The computer plays against you, say defending the rook side of a KQKR chess endgame. Initially, it simply uses the most stubborn defense (probably Distance To Conversion), but when it detects you have learned how to play against that, it switches to a different defense designed to confuse your strategy.
How can the computer detect you have learned a certain strategy, when the space of possible strategies is infinite? I think the saving grace is that the space of possible board positions, and board motifs, is not infinite, and the computer can detect you are steering toward or avoiding certain motifs.
2 comments :
Given that the space of possible board positions (and indeed possible ways the game can play out) is finite (albeit large), is it really valid to say that the space of possible strategies is infinite? I think of a "strategy" as a description of what move to make from any possible position, so the number of possible strategies seems finite (albeit large). Are you considering different representations as distinct (e.g. "always check the opponent's king" is different from "always put the opponent's king in check") to get infinitely many strategies?
That is a good point, and the question I am posing remains vague in my head. On one hand, the number of game theoretic strategies is finite but so huge as to be useless as a direct method of implementing an adapting AI opponent. On the other hand, it could mean all possible representations of strategies inside a human's brain, which while not infinite, seems impossible to precisely define.
Post a Comment