D. W.H. After some milnutes in the dark the rhythm is presents with the eyes open. Closing them does not alter the rhythm.
Figure 3.2.Alpha rhythms in the brain, showing the effect of opening and closing the eyes. Source: E. D. Adrian and B. H. C. Matthews, "The Berger Rhythm: Potential Changes from the Occipital Lobes in Man," Brain, 57 (1934), 355–85. (By permission of Oxford University Press.)
In 1939 Walter and Golla moved together to the newly established Burden Neurological Institute near Bristol, with Golla as its first director and Walter as director of its Physiology Department (at annual salaries of £1,500 and £800, respectively). The Burden was a small, private institution devoted to "clinical and experimental neuroscience" (Cooper and Bird 1989), and Walter remained there for the rest of his working life, building a reputation as one of the world's leaders in EEG research and later in research using electrodes implanted in the brain (rather than attached to the scalp).6 Walter's best-recognized and most lasting contribution to brain science was his discovery in the 1960s of contingent negative variation, the "expectancy wave," a shift in the electrical potential of the brain that precedes the performance of intentional actions. He was awarded the degree of ScD by Cambridge in 1947 and an honorary MD degree by the University of Aix-Marseilles in 1949.
Besides his technical work, in 1953 Walter published an influential popular book on the brain, The Living Brain, with a second edition in 1961, and in 1956 he published a novel, Further Outlook, retitled The Curve of the Snowflake in the United States.7 He was married twice, from 1934 to 1947 to Katherine Ratcliffe, with whom he had two children, and from 1947 to 1960 to Vivian Dovey, a radiographer and scientific officer at the Burden, with whom he coauthored papers and had a son. From 1960 to 1974 he lived with Lorraine Aldridge in the wife swap mentioned above (R. Cooper 1993; Hayward 2001a, 628). In 1970 Walter's research career came to an end when he suffered a serious head injury as a result of a collision between the scooter he was riding (at the age of sixty, let us recall) and a runaway horse. He was in a coma for a week, suffered serious brain damage, and never fully recovered. He returned to work at the Burden as a consultant from 1971 until his retirement in 1975 and died suddenly of a heart attack in 1976 (Cooper and Bird 1989, 60).
Walter's most distinctive contribution to cybernetics came in 1948, with the construction of the first of his robot tortoises. He was one of the founders of the Ratio Club, the key social venue for the British cyberneticians, which met from 1949 until 1955 (Clark 2002, chap. 3, app. A1). He was an invited guest at the tenth and last of the U.S. Macy cybernetics conferences in 1953 (Heims 1991, 286), and he was a member of the four-man scientific committee of the first meeting of the European counterpart of the Macys, the 1956 Namur conference—the First International Congress on Cybernetics—where he presided over section IV, devoted to "cybernetics and life."
The Tortoise and the Brain
How might one study the brain? At different stages of his career, Walter pursued three lines of attack. One was a classically reductionist approach, looking at the brain's individual components. Working within a well-established research tradition, in his postgraduate research at Cambridge he explored the electrical properties of individual neurons which together make up the brain. One can indeed make progress this way. It turns out, for example, that neurons have a digital character, firing electrical signals in spikes rather than continuously; they have a certain unresponsive "dead time" after firing; they have a threshold below which they do not respond to incoming spikes; they combine inputs in various ways. But if one is interested in the properties of whole brains, this kind of understanding does not get one very far. A crude estimate would be that the brain contains 1010 neurons and many, many more interconnections between them, and no one, even today, knows how to sum the properties of that many elements to understand the behavior of the whole. As Walter put it, "One took an anatomical glance at the brain, and turned away in despair" (1953, 50). We could see this as a simple instance of the problem of complexity which will appear in various guises in this chapter and the next: there exist systems for which an atomic understanding fails to translate into a global one. This is the sense in which the brain counted for Stafford Beer as an exemplary "exceedingly complex system."
