Book review: Advances in Computer Chess 4. Edited by D.F. Beal (Pergamon Press 1986 )

Abstract

This book contains the papers that were presented on the fourth conference on Advances in Computer Chess, held at Brunel University, London in April 1984. Together with the main tournaments, this tri-annual conference is the main event in the computer chess world. The 13 contributions cover a broad spectrum of topics in computer chess, and some are apprehendable with only elementary chess knowledge. Among these is an article by Kopec, Newborn and Yu, on the comparison of cognition of selected chess positions between humans and computers. More extreme positions, in which it is argued that real understanding (Intelligence) is needed instead of calculating deep variations, are given in 'Artificial Stupidity' by Hartson. The art of problem solving (including phenomena such as self-mates and help-mates) with results of the best programs in this discipline can be found in the article by Lindner. Chess Master David Levy reports his success in defeating the computer world champion Cray Blitz in a 4 game match which was held during the conference. Most interesting in this respect is his description of the special preparation he put to use. In general, his strategy is probably the best method to exploit a thusfar very basic weakness of computer programs. Hyatt et al. present some stunning results of their program, Cray Blitz, which runs on the fastest available computer. They discuss the use of hash tables, and the way in which multi-processing has been taken advantage of. Description and implementation, together with results of multi processing-techniques in two chess programs are described by Marsland et al. Berliner, author of Hitech, describes in his paper the application of special hardware (VLSI) in the design of an ultra-fast move generator that produces all pseudo-legal moves for a given position in one clock cycle (250 nanoseconds). He furthermore shows results of the B* tree-search algorithm, and describes another one of his programs: Patsoc. A new trend in computer chess cognition seems to be the notion of 'chunks': Groups of pieces that can be identified as one entity. Berliner describes this for pawns in the endgame, and Brakto et al. devote their entire paper to this subject. Chunking is thought to be important, because humans identify them as such which enables strong players to correctly reproduce a given position after only a short moment of examination. A strongly represented subject is the generation and induction of patterns and plans. Michie presents his 'Michie Road' in which rules are to be generated from examples provided by an expert. In some cases (special endgames) this expert can be a computer. Reidel's contribution describes the deduction of patterns in the King+Rook versus King endgame. Owsnicki and von Luck, both with a thesis on computer chess, discuss a method of plan construction with results from its application in a program called N.N. They illustrate three catagories of possible errors, each with its own cause and each with a different level of solvability. Shapiro and Michie describe a self-commenting facility whereby expert supplied text fragments attached to individual attributes are recognised to form run-time explanations of decisions made by rules previously induced. A theoretical paper is given by Schrueter. He analyses the apparent paradox that experience shows that a deeper tree-search yields better play, whereas mathematical investigation of the problem predicts less reliable results for deeper searches.