Go to ICLR 2025 Conference homepageLearning to Imitate with Less: Efficient Individual Behavior Modeling in Chess
As humans seek to collaborate with, learn from, and better understand artificial intelligence systems, developing AI agents that can accurately emulate individual decision-making becomes increasingly important. Chess, with its long-standing role as a benchmark for AI research and its precise measurement of skill through chess ratings, provides an ideal environment for studying human-AI alignment. However, existing approaches to modeling human behavior require large amounts of data from each individual, making them impractical for new or sparsely represented users. In this work, we introduce Maia4All, a model designed to learn and adapt to individual decision-making styles efficiently, even with limited data. Maia4All achieves this by leveraging a two-stage fine-tuning method to bridge population and individual-level models and uses a meta-network to initialize and refine these embeddings with minimal data. Our experimental results show that Maia4All can accurately predict individual moves and profile behavioral patterns with high fidelity, establishing a new standard for personalized human-like AI behavior modeling in chess. Our work provides an example of how population AI systems can flexibly adapt to individual users using a prototype model as a bridge, which could lead to better and more accessible human-AI collaboration in other fields like education, healthcare, and strategic decision-making.