(low-quality MP3 recording of the talk: 1hr, 35MB)

Jeff Hawkins practically bounces into the room, full of energy, even though there is hardly anyone is there. He is unfazed. “Well, at the very least my mom is coming,” he says. (Sure enough, his mom soon arrives — with a friend, even.) He needn’t worry though, the room fills up as it gets closer to the start time.

Hawkins is here to talk about his new book, On Intelligence, which, he says, is “a book about brains … a functional, biological theory about how the neocortex works”. The book exudes the same cheerful, almost goofy enthusiasm Hawkins does. “It can be a very technical field and I didn’t want it to be a very technical book,” Hawkins says, so he hired New York Times science writer Sandra Blakeslee as his co-author.

Hawkins seems like an odd person to write a book about brains. After all, he is a self-admitted “computer guy”. But upon reading a special issue of Scientific American about the brain, he realized he wanted to be the guy who figures out how the brain actually works. He tried to get started immediately but found there wasn’t much of an opportunity to do this without a lot of money, so he decided to make some money in the computer industry first.

As he well did. During his time in the industry, he invented the PalmPilot, the Treo smart phone, and other industry successes. “But all this time as an entrepreneur I was working part-time. The other part of my time was on neuroscience and I had a lot of friends in the neuroscience community.” About three years ago those friends convinced him to create a new institution to study this kind of stuff. It’s called the Redwood Neuroscience Institute (RNI) and it’s nearby in Menlo Park. “And over the last few years a lot of progress has been made — a tremendous amount — and that’s in the book.”

The book is about the neocortex, the white wrinkly stuff that covers the brain and handles high-level thought. He pulls out a large dinner napkin. “This is a model of the neocortex,” he says. “It gets crumpled up to fit into your head, that’s why it’s all wrinkly.” There are two important things to understand about the cortex. First, there are different regions of the cortex that do different things (sight, motor control, listening, speech, etc.). Second, the entire cortex — the whole dinner napkin — is pretty much identical everywhere.

The whole thing is a big pattern-matching machine. The pieces at the bottom of the hierarchy are connected directly to sensory inputs, like the eyes. They get messy, blurry signals back and try to make sense out of them: “that’s a line, that’s a circle”. Then they pass this up to the next level of the hierarchy. The next region is about half as small as the first one, because each piece combines a couple of the results from a piece below to make a larger result. “Line, line, line, line? That’s a rectangle.” And so on. Then they all pass it up the chain again, to another part which again puts them together. “Oh, that’s a piece of paper. That’s a bottle of water.” Finally, they get passed up to the higher-level portions of the brain. “Oh, this is my room.”

There are several key points here. First, each piece tries to match relationships, not things themselves. You listen to the musical distance between notes, not the notes. This is why you can’t notice if a song is sung in a different key without perfect pitch and why you can recognize people’s faces even under different colored lights. Even though the levels all change, the relationships between them stay the same.

Second, the connections all go in two directions. Once it’s bubbled up and your brain realizes its in your room, it passes the information about the layout of your room back down so that your eyes no where to look and your head knows where to turn. If you think “fan”, it passes that information back down the chain so that the eyes know to look for a white circle and the neck muscles know to look left.

Third, if any part sees something it doesn’t recognize, it passes it up to the next highest level. If nothing recognizes it, it goes to the hippocampus which makes a new memory for it. If you see the thing over and over again, the same pattern eventually gets learned at lower and lower levels, making your responses to things faster and more ingrained. This explains what an expert is: they’re someone who has pushed the details of a specific field down into their brain so they can process things more quickly.

(I’ve cut some details out of this report for length; you can see them by viewing source and reading the comments.)

posted January 23, 2005 08:16 PM (Education) (1 comments) #

Nearby

D.J. Bernstein: The Good News Archive
Pick A Side
In His Own Words
Newspaper Writers on the Election
Stanford: Day 62
Jeff Hawkins on the Brain
Stanford: Day 63
Stanford: Day 64
Stanford: Day 65
Quick Takes
Stanford: Day 66

Comments

I just wanted to thank you for posting this. I’ve been interested in Hawkins’ work ever since I relatively recently found out about it. The notion of a guy deciding to make a lot of money (and then doing it!) before going back to study the nature of intelligence is amazing.

It is clear to me that computational power is increasing far faster than our ability to write ‘intelligent’ software for it. There’s just so much yet to discover there! I’m certainly envious of the guy, though I really shouldn’t be.

posted by Rich at February 28, 2005 02:46 AM #

Subscribe to comments on this post.

Add Your Comment

If you don't want to post a comment, you can always send me your thoughts by email.


(used only to send you my reply, never published or spammed)

Remember personal info?


Note: I may edit or delete your comment. (More...)

Aaron Swartz (me@aaronsw.com)