Re: computer-go: A problem with understanding lookahead

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

   From: "David Mechner" <mechner@xxxxxxxxxxxxxxxxx>

   If you are saying that search is useful when you have a sufficiently good
   evaluation function, and when the structure of the game you're playing is
   well-behaved, then we have no disagreement.

   My only point was that it's interesting that in computer chess, search using
   an evaluation much cruder than expert human positional judgement leads a
   program to better positions not just by its crude evaluation function, but
   also by the more sophisticated judgement of the human.

   -David



Yes,  that is very interesting.   I've seen my  program easily outplay
humans who had far more "understanding"  than the program did.  But as
I  mentioned before, knowledge can be   "simulated" by search, because
concepts can be learned by "demonstration."   (if you have a weak pawn
and don't know it, I'll show you by  either winning it, or forcing you
to defend it while I do something else!)



Don