Copyright: 2004
Publisher: Times Books
ISBN: 0-8050-745602
Jeff Hawkins is the founder of both Palm Computing and Handspring. His lifelong fascination with the brain and how it works led to his study of neuroscience and ultimately to this book. In "On Intelligence", Hawkins presents what he believes is a solid framework for understanding how the brain works. Specifically he explores what it means to "have intelligence." Many of the artificial intelligence efforts underway today focus on ways to mimic the way the brain behaves without regard to how it actually works. Hawkins manages to make a fairly complex and detailed topic manageable for those of us who aren't neuroscientists but are still interested in understanding our brains!
The Premise
Behavior is a manifestation of intelligence but it is not the central or primary characteristic. Thus, those scientists seeking to create computers that simply behave like humans are barking up the wrong tree. According to Hawkins, prediction, specifically prediction based on hierarchical memory, is the essence of intelligence. Thus, until we make a machine that can make predictions based on invariant memory patterns, we haven't really created an intelligent machine.
Cognitive Wheel
Backing up a bit, Hawkins points out that many scientists have been approaching intelligence from the "cognitive wheel" perspective. They argue that as humans we have learned to travel as fast as cheetahs (actually faster) but we didn't do it by learning to run like a cheetah or building a machine that could. Instead we invented the wheel. The wheel does not occur in nature and yet we have used it to exceed nature's built in speedsters.
The cognitive wheel approach says that just like we didn't learn to run fast like a cheetah, we won't replicate the brain in creating artificial intelligence. Instead we need the cognitive equivalent of a wheel invention.
Hawkins effectively rejects this notion by pointing out that until we understand how the brain really creates intelligence, we can't really start inventing our wheel because we don't know yet what intelligence is.
What Intelligence is Not
There are many aspects of being human that are not necessary for intelligence. For instance, we do not need to build machines that experience hunger in order to create an intelligent machine. In fact, emotions and other human qualities are by-products of our specific implementation of intelligence but are not necessary for intelligence to exist.
Another interesting concept that Hawkings brings up is that the brain does not really computer solutions. We often think of the brain as just another computer and yet we know, given the physical limitations of axons and how fast they can fire, that the algorithms for solving complex problems simply won't fit in a "brain as a computer" model.
For instance, when shown a picture of a dog and another of a cat, humans can recognize the two in less than a second. No computer can do it that fast, if one can do it at all! Computer scientists like to say that the problem is learning how to parallel computer problems or make faster computers. Realistically, computers are already faster than the brain in terms of how fast they can perform calculations. In addition, even the most parallel systems cannot chunk this simple recognition problem down to the less than 100 steps that the brain can perform in under a second.
Pattern Recognition
So how does the brain recognize the difference between a dog and a cat? In a nutshell, Hawkins argues that the brain doesn't compute the solution but rather recalls it from memory. By recognizing patterns and pulling the closest match from memory, the brain doesn't need to compute a solution, it simply pulls out the one it recognizes.
So how does the neocortex store it's memories?
- It stores sequences of patterns (temporal as well as spatial memory)
- It recalls auto-associatively (builds on it's own memories)
- It stores patterns in invariant forms (so "close" matches count)
- It stores patterns in a hierarchy (low level patterns get names that are passed to higher levels)
Your conscious flow then is a flow of pattern recognition. Thus your brain can ignore most of the stimulus that comes in without being consciously aware of it... it simply fits the pattern that is predicted. Only when something out of the ordinary occurs do you need to become conscious of the unusual event. This leaves your consciousness available for imagination and prediction activities that lead to intelligent behavior.
HCI Implications
Interestingly Jeff Hawkins pointed out that many human computer interaction specialists argue that the computer should always adapt to the user, rather than the user adapting to the computer. He disagrees with this assertion.
As part of his work with the Palm hand held computers, Hawkins devised Graffiti, a pseudo handwriting recognition program. Rather than trying to write software that would recognize everyone's unique handwriting style, Hawkins came up with a simple to learn set of pen strokes that humans could easily learn and computers could easily recognize. He came upon this idea when he realized that computer keyboards are not intuitive for humans and yet millions of people learn how to touch type because it is a skill that is very useful. In a similar fashion, thousands of people have learned Graffiti (including myself) because it is a useful skill when you have a Palm device.
As a software developer I am constantly striving to make my software easier for humans to use, but Hawkins' point is well taken. There are times when a human is simply better at something than the computer is, so it makes more sense to have the human adapt than the computer.
Conclusion
I suppose I am a bit of a geek and as such I love this kind of book. It digs deep into a field that is fascinating, the brain and how it works. Along the way Hawkins provides some very interesting insights into human behavior. If you are looking for "the answer" you won't find it in this book. Hawkins even says several times in the book that he has no illusion that this book contains the final answer. Rather he presents it as a possible framework for developing the final answer. If you like reading about the brain, how it works and if you like pondering things like "what does it mean to be intelligent", then this book is for you!