New Player of Games AI Mastering Both Perfect and Imperfect Information
Researchers introduce Player of Games, a general-purpose AI algorithm that combines search, self-play, and game-theoretic reasoning to master diverse game types. It achieves strong performance in chess, Go, poker, and the complex imperfect information game Scotland Yard.
Games serve as a long-standing benchmark for artificial intelligence progress, but until now, no single algorithm excels across both perfect and imperfect information games. Researchers introduce Player of Games, a general-purpose algorithm that unifies previous approaches by combining guided search, self-play learning, and game-theoretic reasoning into one cohesive system.
This unified approach allows the AI to achieve strong empirical performance across large perfect and imperfect information games, marking a major step toward truly general algorithms for arbitrary environments. The researchers prove that the algorithm is sound, meaning it converges to perfect play as its computation time and approximation capacity increase.
In practical tests, Player of Games demonstrates strong performance in classic perfect information games like chess and Go. Additionally, it beats the strongest openly available agent in heads-up no-limit Texas hold'em poker and defeats the state-of-the-art agent in Scotland Yard, an imperfect information game that perfectly highlights the value of combining guided search, learning, and game-theoretic reasoning.