Walter's second line of attack emerged on his move to London. His EEG work aimed at mapping the properties of the brain. What does the brain do? Well, it emits small but complicated electrical signals that are detectable by sensitive electronic apparatus. Such signals, both oscillatory (waves) and singular, were what Walter devoted his life to studying. This proved to be difficult. Other rhythms of electrical activity—the so-called beta, theta, and delta bands of brainwaves at frequencies both above and below the alphas—were discovered, but EEG readouts revealed the brain to be very noisy, and distinguishing correlations between inputs and outputs was problematic. Echoing the findings of Adrian and Matthews in 1934, Walter (1953, 90) observed that "very few of the factors affecting the spontaneous rhythms were under the observation or control of experimenter or subject. Usually only the effects of opening or closing the eyes, of doing mental arithmetic, of overbreathing and of changes in the blood sugar could be investigated. . . . The range and variety of methods were not comparable with the scope and sensitivity of the organ studied, and the information obtained by them was patchy in the extreme." The electric brain, one could say, proved more complex than the variables in terms of which researchers might hope to map it.8
We can return to Walter's EEG work at various points as we go along, but I can enter a couple of preliminary comments on it here. As ontological theater, it evidently stages for us a vision of the brain as a performative organ rather than a cognitive one—an organ that acts (here, emitting electrical signals) rather than thinks. Equally evidently, such a conception of the brain destabilizes any clean dualist split between people and things: the performative brain as just one Black Box to be studied among many.9 At the same time, though, as we will see shortly, Walter's ambition was always to open up the Black Box, in pursuit of its inner go. This is what I mean by referring to the hybrid quality of his cybernetics.
Walter's third line of attack on the brain was the one that I have talked about before: the classically scientific tactic of building models of the brain. The logic here is simple: if a model can emulate some feature of the system modelled, one has learned something, if only tentatively, about the go of the latter, its inner workings. As Roberto Cordeschi (2002) has shown, one can trace the lineage of this approach in experimental psychology back to the early years of the twentieth century, including, for example, the construction of a phototropic electric dog in 1915. The early years of cybernetics were marked by a proliferation of such models, including the maze-learning robots built by Claude Shannon and R. A. Wallace—which Walter liked to call Machina labyrinthea—and Ashby's homeostat (Machina sopora)(Walter 1953, 122–23), but we need to focus on the tortoise.10
The tortoises (or "turtles") were small electromechanical robots, which Walter also referred to as members of a new inorganic species, Machina speculatrix.He built the first two, named Elsie and Elmer, at home in his spare time between Easter of 1948 and Christmas of 1949. In 1951, a technician at the Burden, W. J. Warren—known as Bunny, of course—built six more, to a higher engineering standard (Holland 1996, 2003). The tortoises had two back wheels and one front (fig. 3.3). A battery-powered electric motor drove the front wheel, causing the tortoise to move forward; another motor caused the front forks to rotate on their axis, so the basic state of the tortoise was a kind of cycloidal wandering. If the tortoise hit an obstacle, a contact switch on the body would set the machine into a back and forth oscillation which would usually be enough to get it back into the open. Mounted on the front fork was a photocell. When this detected a source of illumination, the rotation of the front fork would be cut off, so the machine would head toward the light. Above a certain intensity of illumination, however, the rotation of the forks would normally be switche
d back on, so the life of the tortoise was one of perpetual wanderings up to and away from lights (fig. 3.4). When their batteries were low, however, the tortoises would not lose interest in light sources; instead, they would enter their illuminated hutches and recharge themselves.
The tortoises also executed more complex forms of behavior which derived from the fact that each carried a running light that came on when the tortoise was in search mode and went off when it locked onto a light. The running lights were originally intended simply to signal that a given tortoise was working properly, but they bestowed upon the tortoise an interesting sensitivity to its own kind. It turned out, for example, that a tortoise passing a mirror would be attracted to the reflection of its own light, which light would then be extinguished as the tortoise locked onto its image; the light would then reappear as the scanning rotation of the front wheel set back in, attracting the tortoise's attention again, and so on (fig. 3.5). The tortoise would thus execute a kind of mirror dance, "flickering, twittering and jigging," in front of the mirror, "like a clumsy Narcissus." Likewise, two tortoises encountering one another would repetitively lock onto and then lose interest in one another, executing a mating dance (fig. 3.6) in which "the machines cannot escape from one another; but nor can they ever consummate their 'desire' " (Walter 1953, 128, 129).
Figure 3.3.Anatomy of a tortoise. Source: de Latil 1956, facing p. 50.
So much for the behaviors of the tortoises; now to connect them to the brain. One can analogize a tortoise to a living organism by distinguishing its motor organs (the power supply, motors, and wheels), its senses (the contact switch and the photocell), and its brain (connected to the motor organs and senses by nerves: electrical wiring). The brain itself was a relatively simple piece of circuitry consisting of just two "neurons," as Walter (1950a, 42) put it, each consisting of an electronic valve, a capacitor, and a relay switch (fig. 3.7). In response to different inputs, the relays would switch between different modes of behavior: the basic wandering pattern, locking onto a light, oscillating back and forth after hitting an obstacle, and so on.
What can we say about the tortoise as brain science? First, that it modelled a certain form of adaptive behavior. The tortoise explored its environment and reacted to what it found there, just as all sorts of organisms do—the title of Walter's first publication on the tortoises was "An Imitation of Life" (1950a). The suggestion was thus that the organic brain might contain similar structures to the tortoise's—not valves and relays, of course, but something functionally equivalent. Perhaps, therefore, it might not be necessary to descend to the level of individual neurons to understand the aggregate properties of the brain. This is the sense in which Jerome Lettvin (once a collaborator of Warren McCulloch) could write in 1988 that "a working golem is . . . preferable to total ignorance" (1988, vi). But the tortoises also had another significance for Walter.
The tortoise's method of finding its targets—the continual swiveling of the photocell through 360 degrees—was novel. Walter referred to this as scanning, and scanning was, in fact, a topic of great cybernetic interest at the time. The central question addressed here was how the brain goes from atomistic sensory impressions to a more holistic awareness of the world. In the United States in 1947 Walter Pitts and Warren McCulloch published an influential paper, "How We Know Universals," which aimed to explain pattern recognition—for example, recognizing individual letters of the alphabet independently of their size and orientation—in terms of a scanning mechanism. More relevant to Walter, in his 1943 book The Nature of Explanation,Kenneth Craik (1943, 74), the British experimental psychologist, speculated about the existence of some cerebral scanning mechanism, always, it seems, explained by an analogy with TV. "The most familiar example of such a mechanism is in television, where a space-pattern is most economically converted for transmission into a time sequence of impulses by the scanning mechanism of the camera" (Walter 1953, 108). The basic idea was that the brain contains some such scanning mechanism, which continually scans over its sensory inputs for features of interest, objects, or patterns in the world or in configurations internal to itself.11
Figure 3.4.The tortoise in action. Source: de Latil 1956, facing p. 275.
Figure 3.5.The mirror dance. Source: Holland 1996.
One of the tortoise's most striking features, the rotation of the front forks and the photocell, was thus an implementation of this cybernetic notion of scanning. And beyond that, scanning had a further degree of significance for Walter. Craik visited Walter in the summer of 1944 to use the Burden's automatic frequency analyzers, and from that time onward both of them were drawn to the idea that the brainwaves recorded in Walter's EEGs were somehow integral to the brain's scanning mechanism (Hayward 2001b, 302). The basic alpha rhythm, for example, which stopped when the eyes were opened, could be interpreted as a search for visual information, a search "which relaxes when a pattern is found," just as the tortoise's photocell stopped going around when it picked up a light.12 This interpretation found some empirical support. As Walter noted (1953, 109), "There was the curious coincidence between the frequency of the alpha rhythms and the period of visual persistence. This can be shown by trying how many words can be read in ten seconds. It will be found that the number is about one hundred—that is, ten per second, the average frequency of the alpha rhythms" (Walter 1953, 109). He also mentioned the visual illusion of movement when one of a pair of lights is turned off shortly after the other. Such data were at least consistent with the idea of a brain that lives not quite in the instantaneous present, but instead scans its environment ten times a second to keep track of what is going on.13
Figure 3.6.The mating dance. Source: Holland 1996.
