Verifying and Codifying Strategies in a Chess Endgame

Abstract

Only a short time ago, the game of chess was a research area in which artificial intelligence (AI) workers without specific chess knowledge could achieve significant results. Notably, AI techniques (representation, search heuristics including their sophisticated refinements, etc.) were tested in endgame domains of three or at most four pieces. Anything beyond this number was practically excluded by the extreme complexity and huge fast-storage requirements. The construction of omniscient databases, as described by van den Herik and Herschberg (1985) and by Thompson (1986), was made possible by the arrival of supercomputers which overcame the previous storage limitations, and which had enough computational power to handle the complexity issue in reasonable time. For a treatment of these issues, see the works of Dekker, van den Herik and Herschberg (1987b), and van den Herik and Dekker (1988).