Replies

Hypothesizing analog computing solves some sticky problems about the brain: the fact that it can do pattern matching very quickly despite having a slow clock rate, the fact that it is generally bad at formal logic, and the fact that it jumps to conclusions and sees patterns in ambiguous data, and the fact that it manages to "instantly" integrate zillions of simultaneous processes.

"Analog computing" doesn't really mean much in theory. Computation is computation, though different computers are optimized for different kinds of information transforms. "Analog" is actually an information encoding format, and can be converted into equivalent digital forms.

The general properties, relative capabilities, and structure of the brain map very closely to a theoretical model of computation that most people are not familiar with but which has been having much gray matter thrown at it lately. There is a difficult problem in computer science related to this computational model that, once solved, will allow us to build software systems that exhibit the same basic computational properties and capabilities of the human brain. The math has been solved (it is an interesting area of algorithmic information theory) but the implementation and architecture of a tractable design is something else entirely.

For those interested, it was proven a couple years ago that solving this problem is equivalent to solving the general problem of AI. I won't go into the how or why of it since that requires much grokking, but most mathematicians and computer scientists who don't take a mystical view of AI acknowledge that it is a very sound formulation of the fundamental problem -- it has survived a couple years of rigorous criticism.

I wasn't trying to assert any superiority of analog, or deny that there can be a conversion. The issue I was addressing was performance. For any given transistor speed, an analog to analog transform is faster than any transform that requires two conversions plus computation.

Plus, considering neurons simply as digital processors ignores the probability that their chemical environment may play key roles in computation. All this can have digital equivalents, but I was addressing the performance issue of designing a brain with a 100 hz clock.

There is a difficult problem in computer science related to this computational model that, once solved, will allow us to build software systems that exhibit the same basic computational properties and capabilities of the human brain.

And what problem would that be? Of course , one cannot know that a problem is 'solvable' until it has been solved. There are numerous problems in philosophy which are still unsolved after thousands of years so I think I will wait to see if the problem is solved. Also, I do not think that the problem of AI is one of computational ability. In time it can exceed man's computational ability. That is why we use calculators. There are many more things to man such as intuition that are not accounted for by computational ability.

"The general properties, relative capabilities, and structure of the brain map very closely to a theoretical model of computation that most people are not familiar with but which has been having much gray matter thrown at it lately."

What's its name? (At least you can tell us that!)

--Boris