TL;DR
Get school and study supplies delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
Ataraxos, an AI developed by researchers from Carnegie Mellon, MIT, NYU and Stanford, defeated top Stratego player Pim Niemeijer in a reported match series: 15 wins, one loss and four draws. The team says it trained the system using 16 GPUs and a few thousand dollars, tackling Stratego’s long games, hidden piece identities and bluffing.
A research team from Carnegie Mellon, MIT, New York University and Stanford says its AI system Ataraxos defeated Stratego champion Pim Niemeijer in a match series, winning 15 games, losing one and drawing four. The result marks a reported breakthrough in a game where pieces’ identities are hidden and strategic choices can unfold over thousands of moves; the researchers say training the system required 16 GPUs and a few thousand dollars.
Stratego gives each player 40 pieces of different ranks, along with bombs and a flag. A player wins by capturing the opponent’s flag. Although both players can see where the other’s pieces are, they do not know their identities until pieces fight. The lower-ranked piece is removed, and the surviving piece’s identity is revealed. That combination of hidden information and direct confrontation makes the game different from chess, where the board state is visible to both players.
The researchers’ reported result is a 15-1 record with four draws against Niemeijer, whom the source report describes as arguably the strongest Stratego player in history. The report says the team trained Ataraxos with 16 GPUs and a budget of a few thousand dollars. Those figures describe the reported training effort; the source material does not provide a full accounting of development costs, compute time or the match conditions.
The team says Stratego’s challenge is not just the number of possible piece arrangements. Players also bluff: a weak piece can be moved as though it were a powerful one, while repeated or predictable threats can lose their effect. The system had to make decisions across long games with incomplete information, rather than rely on a short sequence of visible tactical calculations.
A Lower-Cost Route to Stratego
The match suggests that strong play in a game with extensive hidden information may not require the scale of resources commonly associated with headline-grabbing AI victories. The team’s reported use of 16 GPUs and a few thousand dollars contrasts with the high-profile resources behind some earlier game-playing systems. It does not, by itself, establish that comparable results are affordable across other games or real-world tasks.
Stratego also tests abilities that are hard to measure in games with fully visible boards: maintaining beliefs about unseen information, adapting to an opponent and deciding when to mislead. A system that performs well under those conditions is a research result about strategic decision-making. It should not be taken as evidence that AI has solved hidden-information problems generally, or that the same approach transfers directly to settings outside games.
The result is relevant to AI research because the team reports both a strong performance against a highly accomplished human and a comparatively modest training budget. Independent details about the evaluation will help readers judge how broadly that achievement applies.
As an affiliate, we earn on qualifying purchases.
Why Stratego Resisted Earlier Systems
Computers have surpassed leading human players in several prominent games. IBM’s Deep Blue beat chess champion Garry Kasparov in 1997, and DeepMind’s AlphaGo defeated Lee Sedol at Go in 2016. Poker systems have also reached professional level. Those milestones did not settle Stratego, where the opponent’s pieces are visible but their ranks are concealed.
Researchers cited in the source report contrasted Stratego’s information load with Texas Hold’em. MIT computer scientist and study co-author Gabriele Farina said Texas Hold’em has two hidden cards and 1,326 possible hands. In Stratego, he said, the 40 pieces can be arranged in more than a decillion possible ways. The source also notes that a Stratego game can last around 2,000 moves, compared with roughly 40 moves in a typical chess game.
DeepMind introduced its Stratego-playing system DeepNash in 2022, but the source report says it did not reliably beat the best human players. NYU researcher and study co-author Eugene Vinitsky described Stratego as distinctive because a large amount of hidden information unfolds over a long time scale. That challenge combines uncertainty with the need to manage a changing opponent’s expectations.
“There’s something super distinctive about Stratego, which is that it is a massive amount of hidden information that unfolds over a very long time scale.”
— Eugene Vinitsky, NYU researcher and study co-author
As an affiliate, we earn on qualifying purchases.
Match Conditions Still Need Detail
The reported score establishes the outcome presented in the source, but the supplied material does not give the dates of the games, the full match format or details such as how sides and starting layouts were assigned. It also does not describe the human opponent’s preparation or whether the games were played under tournament-standard conditions. Those details matter when interpreting a match record.
The source identifies Ataraxos as the researchers’ system and gives a broad account of its training cost, but does not include the complete technical method, training duration or a detailed breakdown of hardware and expenses. The claim that it beat the best Stratego player in history is framed as an assessment of Niemeijer’s standing, not as a formal, universally agreed ranking. The result is a reported match achievement, not proof that the system will win consistently against every elite player.
educational strategy board games for kids
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
More Evidence on Ataraxos
The next useful evidence will be the team’s full research paper and any accompanying match records, including rules, game logs and evaluation procedures. Those materials can clarify how Ataraxos was trained, what the stated budget covers and how the researchers measured its performance against human opponents.
Further matches against other top Stratego players would show whether the reported result holds beyond this series. Until those details and tests are available, the confirmed development is the team’s reported 15 wins, one loss and four draws against Niemeijer—not a settled claim that Ataraxos is unbeatable or that the approach has solved imperfect-information AI as a whole.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Ataraxos?
Ataraxos is an AI system developed by researchers from Carnegie Mellon, MIT, NYU and Stanford to play Stratego, a board game in which players cannot see the ranks of their opponent’s pieces.
What was the reported result against Pim Niemeijer?
The source report says Ataraxos won 15 games, lost one and drew four against Niemeijer. It does not provide the full match conditions in the supplied material.
Why is Stratego difficult for AI?
Stratego combines hidden piece identities, bluffing and long games. Players must infer what an opponent’s pieces might be while making plans that can extend over many moves.
How much computing power and money did training take?
The team is reported to have trained Ataraxos using 16 GPUs and a few thousand dollars. A detailed cost breakdown and training duration were not included in the supplied source material.
Does this mean AI has solved imperfect-information games?
No. The reported win is a result in one game and one match series. More evaluation would be needed to establish how reliably Ataraxos performs against other elite players or whether its methods transfer to other settings.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
