Thursday, 23 December 2010

The wrong kind of chess computer

Somewhere along the line we missed the point of building chess computers. We were meant to use the game's structure as a way to investigate human pattern recognition, reasoning and decision making. Instead, we powered ahead with brute force algorithms, beating human players and learning almost nothing in the process.

A real, intuitive chess program needs to group pieces together into conceptual formations, evaluate the board considering only about seven possible moves, and think a few moves ahead. Perhaps just placing those limits on chess programs - seven potential moves, maybe seven moves ahead - will start leading us in the right direction.

Mokalus of Borg

PS - Applied to current methods, those limits just make bad chess programs.
PPS - But maybe someday we'll start figuring it out.

No comments: