AI beats Stratego's greatest player, ending one of the last human strongholds in board games
decoder ·
Researchers from Carnegie Mellon, NYU, Stanford, and MIT have developed Ataraxos, an AI that achieves superhuman performance in the board game Stratego. In a 20-game series against Pim Niemeijer, widely regarded as the greatest player in history, Ataraxos secured 15 wins, one loss, and four draws. The system was trained for less than $8,000 using 16 Nvidia H100 GPUs, costing roughly 1/500th of what DeepMind’s previous attempt required. Unlike earlier poker-derived methods, Ataraxos handles Stratego's massive hidden information space through self-play regularization and a belief network that predicts opponent positions. The AI demonstrated its strength by winning 38 of 40 games in an exhibition at the 2025 World Championship and excelling in other imperfect information games like Hanabi and Dou dizhu. This success suggests that large-scale hidden information is no longer a barrier for reinforcement learning, potentially opening doors for AI applications in finance and military strategy. Although Ataraxos played with a structural disadvantage regarding adaptation, its effective win rate of 85 percent remains unprecedented at the highest competitive level.
decoder