By Interestana AI Editorial — AI-drafted, human-overseen. How we report
AI Cracks Stratego, Defeating Top Human Player

An artificial intelligence named Ataraxos has achieved a significant breakthrough in the domain of imperfect-information games by defeating Pim Niemeijer, widely considered the best human Stratego player, with a decisive score of 15 games to one, alongside four draws. This accomplishment, detailed by a collaborative team of researchers from Carnegie Mellon, MIT, New York University, and Stanford University, marks the first time an AI has reliably conquered Stratego, a game that had previously resisted advanced artificial intelligence efforts, even from DeepMind.
Stratego is a classic board game where each player commands an army of 40 pieces, each with a specific military rank ranging from marshal to spy, along with bombs and a flag. The objective of the game is to capture the opponent's flag. A key characteristic of Stratego is its "imperfect information" nature: players can see the positions of their opponent's pieces but not their identities. These identities are only revealed when two pieces engage in combat, at which point the lower-ranked piece is removed, and the winner's identity is disclosed. This mechanic makes Stratego similar to games like poker, which have previously been mastered by AI.
Eugene Vinitsky, a researcher at NYU and a co-author of the study, highlighted the unique challenge presented by Stratego, describing it as "a massive amount of hidden information that unfolds over a very long time scale." The development of Ataraxos involved a relatively modest computational resource, requiring only 16 GPUs and an estimated training cost of a few thousand dollars. This contrasts with the extensive resources often associated with developing advanced AI for complex games.
Prior to this achievement, AI had demonstrated considerable success in games with perfect information, such as chess, where IBM's Deep Blue defeated Garry Kasparov in 1997, and Go, where DeepMind's AlphaGo triumphed over Lee Sedol in 2016. Furthermore, AI has also proven adept at poker, a game with imperfect information, by consistently outperforming professional players for years. The success of Ataraxos in Stratego signifies a notable advancement in AI's ability to strategize and adapt in environments with incomplete data and long-term planning requirements, potentially paving the way for applications in other complex, real-world scenarios where information is not fully known.
Original source — read the full reporting at the publisher:
Read on Ars TechnicaGet the weekly AI digest
AI news + new model releases, weekly. Drafted by our agents, reviewed by humans.