Causal Models for Evaluating Combinations of Patterns in Chess

In most algorithms for game playing, evaluation is based on a linear combination of features. We argue that this is insufficient in complex games like chess. We hypothesize that the influence of a feature or pattern on the game result is completely determined by a limited number of advantageous effects. These relations follow the markov condition. The evaluation of a pattern in a certain context should be made by testing if the effects of the pattern can be attained during the continuation of the game. A causal model captures all information for this evaluation. This is demonstrated for combinatorial situations and for game analysis.