Applying genetic algorithms to game search trees

Abstract

 Computer game playing is an important artificial intelligence research field in that the results can usually be applied to other related fields. One of the key computer-game-playing issues is designing effective search algorithms. Traditional search algorithms incur great temporal and spatial complexities when exploring deeply into search trees to find good next moves. Searches are thus usually not deep enough to derive good playing strategies. In this paper, we focus on one-player game search trees, and propose a genetic-algorithm-based approach to enhancing the speed and accuracy of game tree searches. Experiments show that our algorithm can improve solution accuracy and search speed.