Ashby's first cybernetic paper, then, discussed some very simple instances of dynamic equilibrium and portrayed them as models of the brain. One is reminded here of Wiener's cybernetics, in which feedback systems stood in as model of the brain, and indeed the thermostat as discussed by Ashby was none other than such a system. And two points are worth noting here. First, a historical point: Ashby's essay appeared in print three years before Arturo Rosenblueth, Wiener, and Julian Bigelow's classic article connecting servomechanisms and the brain, usually regarded as the founding text of cybernetics. And second, while Rosenblueth, Wiener, and Bigelow (1943) thought of servomechanisms as models for purposive action in animals and machines, Ashby's examples of homeostatic mechanisms operated below the level of conscious purpose. The brain adumbrated in Ashby's paper was thus unequivocally a performative and precognitive one.
_ _ _ _ _
I quoted Ashby as saying that he solved the problem of how the brain can be at once mechanistic and adaptive in 1941, and his major achievement of that year is indeed recorded in a notebook entitled "The Origin of Adaptation," dated 19 November 1941, though his first publication on this work came in an essay submitted in 1943 and only published in 1945, delayed, no doubt, by the exigencies of war (Ashby 1945a). The problematic of both the notebook and the 1945 publication is this: Some of our biological homeostatic mechanisms might be given genetically, but others are clearly acquired in interaction with the world. One of Ashby's favorite adages was, The burned kitten fears the fire. The kitten learns to maintain a certain distance from the fire—close enough to keep warm, but far away enough not get to burned again, depending, of course, on how hot the fire is. And the question Ashby now addressed himself to was how such learning could be understood mechanistically—what could be the go of it? As we have seen, Walter later addressed himself to the question of learning with his conditioned reflex analogue, CORA. But Ashby found a different solution, which was his first great contribution to brain science and cybernetics.
The 1945 essay was entitled "The Physical Origin of Adaptation by Trial and Error," and its centerpiece was a strange imaginary machine: "a frame with a number of heavy beads on it, the beads being joined together by elastic strands to form an irregular network." We are invited to think of the positions and velocities of the beads as the variables which characterize the evolution of this system in time, and we are invited also to pay attention to "the constants of the network: the masses of the beads, the lengths of the strands, their arrangement, etc. . . . These constants are the 'organization' [of the machine] by definition. Any change of them would mean, really, a different network, and a change of organization." And it is important to note that in Ashby's conception the "constants" can change; the elastic breaks if stretched too far (Ashby 1945a, 15–16).11
The essay then focuses on the properties of this machine. Suppose we start it by grabbing one of the beads, pulling it against the elastic, and letting go; what will happen? There are two possibilities. One is that the whole system of beads and elastic will twang around happily, eventually coming to a stop. In that case we can say that the system is in a state of dynamic equilibrium, as defined in the 1940 essay, at least in relation to the initial pull. The system is already adapted, as one might say, to that kind of pull; it can cope with it.
But now comes the clever move, which required Ashby's odd conception of this machine in the first place. After we let go of the bead and everything starts to twang around, one of the strands of elastic might get stretched too far and break. On the above definition, the machine would thus change to a different state of organization, in which it might again be either stable or unstable. In the latter case, more strands would break, and more changes of organization would take place. And, Ashby observed, this process can continue indefinitely (given enough beads and elastic) until the machine reaches a condition of stable equilibrium, when the process will stop. None of the individual breaks are "adaptive" in the sense of necessarily leading to equilibrium; they might just as well lead to new unstable organizations. In this sense, they are random—a kind of nonvolitional trial-and-error process on the part of the machine. Nevertheless, the machine is ultrastable—a technical term that Ashby subsequently introduced—inasmuch as it tends inexorably to stable equilibrium and a state of adaptedness to the kinds of pull that initially set it in motion. "The machine finds this organization automatically if it is allowed to break freely" (1945a, 18).
