Sunday, June 7, 2009

July 05, 2006: Chapter 6 of "On Intelligence:" Where the Action Is

I have now made it through Chapter 6 of On Intelligence; and I am glad that I persisted (even in the midst of all other stuff competing for my attention). Personally, I probably would have been happier with a monograph that consisted only of this chapter, since, in terms of such factors as rhetoric and style, I think that Chapter 6 is stronger than anything preceding it. The fact that it is disproportionately longer than any other chapter in the book seems to indicate that it received much more attention, which is why I think it would have been better off delivered as a monograph. In that setting there would have been less temptation to pad out the preceding material in ways that damaged those factors of rhetoric and style.

This is not to say that I was altogether happy with Chapter 6, just that this chapter gave me a better sense of what Hawkins felt his message was. As a matter of fact, at the end of the day, I would characterize Chapter 6 as "Edelman-lite," meaning that it covers much the same ground that Gerald Edelman covered in The Remembered Present, but in a generally more accessible manner. The stories are about the same, as are the settings "in terms of regions of the brain." As a matter of fact, the most important difference involves the prize on which the respective authors eyes are focused. Edelman wants to talk about consciousness, and Hawkins wants to talk about intelligence. I want to make the case that this difference is insignificant because, at the end of the day, that terminology does not matter very much.

My argument is based on Drew McDermott's paper "Artificial Intelligence Meets Natural Stupidity." Like the little boy who saw the emperor as naked, McDermott had the courage to describe an "expert system" in terms of what it really did (essentially the operationalization of a rule base) and then make the argument that there was no sound reason for assuming any connection between that operationalization and such elevated concepts as "intelligence" or "expertise." I think there is much to be gained from reading both Edelman and Hawkins with such dispassionate insight.

From that point of view, both are writing about a single functional capacity. Since Edelman got there first (but also because Hawkins never calls it out as a concept unto itself), I shall use Edelman's label for that capacity: perceptual categorization. The bottom line is that this is what the neocortex does. This affirms Mountcastle's observation about its uniformity and his hypothesis that, by doing that one thing well (like the hedgehog of Isaiah Berlin's essay), it can cover most of the ground necessary to account for not only interpreting sensory signals but also acting on them. Edelman goes deeper into details (and has tried to use simulation as a tool for better understanding of those details); but his peer group expects that of him. More importantly, Edelman devotes more attention to the whole body, rather than just worrying about the neocortex. So he puts a lot more weight in analyzing the tight coupling between perception and motor behavior and less on the Hawkins idea of a "memory-driven" system (since the system is driven by perceptual categories and Edelman argues that memory is a higher-level phenomenon). Also Edelman argues for a tight coupling between any "cognitive" functionality in neocortex and activity in the limbic system associated with emotional functionality. This part is totally beyond the scope of Chapter 6 (I shall have to wait for Chapter 8); but Edelman now has a lot of support for his position from Damasio.

So my strongest quibble with Hawkins is probably over his title! The fact is that Turing is the one who got it right: He began his classic "Turing test" paper by positing that we cannot define the concept of intelligence and probably will not be able to do so for some time, if ever. He then said (in paraphrase), "Well, if we cannot answer the question 'Can a machine think?' let us consider a simpler question whose answer may be within our grasp;" and that is when he lays out the Turing test. Searle then got all bent out of shape because he felt that too many people were taking the simpler question to be a question about intelligence; and, if you look at the kind of discourse that was flying around the AI community at that time, he was probably right. Unfortunately, the more heat that got generated over the Turing test, the less attention was paid to why we were thinking about it in the first place! Ironically, I probably would have been happier if Hawkins had called his book How the Brain Works. A serious brain theorist would probably have found that title just as hyperbolic as On Intelligence, but it would have been a more straightforward account of the most important subject matter in the book.

No comments:

Post a Comment