Can a computer learn to play backgammon at a world-class level? This paper explores the application of temporal difference (TD) learning to the game of backgammon, resulting in the development of TD-Gammon. The authors discuss the potential of complex board games, such as chess and backgammon, as ideal testing grounds for artificial intelligence and machine learning. TD-Gammon is presented as a pioneering effort in this domain. The success of TD-Gammon demonstrates the power of temporal difference learning in mastering complex tasks. This approach contributes to the advancement of machine learning algorithms capable of achieving expert-level performance in challenging domains. The techniques and insights gained from TD-Gammon have influenced subsequent research in reinforcement learning and game playing.
Published in Communications of the ACM, this paper's exploration of temporal difference learning and its application to backgammon fits well with the journal’s focus on advances in computer science and artificial intelligence. The article likely resonated with the ACM readership interested in game-playing algorithms and machine learning techniques.