Sunday, April 24, 2005

Intelligence is hysteresis: A just-so story

What is this biological capacity we call thought and why do we have it? Well sit right down and I'll tell you what I think: Intelligence is hysteresis.

A techy term fashionable among physicists, "hysteresis" describes the dependence of a thing's behavior on its history; that is, on what happened before. Ice doesn't care what went before: If the temperature now is above 0 degrees C it will melt. But a spring does care what went before: If it's been stretched or compressed it will exert a force, otherwise not.

Events on Earth sometimes seem random, but things do unfold according to physical laws--as well as according to the not always baffling choices of other rational actors--and so one can make predictions of what will happen next based on what's just happened up to now or else happened last time. A multicellular biological system (i.e. organism) that lacks hysteresis recognizes and reacts only to snapshots, which is a terrible handicap to learning or adaptation: "Is that safe falling on my head or floating away?"

So to me cognition and consciousness is most fundamentally about data storage. No doubt nerve cells first were about coordinating distant body parts and about responding fast to environmental circumstances. But my hunch is that they weren't about thought or cognition until they responded and stayed responded awhile. Once that happened, organisms could adapt to and/or learn something from the association, ala Pavlov, between that response or imprint and the crushing weight of the safe, learning or adapting ultimately a dodge reflex. From then on, evolutionarily lineages could play the game of what other useful tricks are there that an individual can do with history.

We had neurons, and we ran with them, and now we have transistors, so we are running with them in attempting artificial intelligence, but I think perhaps the essence--or one essence--is fundamentally abstract: Hysteresis.

Postscript: I now suspect it was hasty to "and/or" learning and adaptation. I was assuming that a process perfectly analagous to natural selection would act during an individual's lifetime to teach a lesson to him, her or it with memory being the chief innovative element evolution-wise. Behind that assumption was my awareness of the simple integrated circuits called "neural nets," in which the circuit elements that stand in for neurons have history-dependent interconnections standing in for synapses. These things learn. But I worry now I was hasty in assuming that in talking about memory or hysteresis in the way I did I was implicitly and necessarily talking about neural nets. I might well be overlooking something terribly obvious, but I haven't figure out what it is.

No comments: