Machine learning using a genetic algorithm to optimise a draughts program board evaluation function

The paper reviews the authors' work in using a genetic algorithm (GA) to optimise the board evaluation function of a game playing program. The test bed used for this study has been the game of draughts (checkers). A pool of draughts programs are played against each other in a round robin (all-play-all) tournament to evaluate the fitness of each 'player' and a GA is used to preserve and improve the best performers. Some solutions to the problems of attempting to compare the absolute performance of possible solutions in this area which is mainly about relative abilities are presented. Comparisons with classical methods and results are also briefly discussed.