Reviews & Press

Giving some thought to thinking

By Mike Langberg
The San Jose Mercury News
Fri, Oct. 01, 2004

Computers don't think, no matter how fast they calculate spreadsheets or solve chess problems. They just crunch numbers. So how do our brains manage to create that most human of characteristics, intelligence?

Jeff Hawkins has been obsessed with finding the answer for most of his life. As a sideline, he became a Silicon Valley celebrity for designing the original Palm Pilot and a string of successor handheld devices, including the current PalmOne Treo 600 smart phone.

But in all the years he has worked at making products and co-founding two companies, the original Palm Computing and Handspring, he always set aside time for serious research in neuroscience.

And now Hawkins, 47, has written a book, appropriately called "On Intelligence," to explain the theory he has developed to bridge the confounding gap between the mind and the physical functions of the brain.

The theory builds on general developments in neuroscience as well as discussions at the Redwood Neuroscience Institute in Menlo Park, which Hawkins launched two years ago. The institute, with a staff of about 10 researchers, is supported almost entirely out of Hawkins' pocket.

He continues splitting his career, working half time as chief technology officer of Milpitas-based PalmOne and half time as director of his institute.

I'm not a neuroscientist, or even a book critic, but I read the 272 pages of "On Intelligence" (www.onintelligence.org), officially published today, and found myself fascinated with the ideas Hawkins is putting forward, and grateful that he writes for non-scientists in a jargon-free style.

In one short chapter at the end of the book, Hawkins talks about how his theory could create new types of computers that would mimic human thinking -- making them much better at tasks that still stump today's computers, such as truly understanding human speech or interpreting what they see when hooked to a video camera.

But after talking to Hawkins earlier this week at his institute office, I'm convinced he's not studying intelligence in search of new investment opportunities. He genuinely wants to advance human understanding of ourselves.

To me, that's one of the best ways to spend the wealth that technology entrepreneurs so often accumulate. Too many engineers, when they're smart enough or lucky enough to become wildly successful, don't seem to know what to do next. Look at Microsoft co-founder Paul Allen, who has frittered away hundreds of millions of dollars on sports teams, a gigantic yacht and a museum honoring Jimi Hendrix.

Even if Hawkins' ideas are wrong, the type of work he's doing is crucial to the future of Silicon Valley. Hawkins' book and institute will stimulate scientific debate, while sending a clear message that Silicon Valley is still the place where the world's brightest people come to struggle with the biggest and ultimately most-rewarding technology challenges.

So what is Hawkins saying? I'll try to summarize, at the risk of grossly over-simplifying.

The cortex, as Hawkins recounts in the book, is a thin sheet of tissue that would be about the size and thickness of a large dinner napkin if unfolded from the surface of our brains. It is the seat of human intelligence, with about 30 billion neurons, a type of specialized brain cell sometimes erroneously compared to transistors in a computer's microprocessor. Each of those neurons has several thousand electrical connections to other neurons.

These connections allow the cortex to store and process information in a fundamentally different way than today's digital computers, using a system Hawkins calls the "memory-prediction framework."

Your brain remembers key elements of everything you've experienced, then applies that knowledge to new situations. That's why you don't have to learn to drive all over again when you get a new car; you know in a general way how cars work and can adapt that information to new circumstances. Today's computers, in contrast, would need specific instructions to explain every change in dashboard layout from the old car to the new.

Hawkins suggests a new type of computer could work like the cortex, building up a library of experiences and using those experiences to analyze new situations. Applied to narrowly defined tasks, such a computer could be more intelligent than humans.

He even described to me a computer that could be "a smarter Einstein," quickly digesting all the physics papers written in the past 50 years and using them to devise new theories.

"It would think about physics like you and I think about physics," Hawkins said from his corner office overlooking the Caltrain station in Menlo Park, on the third floor of the Kepler's Books building. "But it wouldn't be thinking about the other things we think about — like how to get dinner on the table."

To mark today's official publication of his book, Hawkins will descend the stairs this evening at 7:30 to give a talk at Kepler's. The book's Web site lists several future local appearances. If you're interested in the future of computers, or how the brain works, I would highly recommend investing the time to learn more about what Hawkins thinks about the way we all think.