The Tragedy of Bobby Fischer
online.wsj.com
online.wsj.com
That said, maybe it's time to stop using chess as a measurement of human intelligence. Games like Go and contract bridge are still uncracked by AI.
The entire chess theory that tries to cut branches so to say, and find good "strategies" for chess - those methods to my understanding change quite often. Every new chess genius turns the theory upside down. Which only means there is no strategy in chess except tree search.
Correct me if I'm wrong.
Although both humans and AIs are essentially doing a tree search, the methodology is sufficiently different that the resemblance becomes superficial.
An AI's heuristic for evaluating a position is fast and crude, sometimes as crude as just counting material. Instead it spends its processing power on looking as deep as possible.
For humans, it's all about the heuristic. We evaluate a position by examining the "shape" of the board in a way that we don't know how to express well enough to translate into code. The brain circuitry we're using for this is visual/spatial. Whatever we're doing is not just a deeper tree search: we simply don't have the right wiring that we'd be able to do that at a pace that yields the ability level we're getting. We're only fast enough to examine a few positions per minute. An AI that ran at that pace would be laughably easy to beat.
Also, the order in which we search is more like "most interesting first" than "best first": we search the branches which our intuition has the most difficult time predicting the outcome of in advance.
There is an interesting mechanism in our minds: when we find an algorithm for doing something and we do it often, the brain decides to hide it under the water - sort of passes it over to our subconsciousness. After that the algorithm or whatever calculations are performed there, in fact much, much faster - sometimes in orders of magnitude. This starts with walking, our first steps, and then continues on in whatever we do in our lives. Chess is no exception. Good hackers can intuitively - and very quickly - find bugs, etc. That's just subconsciousness with its high-performance computer. The only thing subc. needs from us is a well worked out algorithm - that's all.
So our intuitive moves or evaluations are just a result of a faster tree search which is performed "under the water". We are not aware of it - it's a subconscious process.
As for spatial vision of the board - of course we see the board and the position in its entirety, it's a spatial game after all. But whether this vision is of any help in deciding the next move in chess is a big question for me. Or show me exactly how that works. We sometimes can visually evaluate the position, but how accurate is that assessment? Something tells me that purely visual evaluations can be (and often are) wrong.
And yes, AI's heuristic is terrible, but the thing is, AIs (and why on Earth we call them AI?) do heuristic evaluation to cut branches in the tree search. If thorough search would be feasible, AIs would be able to give very precise evaluations, obviously. Meanwhile, we pre-program our AIs with best chess heuristics known today - something that can be beaten easily by genuises like Bobby Fischer.