Search versus knowledge in human problem solving: A case study in chess

Abstract

This paper contributes to the understanding of human problem solving involved in mental tasks that require exploration among alternatives. Examples of such tasks are theorem proving and classical games like chess. De Groot’s largely used model of chess players’ thinking conceptually consists of two stages: (1) detection of general possibilities, or “motifs”, that indicate promising ideas the player may try to explore in a given chess position, and (2) calculation of concrete chess variations to establish whether any of the motifs can indeed be exploited to win the game. Strong chess players have to master both of these two components of chess problem solving skill. The first component reflects the player’s chess-specific knowledge, whereas the second applies more generally in game playing and other combinatorial problems. In this paper, we studied experimentally the relative importance of the two components of problem solving skill in tactical chess problems. A possibly surprising conclusion of our experiments is that for our type of chess problems, and players over a rather large range of chess strength, it is the calculating ability, rather than chess-specific pattern-based knowledge, that better discriminates among the players regarding their success. We also formulated De Groot’s model as a Causal Bayesian Network and set the probabilities in the network according to our experimental results.