Ataraxos AI Defeats Stratego Champion Pim Niemeijer in Historic Match

Ataraxos, an AI developed by researchers from Carnegie Mellon University, New York University, Stanford University, and MIT, achieved a remarkable victory by winning 15 games, losing 1, and drawing 4 against Pim Niemeijer, the most successful Stratego player of all time. Niemeijer, who has held the top ranking for over 600 weeks and boasts 4 world championships, faced Ataraxos in a series of matches that showcased the AI's capabilities.

The training cost for Ataraxos was less than $8,000, significantly lower than the estimated $3 million to $4.5 million required for DeepMind's previous AI, DeepNash, which had won 19 of 28 games at the 2023 world championship. Ataraxos was trained in just one week using 16 Nvidia H100 GPUs, followed by an additional 4 days on 4 GPUs, demonstrating its compute efficiency at 1/500th the cost of DeepNash.

Ataraxos's effective win rate against Niemeijer, accounting for draws as half a win, stands at 85%. This achievement marks Ataraxos as the first AI to reach superhuman performance in Stratego, as noted by Carnegie Mellon researchers. The AI's ability to handle the game's complexity, with over 10^33 possible setups, indicates a significant advancement in reinforcement learning and search techniques, overcoming challenges previously thought insurmountable.

In an exhibition prior to this match, Ataraxos won 38 out of 40 games against tournament players and secured 4 series victories against top-ranked competitors. This follows a trend of AI systems increasingly dominating traditional board games, further diminishing the human stronghold in competitive play. As noted by Max Roelofs, a three-time world championship runner-up,