A world championship caliber checkers program

Reviewer: John Thomas Ritschdorff

Checkers has never received the amount of attention from artificial intelligence researchers that has been bestowed upon the game of chess. Although Samuel's work is well known, research into computer checkers has not been as extensive as may be warranted. The authors of this paper propose to remedy this. They present their findings from the development of a checkers-playing program named CHINOOK. One point of this work is to reestablish the usefulness of checkers as a task domain for research in game playing, search techniques, and knowledge representation. Checkers is a domain with less complexity than chess, with fewer types of pieces, results that are more easily achieved, and a structure similar to that of chess. Differences between the games include the notion of safe moves, the average number of available moves, and the subtle distinctions that exist when recognizing a win or a draw. The authors consider specialized search needs for checkers programs that employ endgame databases and the development of evaluation functions. Depth of search is compared to human lookahead capabilities. The encoding of grandmaster opening and ending games is discussed. The experience of versions of CHINOOK in competition is also reported. This paper is interesting and easy to read. Anyone who has some knowledge of artificial intelligence terminology and goals can appreciate the authors' work and intentions. Students of artificial intelligence would benefit from this paper. It would provide them with information on current research that exemplifies concepts and techniques that they are currently studying. The authors have successfully embedded their specific work in the general framework of artificial intelligence concerns and provided enough details to inform without overwhelming the reader. A fairly comprehensive reading list is include d.