 |
|
|
|
|
|
|
 |
|
|||||
 |
 |
 |
 |
 |
 |
|
 |
|
|||||
 |
 |
 |
 |
 |
|
|
|
|||||||||||||||||||||||||||||
On Intelligence by Jeff Hawkins
|
|
Most Recent Book Reviews Most Popular Book Reviews Other Types of Reviews: Movie Reviews
|
title: On Intelligence I became interested in the question of what the physical basis for intelligence and consciouness are at an early age. Occasionally I read the literature to see if anyone has come up with good ideas or hard facts on this subject. Usually I am disappointed. On Intelligence does not dissappoint because it is both honest and insightful. Jeff Hawkins does not promise too much, but he shares a great deal. Scientists once thought it would be easy to make computers intelligent. They proved themselves wrong. The problem most likely lies partly in the design of computers and partly in a failure to understand the nature of human intelligence. To date artificial intelligence, or AI, has done some nice tricks, but nothing like what an average human can do. Scientists decided that what they did better than other people - like math and logic - was the true sign of intelligence. But it turns out things that non-scientists may do well, like hitting a pitched baseball with a wooden bat or entering a room and immediately making the perfect complement, require way more real intelligence than adding a column of numbers, or even than finding the nth root of numbers. [See my Indexing Books: Lessons in Language Computations for one example - why computers cannot generate good book indexes.] A little Wittgenstein would have gone a long way here. Computer scientists were asking the wrong questions. Then conducting the wrong experiments. Then doing the same thing over and over again, for decades. The right question is, how does the human brain do things? And, of those things the human brain can do, which are intelligent. Surprisingly, in this era of supercomputers and genetic analysis, neuroscientists still don't know how the neural activity of the brain adds up to intelligent behavior. They know a lot of facts, they have some theories, but a sound testable theory was lacking. Jeff Hawkins, in addition to explaining his topic clearly, presents a theory, or at least the outlines of one, that can be tested. Most of the book is easy reading even if you are not already familiar with intelligence issues, but Chapter 6, How the Cortex Works, was pretty dense. For that chapter I decided to mainly focus on what functions the cortex needs in order to display the systemic capabilities we call intelligence, rather than get bogged down in the details of neural circuitry on the first read. Which means I get to read it again, more slowly. A key takeaway is that intelligence and memory are intertwined in the human brain. Another is that intelligence is about prediction, and the brain and its components, even at the single neuron level, constantly make predictions. In addition both sensory data and memories are built into a hierarchy that makes sequential patterns (like sound or muscle movements) reusable. I once spent considerable time studying computer simulations called neural networks; I still have some tomes about that on my book shelf. One thing I found interesting in Hawkin's theory (I am holding judgement in reserve) is that his design for human brain intelligence is largely localized. Specific memories and patterns should be in specific areas of the cortex. In many neural network models memory is broadly dispersed. It is not in any particular nerve cell or set of synapses. This distributed memory/intelligence model has an intuitive appeal to me, and I wonder if it will have to be abandoned. If you are at all interested in the human brain, computers, or intelligence, you won't want to miss this book.
|
page