I am currently in the process of reading On Intelligence by Jeff Hawkins, and it’s a constant string of “aha!” moments. I could say it was a long time since a non-fiction text blew my mind, but it probably wasn’t. My memory is far too faulty to trust in such matters. It is not a common occurrence, that much I do know.
I have had a fascination with the brain and its workings for a long, long time. Now, this book says it’s about the nature of intelligence, and it is, but mostly it’s about the brand of intelligence we are most familiar with; the one conjured up by evolution. Therefore, the book is pretty much about the human brain. Or rather a part of it, the neocortex.
The structure of the brain as explained by Hawkins is quite easy to understand (at least on a conceptual level; when he talks a lot about “layer 3” and “layer 5” I got a bit lost), mostly because it is so simple and so elegant. The idea that the different areas of the brain aren’t hard-coded for a specific task (such as hearing, sight, etc.) wasn’t that surprising, since people who have suffered strokes have been shown to “move” the function of affected areas to nearby regions. What was a bit surprising, though, was the level of generalization of the neocortex.
I remember reading somewhere some week ago that neurons act a certain way not because they are part of a certain part of the brain, but because they receive certain input. I never quite made the leap to the idea that all tissue of the neocortex is essentially the same, that it is almost exclusively the structure that governs the function. When you think about it, though, it makes perfect sense. Why would evolution produce specific areas of the brain to deal with specific senses? If a similar function can be made with highly generalized modules, then why not? Sure, it has the problem of all generic systems in that it must first be “primed” with input to learn its purpose, but it return a lot of flexibility is won.
The same sort of thinking has been used in the world of computers for a long time. Instead of making many specialized units, it’s almost always better to make a few, robust units that can be combined into a nigh infinite number of complex structures. This is of course not true if speed or size are critical; then flexibility takes the backseat. But since evolution is all about flexibility, it makes sense that the brain is built from many small, identical “blocks”. That each little such block deals with information in the exact same way as all other blocks do seems like such an obvious and efficient way to achieve nested abstractions, the very key to learning and comprehension.
Overall, the book isn’t so much about the structure of the brain as it is about the rules of the structure. A grammar, if you will. A very elegant grammar, at that! The gist of it is that each “block” or “module” of the brain deals with patterns of patterns, or sequences of sequences. These modules are then ordered into a hierarchical structure, with those modules in a higher position receiving the same amount of information as those lower, but more generalized the higher up they are.
For example, the modules perceiving letters sends not the shape of a letter when you see one, but the concept of the letter; no matter how the ‘A’ is written, that module sends an ‘A’. The next module might build several letters it into a word, and so on.
Another key concept was that information passes downward in the hierarchy in the form of predictions. When prediction doesn’t match perception, an error is passed upwards until a unit can resolve the conflict, either by making a ruling on which data is the proper or by forming a memory of this new and unforeseen event (as I understood it). This makes memory and thought efficient processes; concepts are stored as low as possible in the structure, so that they are recognized as quickly as possible. When we learn new things, we must first conciously think about them, but the more we use that memory, the lower in the hierarchy it moves, until we do it as a second nature, no more requiring our consious effort.
I have not read any critique of the book, so I don’t know how valid this model is, but it does look rather elegant, which is always a good sign when it comes to scientific theories.