Evolutionary Tuning of Chess Playing Software

Abstract:

In the ambition to create intelligent computer players, the game of chess

is probably the most well-studied game. Much work has already been

done on producing good methods to search a chess game tree and to statically

evaluate chess positions. However, there is little consensus on how

to tune the parameters of a chess program’s search and evaluation functions.

What set of parameters makes the program play its strongest?

This paper attempts to answer this question by observing the results

of tuning a custom chess-playing implementation, called

Agent, using

genetic algorithms and evolutionary programming. We show not only

how such algorithms improve the program’s playing strength overall,

but we also compare the improved program’s strength to other versions

of Agent.