Here, then, Ashby had gone beyond his earlier conception of a servomechanism as a model for an adaptive system. He had found the solution to the question of how a machine might become a servo relative to a particular stimulus, how it could learn to cope with its environment, just as the burned kitten learns to avoid the fire. He had thus arrived at a far more sophisticated model for the adaptive and performative brain than anyone else at that time.
The Homeostat
The bead-and-elastic machine just discussed was imaginary, but on 19 November 1946 Ashby began a long journal entry with the words "I have been trying to develope [sic] further principles for my machine to illustrate stability, & to develope ultrastability." There followed eight pages of notes, logic diagrams and circuit diagrams for the machine that he subsequently called the homeostat and that made him famous. The next entry was dated 25 November 1946 and began: "Started my first experiment! How I hate them! Started by making a Unit of a very unsatisfactory type, merely to make a start."12 He then proceeded to work his way through a series of possible designs, and the first working homeostat was publicly demonstrated at Barnwood House in May 1947; a further variant was demonstrated at a meeting of the Electroencephalographic Society at the Burden Neurological Institute in May 1948.13 This machine became the centerpiece of Ashby's cybernetics for the next few years. His first published account of the homeostat appeared in the December 1948 issue of the journal Electronic Engineering under the memorable title "Design for a Brain," and the same machine went on to feature in the book of the same name in 1952. I therefore want to spend some time discussing it.
The homeostat was a somewhat baroque electromechanical device, but I will try to bring out its key features. Figure 4.4a in fact shows four identical homeostat units which are all electrically connected to one another. The interconnections cannot be seen in the photograph, but they are indicated in the circuit diagram of a single unit, figure 4.4c, where it is shown that each unit was a device that converted electrical inputs (from other units, on the left of the diagram, plus itself, at the bottom) into electrical outputs (on the right). Ashby understood these currents as the homeostat's essential variables, electrical analogues of blood temperature or acidity or whatever, which it sought to keep within bounds—hence its name—in a way that I can now describe.
The inputs to each unit were fed into a set of coils (A, B, C, D),producing a magnetic field which caused a bar magnet (M) to pivot about a vertical axis. Figure 4.4b is a detail of the top of a homeostat, and shows the coils as a flattened oval within a Perspex housing, with the right-hand end of the bar magnet just protruding from them into the light. Attached to the magnet and rotating with it was a metal vane—the uppermost element in figures 4.4b and 4.4c—which was bent at the tip so as to dip into a trough of water—the curved Perspex dish at the front of figure 4.4b, the arc at the top of figure 4.4c. As indicated in figure 4.4c, an electrical potential was maintained across this trough, so that the tip of the vane picked up a voltage dependent on its position, and this voltage then controlled the potential of the grid of a triode valve (unlabeled: the collection of elements enclosed in a circle just below and to the right of M in figure 4.4c; the grid is the vertical dashed line through the circle), which, in turn, controlled the output currents.
Figure 4.3.Page from Ashby's journal, including his first sketch of the homeostat wiring diagram. Source: Journal entry dated 28 December 1946 (p. 2094). (By permission of Jill Ashby, Sally Bannister, and Ruth Pettit.)
Thus the input-output relations of the homeostat except for on
e further layer of complication. As shown in figure 4.4c, each unit could operate in one of two modes, according to the setting of the switches marked S, the lower row of switches on the front of the homeostat's body in figure 4.4a. For one setting, the input current traveled to the magnet coil through a commutator, X, which reversed the polarity of the input according to its setting, and through a potentiometer, P, which scaled the current according to its setting. The settings for P and X were fixed by hand, using the upper and middle set of knobs on the front of the homeostat in figure 4.4a. More interesting, the switch S could also be set to route the input current through a "uniselector" or "stepping switch"—U in figure 4.4c. Each of these uniselectors had twenty- five positions, and each position inserted a specific resistor into the input circuit, with the different values of the twenty-five resistances being "deliberately randomised, the actual numerical values being taken from a published table of random numbers" (Ashby 1948, 381). Unlike the potentiometers and commutators, these uniselectors were not set by hand. They were controlled instead by the internal behavior of the homeostat. When the output current of the unit rose beyond some preset limit, relay F in figure 4.4c would close, driving the uniselector (via the coil marked G)to its next setting, thus replacing the resistor in the input circuit by another randomly related to it.
