Pattern-based representations of knowledge in the game of chess

✍ Scribed by M.A. Bramer

• Publisher: Elsevier Science
• Year: 1982
• Weight: 562 KB
• Volume: 16
• Category: Article
• ISSN: 0020-7373
• DOI: 10.1016/s0020-7373(82)80050-3

✦ Synopsis

The focus of recent Artificial Intelligence research into computer chess has been on endgames. These afford the possibility of controlled experimentation, whilst retaining much of the complexity of the full game of chess. This paper discusses some of the specific reasons for complexity in the endgame and considers its effects on human chess-playing strategy, textbook descriptions and the development of programs. In programming the endgame the researcher is faced with a range of decisions concerning the quality of play to be aimed at, the balance between knowledge and search to be adopted and the degree to which the playing strategy should be understandable to human chess-players. A model for representing pattern-knowledge is described which has enabled the development of algorithms to play a number of cndgames. Three algorithms representing different levels of performance for the endgame King and Pawn against King are compared, in order to discuss the tradeoff between complexity and completeness, on the one hand, and compactness and comprehensibility, on the other. Finally, the role of search in reducing the amount of knowledge to be memorized is considered and an extension to the basic model to incorporate deeper search is discussed.