November 24, 2004
Just Read: On Intelligence

I just finished Jeff Hawkins' On Intelligence, and it was a great read, reinforcing some of my own beliefs on how one should build "smart" software. [Technical note to self: Among the things it did, without trying to, was to reinfoce my interest in the Sequitur algorithm and remind me of this strongly worded opinion on the purpuse of the visual cortex (from the son of D.M Mackay - one of Hawkins' sources as well).]

The book is, for it's genre, delightfully short and to Hawkins' great credit he does not claim that everybody else has been wrong all the time and he's the only one who gets it. This is the usual bane of popular science books purporting revolution (e.g. Wolframs horrible bloatfest New Kind of Science or Mandelbrots recent book on the stock market), and it is great to read a book that is just earnestly interested in breaking new ground instead of also doubling as a vanity project.

The least satisfying part of Hawkins' book is its philosophical underpinning. Hawkins spends a lot of time talking about Searle's Chinese Room argument, in my opinion a completely pointless piece of overthinking, and he's making a valid point while doing so even if Searle's argument is bogus. To Hawkins, Searle's argument applies to "old AI"; mechanistic, syllogistic, 'electron brain' AI, where we encode the rules of language in a machine in some deterministic way so we can do translation. Hawkins agrees in deriding this kind of simpleminded AI and therefore agrees with Searle. But the rest of his book is one long description of a mechanism for intelligence that I'm positive Searle would also dismiss as non-intelligent by the very same argument!
Hawkins is of the opinion that the model he describes - a sense/memory-based hierarchical prediction machine - captures the notion of understanding in ways the old-school AI does not. I think Hawkins' point is a lot more interesting than Searle's, but I'm sure Searle would disagree with Hawkins in thinking this alternate model any different. After all, it is not understanding anything, it's just playing the odds - predicting.

Posted by Claus at November 24, 2004 01:51 AM | TrackBack (0)
Comments (post your own)
Help the campaign to stomp out Warnock's Dilemma. Post a comment.

Email Address:


Type the characters you see in the picture above.

(note to spammers: Comments are audited as well. Your spam will never make it onto my weblog, no need to automate against this form)


Remember info?