New AI Achieves Superhuman Performance in Stratego, Potentially Aiding Decision-Making
Researchers have developed an AI system that outperformed top human players in the board game Stratego, marking a significant advancement in AI for games with hidden information. This system, named Ataraxos, could be applied to real-world scenarios such as military strategy and business negotiations, where decision-making often involves incomplete information.
A new AI system called Ataraxos has been developed by researchers from multiple universities, including MIT and Carnegie Mellon, to excel in the board game Stratego, which involves hidden information.
Ataraxos defeated top human players with a record of 39 wins and 2 losses at the Stratego world championship and achieved a notable victory against the strongest player with a score of 15-1-4.
The AI uses a combination of self-play reinforcement learning and decision-time planning to enhance its strategic capabilities, allowing it to estimate the identities of hidden opponent pieces and make informed decisions.
The system is designed to be less computationally demanding and cheaper to train compared to previous AI models, achieving superior performance with significantly fewer training examples.
Ataraxos has also been adapted for other games with hidden information, demonstrating its versatility and potential applications in various fields, including military and business contexts.
Future developments aim to improve the interpretability of the AI's decision-making process, ensuring that human users can understand and audit its recommendations.