Re[2]: computer-go: perfect play

[Roland <goprog@xxxxxxx>]

Hello Vlad,

Thursday, September 28, 2000, 11:12:05 AM, you wrote:



VD> As for emulating human play, I don't agree completely. There are (at least)
VD> two main reasons to develop a game-playing program: to solve the game (and
VD> then you are right!), or to gain insight into how the human brain works (and
VD> then even tic-tac-toe is interesting, even if the outcome is known).

I would like to add one more reason: to teach the computer how to
think in abstract terms, which is what a good position analyzer needs.
So the issue is perhaps more how do we teach a program to think like a
human. It's already known how to implement a search tree, but it's
hard to give the program an "understanding" of the game. An example of
a simple abstract concept would be a ladder. Even a go beginner can
understand a ladder and predict it's behavior.

VD> I make a correction for an earlier affirmation I made: it is true that an
VD> empty board is either a winning position for black, for white or a draw
VD> (komi excluded). This is true for any deterministic game with complete
VD> information - so in fact any of them are boring for the perfect player. The
VD> conclusion is then that GoGod and ChessGod are playing bridge!? ;-)

Good point!




-- 
Best regards,
 Roland