© 1981 by British Computer Society
| ||||||||||||||||||||||||||||||||||||||||||||||||||
A theory of evaluative comments in chess with a note on minimaxing
Machine Intelligence Research Unit, University of Edinburgh, Hope Park Square, Meadow Lane, Edinburgh, UK
Classical game theory partitions the set of legal chess positions into three evaluative categories: won, drawn and lost. Yet chess commentators employ a much larger repertoire of evaluative terms than this, distinguishing (for example) a ‘drawn’ from a ‘balanced’ position, a ‘decisive’ from a ‘slight’ advantage, an ‘inaccuracy’ from a ‘mistake’ and a ‘mistake’ from a ‘blunder’. As an extension of the classical theory, a model of fallible play is developed. Using this, an additional quantity can in principle be associated with each position, so that we have not only its ‘game-theoretic value’ but also its ‘expected utility’. A function of these two variables can be found which yields explications for many evaluative terms used by chess commentators. The same model can be used as the basis of computer play. It is shown to be easier to justify, and to adjust to realistic situations, than the minimax model on which state of the art chess programs are based.
Received June 1980.
* Machine Intelligence Research Unit, University of Edinburgh, Hope Park Square, Meadow Lane, Edinburgh EH8 9NW