AI Model Learns to Mimic Individual Chess Champions' Styles

Researchers propose a new architecture that adapts the unified Maia-2 model to champion-specific embeddings, enhanced by a limited Monte Carlo Tree Search (MCTS) process for tactical exploration. This addresses the challenge of emulating individualized decision-making styles of historical chess champions, moving beyond general population models based on skill levels. The standard evaluation metric, move accuracy, is criticized for penalizing natural human variance and ignoring long-term behavioral consistency. A novel behavioral metric based on the Jensen-Shannon divergence is introduced to robustly evaluate stylistic fidelity.