The Bactra Review: Occasional and eclectic book reviews by Cosma Shalizi   24


Experiments in Synthetic Psychology

by Valentino Braitenberg

illustrations by Maciek Albrecht

A Bradford Book / The MIT Press, 1984.

Hume on Wheels, or, One Must Imagine Frankenstein Happy

I'm now convinced that Mary Shelley got Victor Frankenstein wrong. (No doubt it was her sources' fault.) He was German, of course, but rather the cheerful, hearty, bouncy sort of German (you know the ones I mean), one of the ones who can't say anything without making a joke, or at least a pun. Instead of thinking deep and dark thoughts about electricity, magnetism, Mesmer and the alchemists, Frankenstein probably read Voltaire, Buffon, Lavoisier and a bootleg copy of D'Alembert's Dream, and certainly Hume, Condillac, and La Mettrie. Instead of mucking about with corpses and lightning-rods, he must have played with gears and pumps and the ingenious self-governing steam-engine, starting with small little mechanical creatures --- it was the age after Vaucanson's quacking, eating and crapping duck, after all --- and working his way up from there. His creations would have lived in a separate room in the ancestral house, which he'd call the nursery (laughing every time he said it), and made Elizabeth's servant problem more than usually impossible. Eventually, he'd have published a book, probably anonymously and in Holland, which would have been put on the Index, read by Stephen Maturin and Percy Shelley, and been very much like Vehicles.

Of course Vehicles was not inspired by this all-too-sunny version of Frankenstein, rather the reverse. The real source of Braitenberg's "experiments in synthetic psychology" are some decades of pondering how the brain works, of experimenting on real brains to try and see what happens in the head to make a mind, and reflecting on what used to be known as cybernetics. The stated goal is to "let the problem of mind dissolve in your mind" --- "while I was counting fibers in the visual ganglia of the fly or synapses in the cerebral cortex of the mouse, I felt knots unite, distinctions dissolve, difficulties disappear, difficulties I had experienced much earlier while I still held my first naive philosophical approach to the problem of mind." (An old chestnut of the Enlightenment: experience displacing vain metaphysics.) Accordingly Vehicles is entirely devoted to thought experiments, to imaginary contraptions, though it's worth re-iterating that all of them could be built, and some extremely similar devices actually were, as long ago as the '40s and '50s.

The idea is to build very simple machines and see how they behave, gradually increasing their internal complication and sophistication until mental, intentional descriptions become irresistible. (Here we connect to Condillac; also Dennett.) At every step Braitenberg's law of "uphill analysis and downhill synthesis" applies and is illustrated: it's easier to design a mechanism from scratch to do something, than to figure out just how nature has contrived to do it; this suggests that maybe the natural way isn't really insuperably complicated.

The basic machine is a motor connected to a single sensor, which controls the motor's activation. These barely qualify as brutes, and Braitenberg is detained by them for no more than two pages. He then adds multiple motors and multiple sensors, crossing their wires and making some of them inhibitory. This produces creatures which are still extremely simple, but show fear, aggression, love, and affection combined with a wandering eye. (You can put scare-quotes around these words if you wish.)

Having induced our machines to want and act, we now give them thought. In swift succession, we add some logical capacity, in the form of threshold devices (McCulloch-Pitts neurons, for those who care); the possibility of making external symbols and using them to regulate their logical process (pure Vygotsky, but apparently independently arrived at, and tossed off in two paragraphs); and increasing elaboration through a rough sort of natural selection: vehicles get picked up and copied, sloppily, unless they destroy themselves or leave the lab, in which case they're out of the running. Not much use is made of the evolution of vehicles, until almost the very end, but it does give us a plausible excuse for putting in fancier and fancier vehicles.

Now we install association, which could be straight out of Hume, though Donald Hebb (The Organization of Behavior, 1949) is the official inspiration. All the sensors, logic elements, etc. are wired together with "mnemotrix", which forms connections that become stronger the more often two elements are active simultaneously. The vehicle will rapidly come to have patterns of connections, or associations, or concepts (depending on how intentional one wants to be), which reflect patterns of concurrence, or the properties of things (depending on how realist one wants to be) in the world outside it, and these will control its behavior. After this, we add in hard-wired ways of representing time and space; this again is Hume, only in one of the parts of the Treatise on Human Nature which nobody reads. A grasp of causation, of rules and regularities in succession, is added by means of another kind of wire, "ergotrix," which "conducts in one direction only and has an increased conductance when it is interposed between elements that are active in succession within a brief time." (I can't think of any such material, but it'd be easy to simulate with a few circuit elements.) In other words: causation is constant succession, inferred more from an inescapable habit than from anything else. A mechanism to prevent excitation from spreading out of control, or runaway associations (raising thresholds as the over-all level of excitation rises) leads to focused attention and trains of thought, and may, depending on its characteristics, make the vehicle's behavior chaotic in the technical sense, and so unpredictable in the long-run.

The last steps are to add foresight (predictive circuitry, with learning rules), strong, survival-related rules overpowering foresight (derived from taking the brains of a Darwinianly evolved creature and attaching it to the predictor; "reason is, and ought only to be, the slave of the passions") and egoism. (Altruism is not explicitly discussed, but it'd be easy to carry through the usual game-theoretic arguments for why it should evolve under certain circumstances.) Braitenberg tops things off with optimism, which he claims is self-fulfilling and so adaptive; this, if nothing else, proves him a happy man.

After vehicles themselves and the illustrations (imagine Ralph Steadman rendering "Francis Crick Goes to Heaven" or Cosmicomics), Braitenberg has a series of notes on his sources, the biological plausibility of his vehicles, and many curiosities of human brains and psychology, like the neuronal zoom effect and Kanisza contours.

I found myself agreeing with Braitenberg's points, explicit or only slightly implict, almost all the time, but then, I held those views to begin with, and learnt some of them, literally, at my parents' knees. How persuasive, or even intelligible, it would be to someone who was strongly anti-reductionist, anti-mechanist, most of all anti-scientistic or anti-naturalist, I couldn't say. I'd hope that, even if readers come away convinced Braitenberg sheds absolutely no light on human thought whatsoever, they'd at least be charmed to see ideas handled so skillfully and playfully; and for those interested in psychology, neuroscience, AI, artificial life or philosophy of mind, it ought to be required reading.

152 pp., references, no index
Artificial Life / Mind, Consciousness, etc. / Neuroscience / Popular Science / Self-Organization, Complexity, etc.
Currently in print as a paperback, ISBN 0-262-52112-1, US$20 [Buy from Powell's]; out-of-print hardback, ISBN 0-262-02208-7 [Buy from Powell's]. LoC QP356 B74 1984
27 September -- 12 October 1997