I am half way through a book on perception (A Second Way of Knowing by Edmund Blair Bolles, copyright 1991). It makes some interesting statements about the way the brain works, contrasting it with a computer. It touches a bit on AI. I thought some of you might be interested in a few of the ideas mentioned.

Following are some of the statements I found interesting. Some of the statements are exact quotes, other are paraphrases. I have omitted all sorts of stuff from the book which was written between the statements given below. I have added a few comments of my own.

For several thousand years, philosophers assumed than the brain must somehow store data about whatever is learned. In the jargon of the computer age, there must be an (indexed) data base that preserves what we learn. A later statement says that there is no memory storehouse in the brain — The billions of processors in the brain regularly reprogram themselves --- There is probably no indexer for locating items in this nonexistent storehouse.

Cell assemblies violate two basic features of computers. They change the physical material itself (computer programs do not change the hardware). They involve no software.

There is a description of a research project started in the seventies, whose goal was a computer which would read and understand what it read. They stated that reading and understanding a story is one of the most difficult tasks in the world. There was no description of what criteria would be used to conclude that they had succeeded. They never even started working on their final goal.

They hit an immediate snag when they realized that when humans read, they use a vast number of assumptions based on previous experience. This requires a database far more extensive than what is contained in a dictionary. Furthermore, the "database" changes dynamically. Some examples follow (my own examples: they gave none). The word "squeeze" has a meaning associated with the game of bridge. The dictionary definition is one sentence. In a book on bridge, the term requires almost a page of fairly terse statements. If you look up words like conservative, liberal, moderate the dictionary definitions do not change much with time, but the concepts they invoke do differ quite a bit as time goes by. Furthermore, the dictionary definitions do not invoke the vast number of examples and extra data used by a human brain when encountering such words. A moderate Republican is not much like a moderate democrat or a moderate Muslim or a moderate Chinese communist.

I once worked with some linguists who were writing translation software. Consider the following: The airplane flies like a bird; The fruit flies like a banana. English is one of the worst languages in this respect. In one sentence, "flies" is a verb; In the other it is a noun. When I read one sentence first, the other initially invokes thoughts of apples, oranges, & bananas with little wings. When I read them in reverse order, the second makes me imagine a mechanical device chewing the wings off an airplane. Computer translation programs have a terrible time with stuff like this, and translation is a lot easier than understanding for a computer.

By the late eighties, they ended up being satisfied with computer simulation of a neural network which reacts to a few different smells.

Simulating parallel actions on a serial computer is easy enough, but it slows down the simulation. In some of their latest experiments (in 1990), they simulate a second or two of events among ten thousand cells. It takes the computer all night to do the calculations. If you consider that the brain has billions of cells, even with a faster modern computer the AI boys have got some serious problems.

One simulation program showed some of the brain's most confusing characteristics. The model spontaneously created a six-pulse-per-second theta rhythm. Even more astonishingly, it showed the brain's confusing tendency to respond in different ways to identical inputs.