Like communications, the problem with intelligence is the illusion that it exists...
Jeff Hawkins, the person behind PalmPilots and the Redwood Neuroscience Institute, thinks a lot about thinking. In this book, Hawkins addresses two of his loves - and fields of study that have been moving closer together with time - neuroscience and computers. Most people recognize that computers are getting smarter, but don't necessarily understand what this really means, or how this might accelerate. Hawkins's book attempts to describe what intelligence is, so the common ground for this merger is well defined. He describes the key part of the mind as 'auto-associative memory' which essentially is information retrieval based upon pattern recognition. This capability allows the brain to 'fill in' missing information, and make predictions about the future based upon those patterns. This points to Hawkin's biggest difference from others about what constitutes intelligence - he sees it not as a functional capability as much as an ability to learn patterns and leverage them in reasoning about the future.
Hawkins points out that in performing situational awareness (such as we do when walking in a room) a lot of what we do is pattern recognition to identify objects, and focus on 'surprises' (why doesn't that person have his pants on, for example). If the pattern matching is relatively easy (we're familiar with similar circumstances), and we only have a few disconnects to reconcile, we quickly analyze those individual situations and decide what they mean (laugh, leave, etc). If we don't have experience to draw upon, because the situation is so novel or incongruent with our experience, we can easily become overwhelmed, because we don't know what to do. Thus, an important aspect of 'intelligence' is effective pattern-matching, separation of concerns between what's known and what isn't, and prioritization of action to determine what to do to collect new information to fill in the gaps. To the extent we can fill those gaps quickly, we're said to 'be intelligent'.
Combine this auto-associative memory with hierarchical processing and analysis layers (to rank among alternatives at different levels of abstraction, and thus trigger different levels of attention), accumulating knowledge stores, and temporal aspects (an important element in all our senses), and you've defined what Hawkins believes the neocortex does. His book trys to explain this in layman's terms, and reviews facinating details, such as the effect of saccades on vision. These, along with blind spots, are used to higlight how the brain works, and demonstrate that it operates less like a computer than like a neural network.
Along the way, he reviews why traditional 'brute force' approaches to AI haven't worked... simply put, they haven't modeled these pattern recognition and learning approaches of the human mind, but have attempted a more functional approach. This focus changing - a good example of the potential of learning is Google's approach to language translation; instead of trying to semanticly analyze every language component, they instead feed lots of data into learning algorithms, and derive a solution from that.
Of course, understanding language is still a legitimate goal.. this is all about how you get there. But brute force approaches are more feasible with more compute horsepower, and Google will have the horsepower. The next step, at least according to the ideas in the book, would be a more effective architecture, and a logical step in that direciton is to perform some simulations and analyses. Producing a complete functional simulation of the human brain is underway - for thoe hobbyists among you, downloads are available!