Figure 4.4.The homeostat: a, four interconnected homeostats; b, detail of the top of a homeostat unit, showing the rotating needle; c, circuit diagram. Source: W. R. Ashby, "Design for a Brain," Electronic Engineering, 20 (December 1948), 380, figs. 1, 2. (With kind permission from Springer Science and Business Media.)
So what? The first point to bear in mind is that any single homeostat unit was quite inert: it did nothing by itself. On the other hand, when two or more units were interconnected, dynamic feedback interrelations were set up between them, as the outputs of each unit fed as input to the others and thence returned, transformed, as input to the first, on and on, endlessly around the loop. And to get to grips with the behavior of the whole ensemble it helps to specialize the discussion a bit. Consider a four-homeostat setup as shown in figure 4.4a, and suppose that for one of the units—call it homeostat 1—the switch S brings a uniselector into the input circuit, while for the three remaining homeostats the switches S are set to route the input currents through the manually set potentiometers and commutators. These latter three, then, have fixed properties, while the properties of homeostat 1 vary with its uniselector setting.
When this combination is switched on, homeostat 1 can find itself in one of two conditions. It might be, as Ashby would say, in a condition of stable equilibrium, meaning that the vane on top of the unit would come to rest in the middle of its range, corresponding by design to zero electrical output from the unit, and return there whenever any of the vanes on any of the units was given a small push. Or the unit might be unstable, meaning that its vane would be driven toward the limits of its range. In that event, the key bit of the homeostat's circuitry would come into play. As the electrical output of the unit increased above some preset value, the relay would close and drive the uniselector to its next position. This, in effect, would change the electrical properties of homeostat 1, and then we can see how it goes. The unit might again find itself in one of two conditions, either stable or unstable. If the latter, the relay would again drive the uniselector to its next position, inserting a new resistance in the circuit, and so on and so on, until homeostat 1 found a condition of stable equilibrium in which its vane gravitated to the center of its range.
This is the key point about the homeostat: it was a real ultrastable machine of the kind that Ashby had only imagined back in 1941. The uniselectors took the place of the bands that broke in the fantasy machine of his 1945 publication (with the added advantage that the uniselectors were always capable of moving to another position, unlike elastic bands, which never recover from breaking). Started off in any configuration, the homeostat would randomly reorganize itself to find a condition of dynamic equilibrium with its environment, without any external intervention.
The homeostat was, then, a major milestone in Ashby's twenty-year quest to understand the brain as a machine. Now he had a real electromechanical device that could serve in understanding the go of the adaptive brain. It was also a major development in the overall cybernetic tradition then crystallizing around Wiener's Cybernetics, also published in 1948.14 I want to pause, therefore, to enter some commentary before returning to the historical narrative—first on ontology, then on the social basis of Ashby's cybernetics.
The Homeostat as Ontological Theater
ASHBY'S BRILLIANT IDEA OF THE UNPURPOSEFUL RANDOM MECHANISM WHICH SEEKS FOR ITS OWN PURPOSE THROUGH A PROCESS OF LEARNING IS . . . ONE OF THE GREAT PHILOSOPHICAL CONTRIBUTIONS OF THE PRESENT DAY.
NORBERT WIENER,THE HUMAN USE OF HUMAN BEINGS,2ND ED. ( 1967 [1950] ), 54
THERE CAN'T BE A PROPER THEORY OF THE BRAIN UNTIL THERE IS A PROPER THEORY OF THE ENVIRONMENT AS WELL. . . . THE SUBJECT HAS BEEN HAMPERED BY OUR NOT PAYING SUFFICIENTLY SERIOUS ATTENTION TO THE ENVIRONMENTAL HALF OF THE PROCESS. . . . THE "PSYCHOLOGY" OF THE ENVIRONMENT WILL HAVE TO BE GIVEN ALMOST AS MUCH THOUGHT AS THE PSYCHOLOGY OF THE NERVE NETWORK ITSELF.