Figure 3.7.The brain of the tortoise. Source: Walter 1953, 289, fig. 22.
From a scientific perspective, then, the tortoise was a model of the brain which illuminated the go of adaptation to an unknown environment—how it might be done—while triangulating between knowledge of the brain emanating from EEG research and ideas about scanning.
Tortoise Ontology
We can leave the technicalities of the tortoise for a while and think about ontology. I do not want to read too much into the tortoise—later machines and systems, especially Ashby's homeostat and its descendants, are more ontologically interesting—but several points are worth making. First, the assertion that the tortoise, manifestly a machine, had a "brain," and that the functioning of its machine brain somehow shed light on the functioning of the human brain, challenged the modern distinction between the human and the nonhuman, between people and animals, machines and things. This is the most obvious sense in which Walter's cybernetics, like cybernetics more broadly, staged a nonmodern ontology.14 Second, we should reflect on the way the tortoise's brain latched onto its world. The tortoise is our first instantiation of the performative perspective on the brain that I introduced in chapter 1, the view of the brain as an "acting machine" rather than a "thinking machine," as Ashby put it. The tortoise did not construct and process representations of its environment (à la AI robotics); it did things and responded to whatever turned up (cycloidal wandering, locking onto lights, negotiating obstacles). The tortoise thus serves to bring the notion of a performative brain down to earth. In turn, this takes us back to the notion of Black Box ontology that I introduced in chapter 2. The tortoise engaged with its environment as if the latter were a Black Box, in Ashby's original sense of this word—a system to be performatively explored.15 As ontological theater, the tortoise staged a version of this Black Box ontology, helping us to grasp it and, conversely, exemplifying a sort of robotic brain science that might go with such an ontology.
Now we come to the complication I mentioned in chapter 2. In one sense the tortoise staged a nonmodern Black Box ontology, but in another it did not. For Walter, the point of the exercise was to open up one of these boxes, the brain, and to explore the inner go of it in the mode of modern science. How should we think about that? We could start by remembering that in Walt
er's work the world—the tortoise's environment—remained a Black Box. In this sense, Walter's cybernetics had a hybrid character: nonmodern, in its thematization of the world as a performative Black Box; but also modern, in its representational approach to the inner workings of the brain. My recommendation would then be to pay attention to the nonmodern facet of this hybrid, as the unfamiliar ontology that cybernetics can help us imagine. But there is more to think about here. The question concerns the extent to which Walter's brain science in fact conformed to the stereotype of modern science. As I mentioned in chapter 2, cybernetic brain science was an odd sort of science in several ways. First, the scientifically understood brain had as its necessary counterpart the world as an unopened Black Box, so that the modern and the nonmodern aspects of this branch of cybernetics were two sides of a single coin. Second, the style of scientific explanation here is what I called "explanation by articulation of parts." Walter's brain science did not emulate physics, say, in exploring the properties of the fundamental units of the brain (neurons or their electromechanical analogues); instead, it aimed to show that when simple units were interconnected in a certain way, their aggregate performance had a certain character (being able to adapt to the unknown). Again, this sort of science thematizes performance rather than knowledge of individual parts. And third, this style of explanation had a tendency to undermine its own modern impulse in what I call the cybernetic discovery of complexity, to which we can now turn.
_ _ _ _ _
IT IS ONE OF THE INTERESTING CONSEQUENCES OF THIS KIND OF MODELMAKING—THOUGH I ONLY REALISED IT AFTER I STARTED MAKING THESE TOYS— THAT A VERY SMALL NUMBER OF NERVE ELEMENTS WOULD PROVIDE FOR AN EXTREMELY RICH LIFE.
The Cybernetic Brain Page 6