Summary and Contributions

Abstract

This book presents the results of our research aimed at increasing the scalability and performance of game-tree search in computer chess. Game-playing programs generally push complicated search tasks to their limits and computer chess, in particular, is widely regarded as the “drosophila” of tree searching in the field of Artificial Intelligence. Modern chess programs are sophisticated depth-first searchers that rely on many refinements of the alpha-beta paradigm in order to reduce the sizes of their search trees. All decent programs cache as much data as possible in different hash tables and perform complex dynamic move ordering during the search. Most programs are only mildly selective in the so-called “full-width” parts of their searches where they apply various search extensions and limited forward pruning. In view of the abundant related research our work concentrated on three main areas: (a) selective pruning, (b) the efficient application of game-theoretical knowledge during the search, and (c) the behaviour of the search at increasing depths. Throughout the work our high-speed chess program DarkThought [98] (see Appendix A) served as a realistic test vehicle.