ROSS ASHBY,DISCUSSION AT THE 1952 MACY CONFERENCE(ASHBY 1953B, 86–87)
My ontological commentary on the homeostat can follow much the same lines as that on the tortoise, though I also want to mark important differences. First, like the tortoise, the homeostat stages for us an image of an immediately performative engagement of the brain and the world, a little model of a performative ontology more generally. Again, at the heart of this engagement was a process of random, trial-and-error search. The tortoise physically explored its environment, finding out about distributions of lights and obstacles; the homeostat instead searched its inner being, running through the possibilities of its inner circuitry until it found a configuration that could come into dynamic equilibrium with its environment.
Next we need to think about Ashby's modelling not of the brain but of the world.15 The world of the tortoise was largely static and unresponsive—a given field of lights and obstacles—but the homeostat's world was lively and dynamic: it was, as we have seen, more homeostats! If in a multiunit setup homeostat 1 could be regarded as a model brain, then homeostats 2, 3, and 4 constituted homeostat 1's world. Homeostat 1 perturbed its world dynamically, emitting currents, which the other homeostats processed through their circuits and responded to accordingly, emitting their own currents back, and so on around the loop of brain and world. This symmetric image, of a lively and responsive world to be explored by a lively and adaptive brain, was, I would say, echoing Wiener, the great philosophical novelty of Ashby's early cybernetics, its key feature.
As ontological theater, then, a multihomeostat setup stages for us a vision of the world in which fluid and dynamic entities evolve together in a decentered fashion, exploring each other's properties in a performative back-and-forth dance of agency. Contemplation of such a setup helps us to imagine the world more generally as being like that; conversely, such a setup instantiates a way of bringing that ontological vision down to earth as a contribution to the science of the brain. This is the ontology that we will see imaginatively elaborated and played out in all sorts of ways in the subsequent history of cybernetics.16 Biographically, this is where I came in. In The Mangle of Practice I argued that scientific research has just this quality of an emergent and performative dance of agency between scientists and nature and their instruments and machines, and despite some evident limitations mentioned below, a multihomeostat setup is a very nice starting point for thinking about the ontological picture I tried to draw there. It was when I realized this that I became seriously interested in the history of cybernetics as elaborating and bringing that ontological picture down to earth.
Three further remarks on homeostat ontology might be useful. First, I want simply to emphasize that relations between homeostats were entirely noncognitive and nonrepresentational. The homeos
tats did not seek to know one another and predict each other's behavior. In this sense, each homeostat was unknowable to the others, and a multihomeostat assemblage thus staged what I called before an ontology of unknowability. Second, as discussed in chapter 2, paradigmatic modern sciences like physics describe a world of fixed entities subject to given forces and causes. The homeostat instead staged a vision of fluid, ever-changing entities engaged in trial-and-error search processes. And a point to note now is that such processes are intrinsically temporal. Adaptation happens, if it happens at all, in time, as the upshot of a temporally extended process, trying this, then that, and so on. This is the sense in which the homeostat adumbrates, at least, an ontology of becoming in which nothing present in advance determines what entities will turn out to be in the future. This is another angle from which we can appreciate the nonmodernity of cybernetics. Third, we could notice that the brain/world symmetry of Ashby's setups in fact problematized their specific reference to the brain. We can explore Ashby's response to this later, but to put the point positively I could say now that this symmetry indexes the potential generality of the homeostat as ontological theater. If the phototropism and object avoidance of the tortoise tied the tortoise to a certain sort of brainlike sensing entity, very little tied the homeostat to the brain (or any other specific sort of entity). A multihomeostat configuration could easily be regarded as a model of a world built from any kind of performatively responsive entities, possibly including brains but possibly also not. Here, at the level of ontological theater, we again find cybernetics about to overflow its banks.
The Cybernetic Brain Page 13