Chess as a bridge between human cognition and artificial intelligence

Workshop Description

Chess presents an intriguing case study in complex decision-making and multi-step planning. Historically, the psychology underlying chess play has been a topic of great interest, with chess even being referred to by Chase and Simon as the “Drosophila of psychology” - a standard task around which knowledge can accumulate much like model organisms in biology. Despite seminal work focused on expertise, the promise of chess in cognitive science has not yet been fulfilled, primarily due to its intractability for computational modeling. However, two recent developments have opened the door for more precise characterizations of human reasoning in chess. In particular, the popularity of online chess platforms has resulted in large-scale data sets, and modern chess engines provide better tools for developing detailed computational models of human decision-making and planning in this setting. In parallel, the field of artificial intelligence similarly has a long history of work related to chess. Early efforts such as DeepBlue, a chess-playing computer that implemented a form of heuristic search, have been complemented by contemporary advancements such as AlphaZero, a system that can achieve superhuman performance by combining artificial neural networks with tree search. These cutting-edge methods have been used in the chess community to create publicly available chess engines such as Stockfish, and there have been attempts to build such engines that are optimized to match human play rather than performance. Thus, chess is now at the forefront of research in both cognitive science and artificial intelligence.

This workshop will bring together researchers using chess to understand human cognition and build state-of-the-art artificial intelligence systems. Topics will include: (1) superhuman chess play and the mechanistic interpretability of artificial intelligence that can do so, (2) algorithms that are specifically designed to be more human-like in their performance, and (3) characterizing human behavior in chess and similar two-player combinatorial games. A key goal of the workshop is to establish deeper connections between these two fields by addressing open questions and challenges that may benefit from interdisciplinary collaboration.