[By Nic Lindh on Wednesday, 10 October 2007]
On Intelligence is an attempt to provide a framework for how human brains work in order to help us one day build artificial intelligences. The book was co-written by Jeff Hawkins, the inventor of the Palm Pilot, who obviously knows a thing or two about how to make machines act smart.
The basic thesis is that the brain is very much not like a computer—that computers and brains are fundamentally different, and that it’s only by understanding how the brain really works that we’ll be able to synthesize something that acts like it. According to the book, the brain, at the most basic level, only does two things: it evaluates patterns, and it remembers patterns. All sensory input boils down to patterns over time.
This would explain how, when you sit down at your desk at work in the morning, you realize that something is wrong if somebody has moved your coffee cup. If asked before you entered the office, you probably couldn’t have said where, exactly, the coffee cup was, but as you decide to reach for the cup, your brain runs through the sequence of how reaching for the coffee cup “should” go, and once that fails, you realize that it’s been moved. Everything you do, your brain is running through a memory of how it “should” be, and you notice when the memory doesn’t coincide with reality.
This could also explain why new experiences are tiring—your brain is struggling to absorb the new patterns and mesh them with existing memories.
Without being a neuroscientist, it’s hard to disagree with Hawkins’s ideas. They make sense.
They also could usher in a very exciting era of truly smart machines.
For completeness, On Intelligence also includes a section on the ethics and risks involved in building these kinds of machines.
It’s a very interesting read.