Modeling Strong and Human-Like Gameplay with KL-Regularized Search
This addresses the challenge of building AI agents that are effective yet understandable and cooperative for humans in strategic games, representing an incremental advance over existing methods.
The paper tackled the problem of creating policies that are both strong and human-like in multi-agent games, showing that KL-regularized search improves human prediction accuracy and strength over imitation learning alone in chess and Go, and a novel regret minimization algorithm in Diplomacy matches human prediction accuracy while being substantially stronger.
We consider the task of building strong but human-like policies in multi-agent decision-making problems, given examples of human behavior. Imitation learning is effective at predicting human actions but may not match the strength of expert humans, while self-play learning and search techniques (e.g. AlphaZero) lead to strong performance but may produce policies that are difficult for humans to understand and coordinate with. We show in chess and Go that regularizing search based on the KL divergence from an imitation-learned policy results in higher human prediction accuracy and stronger performance than imitation learning alone. We then introduce a novel regret minimization algorithm that is regularized based on the KL divergence from an imitation-learned policy, and show that using this algorithm for search in no-press Diplomacy yields a policy that matches the human prediction accuracy of imitation learning while being substantially stronger.