Checkers using a co-evolutionary on-line evolutionary algorithm

The game of checkers has been well studied and many computer players exist. The vast majority of these 'software opponents' use a minimax strategy combined with an evaluation function to expand game tree for a number of moves ahead and estimate the quality of the pending moves. In this paper, an alternative approach is described where an on-line evolutionary algorithm is used to co-evolve move sets for both players in the game, playing the entire length of the game tree for each evaluation, thus avoiding the need for the minimax strategy or an evaluation function. The on-line evolutionary algorithm operates in essence as a 'directed' Monte-Carlo search process and although demonstrated on the game of checkers, could potentially be used to play games with a larger branching factor such as Go.