata" residing at the intersections of a Cartesian grid. These automata evolved in discrete time steps according to simple rules for persistence, movement, reproduction, and death according to their success in exploiting a finite supply of "food." The early chemical-computer publications reported "hand simulations" of populations of automata, and in 1969 Pask reported on a set of computer simulations which prefigured more visible developments in work on cellular automata and artificial life a decade or two later (Pask 1969a).27 Interestingly, however, Pask framed his account of these computer simulations not as an exploration of chemical computing but as a study of the emergence of norms and roles in social systems. Over the past decade there has been something of an explosion of social-science research on computer simulations of populations of automata.28 It is not clear to me whether Pask's work was a formative historical contribution to this new field or whether we have here another instance of independent reinvention. What is clear is that this contemporary work on social simulation, like Pask's, can be added to our list of examples of ontology in action.29

  Pask's interest in automata and simulation will reappear below in his work in architecture. But let me close this section with two remarks. First, following Cariani (1993, 30), we can note that the move from chemical computers to numerical simulation was not without its cost. The chemical computers found their resources for developing new senses in their brute materiality; they could find ways to reconfigure themselves that had not been designed into them. Pask's simulated automata did not have this degree of freedom; their relevance conditions were given in advance by the programs that ran them. No doubt this, too, had a bearing on the erasure of the chemical computers even from the consciousness of cybernetics.

  Second, while one should beware of exaggeration, we can observe that cybernetic controllers are back in the news again. "Brain in a Dish Flies Plane" (Viegas 2004) is one of many media reports on a project strongly reminiscent of Pask, Beer, and even Ashby (who, we recall, discussed the virtues of homeostatic autopilots). In work at the University of Florida, rat neurons (in the style of Beer's biological computers) were grown in a dish and connected into the world via a sixty-channel multielectrode array (à la Pask). When this device was hooked up to an F-22 fighter jet flight simulator, "over time, these stimulations modify the network's response such that the neurons slowly (over the course of 15 minutes) learn to control the aircraft. The end result is a neural network that can fly the plane to produce relatively stable straight and level flight." Another version of the philosopher's apocryphal brain in a vat, though not so apocryphal any more, and robustly connected into the world of performance rather than seeking to represent a world of which it is not a part.30

  The Arts and the Sixties

  We have traveled a long way from Musicolour to chemical computers via typing trainers and teaching machines. For the remainder of this chapter I want to return to Pask's work in the theater, the arts, and architecture, picking up the story in the early 1960s (that is, in work that ran in parallel to his work on trainers and teaching machines). I am interested in three projects in particular: Pask's plans for a cybernetic theater; his robotic artwork, the Colloquy of Mobiles; and his contributions to architecture, beginning with the London Fun Palace. These projects are interesting in themselves as fresh instances of ontology in action, and they are also worth contemplating as yet more instances of crossovers from cybernetics to the distinctive culture of the 1960s. At an early stage in their careers, the Rolling Stones were apparently "roped in" to try out the adaptive machines at System Research (Moore 2001, 770).31 Pask's student Ranulph Glanville had a fleeting association with Pink Floyd, who lived nearby, and built a piece of electronic equipment for them—a ring modulator; he also did a sound mix for Captain Beefheart (Glanville, email, 16 August 2005). More consequential than such contacts with the iconic bands of the sixties, however, were Pask's contacts dating back to his undergraduate days with Cedric Price (Price 2001, 819), one of Britain's leading postwar architects, and with the radical theater director Joan Littlewood. If we pursued Pask's projects chronologically, the order would be Fun Palace, cybernetic theater, Colloquy of Mobiles, but for the purposes of exposition it is better to begin with the theater and to end with architecture.32

  Cybernetic Theater

  Joan Littlewood (1914–2002), the founder of the postwar Theatre Workshop in Britain and of the Theatre Royal in Stratford, London, writer and producer of Oh, What a Lovely War! and many other plays that marked an era, occupies an almost legendary place in the history of British theater (Ezard 2002).33 She recalled
that she had heard stories about Pask in the 1950s and that he had "flitted across my life from time to time like a provocative imp. . . . He had some idea of what we were up to. I wrote to him a couple of times. He seemed to be as de trop in English society as we were. They simply did not know how to use him—the Yanks did." The reference to the Yanks is an exaggeration, but, as usual for our cyberneticians, de trop sounds about right. Littlewood and Pask first met in person, presumably in the late 1950s, at System Research, "a normal looking house, from the outside, but we were standing in a labyrinth of wires, revolving discs of cardboard, cut from shredded wheat packets, little pots and plugs, while through it all a small, perfectly normal baby girl [Hermione Pask] was crawling in imminent danger of being electrocuted from the looks of things, though she was cooing contentedly" (Littlewood 2001, 760).

  Here is Littlewood's recollection of a subsequent conversation with Pask (Littlewood 2001, 761): "I told him about two Red Indians taking their morning coffee in the Reservation Cafe and discussing last night's film. 'I thought we were going to win till that last reel,' said one. 'It would be fun,' I said, 'if the Red Indians did win for a change.' This caused a spark. He knew that I worked with inventive clowns. 'We could have a set of different endings,' he said. 'At least eight and the audience could decide which they wanted,' 'How?' 'By pressing a button attached to their seat, quite simple.' " The upshot of this conversation was a thirty-page 1964 document entitled "Proposals for a Cybernetic Theatre," written by Pask on behalf of Theatre Workshop and System Research. In it Pask describes in considerable detail how audience and actors could be coupled together via feedback loops in determining the substance of any specific performance. Pask's basic idea here was that of a new kind of play, which would retain certain set-piece structural elements, specified in advance, but which would include alternative routes of plot development between the set pieces in an entailment-mesh-like structure, including the possibility that the trajectories might redefine the significance of the fixed elements.34 During the performance, members of the audience could signal their identification with one or another of the principal actors (designated Aand Bin fig. 7.14). At specified branchpoints, the audience could also use levers to advocate different choices of action for their chosen character, drawing upon both their understanding of how the play had developed thus far and also upon "metainformation" on their character's thinking at this point, developed in rehearsal and provided in real time by "interpreters" (alpha and beta in fig. 7.14) via headphones or earpieces. The interpreters in turn would then use hand signals, or perhaps radio, to let the actors know their supporters' inclinations, and the play would proceed accordingly. Depending on how the play developed from these branch points, the audience was free to change identifications with actors, to make further plot decisions, and so on.