AIChessCoach: Software development aimed at competitive level chess players using Generative AI

Minaya P.; Moreno L.; Salas C. · Proceedings - 2025 IEEE International Conference on Future Machine Learning and Data Science, FMLDS 2025 (2025) · DOI: 10.1109/fmlds67896.2025.00015

Resumen

Current chess engines have reached superhuman levels of performance, but their usefulness as pedagogical tools is limited, especially for tournament players seeking to understand the reasoning behind the moves. These systems offer accurate numerical evaluations, but lack accessible explanations tailored to the level of the user. In response to this limitation, we propose AIChessCoach, a web application that combines the analytical capabilities of Stockfish with a generative natural language model, with the goal of generating understandable explanations of complex positions. The system is designed to emulate the interaction between a human coach and a player, providing detailed analysis and contextualized tactical and strategic reasoning. To validate its effectiveness, an experimental comparison was made between AIChessCoach and ChessBuddy, using tactical problems classified by difficulty level as: mate in 2, 3 and 4. The results show that AIChessCoach achieved an average performance of 80.43%, surpassing ChessBuddy (61.83%) by 30%, and providing clearer and more substantiated explanations, even in highly complex positions. This approach proves to be a promising alternative for AI-assisted learning in chess, promoting a deeper understanding of positional analysis and facilitating personalized and effective training.

Instituciones: Universidad Peruana de Ciencias Aplicadas