TL;DR
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
Ataraxos, an AI developed by researchers from Carnegie Mellon, MIT, NYU and Stanford, beat Stratego player Pim Niemeijer 15 games to one, with four draws. The team says it trained the system using 16 GPUs and a few thousand dollars, addressing a game shaped by extensive hidden information, bluffing and long matches.
Researchers from Carnegie Mellon, MIT, New York University and Stanford say their AI, Ataraxos, beat elite Stratego player Pim Niemeijer by 15 games to one, with four draws. The result addresses a long-running challenge for game-playing AI: Stratego combines concealed piece identities with bluffing and matches that can stretch to thousands of moves, while the team says the system was trained on 16 GPUs for a few thousand dollars.
Stratego gives each player 40 pieces of different ranks, including bombs and a flag. Players can see where an opponent’s pieces are but not what each one is; identities are disclosed when pieces meet in battle. The objective is to capture the opposing flag, so players must make decisions while inferring what the other side has deployed and how it may respond.
The reported match against Niemeijer ended with 15 Ataraxos wins, one loss and four draws. The researchers’ account describes Niemeijer as arguably the best Stratego player of all time; that characterization is an assessment, not a formal ranking stated in the available material. The team also says training used 16 GPUs and cost a few thousand dollars, a comparatively modest resource claim that has not been independently detailed in the source material.
The researchers say Stratego’s hidden-information problem is much larger than the uncertainty in many familiar card games. MIT computer scientist and co-author Gabriele Farina said Stratego has 40 pieces whose arrangement can vary, producing more than a decillion possible setups. The game can also run to about 2,000 moves, according to Farina, requiring a system to manage uncertainty over a long stretch rather than solve a short sequence of decisions.
A Harder Test for Game-Playing AI
The reported win matters because it shows a system succeeding in a game where the challenge is not simply calculating legal moves or searching for a known board position. Ataraxos had to act without knowing the opposing pieces’ ranks, interpret moves that might be bluffs, and revise its judgments as battles revealed information. Those demands make Stratego a test of decision-making under uncertainty over time.
The resource claim also draws attention: the researchers say they reached the result with 16 GPUs and a few thousand dollars, rather than a large-scale computing budget. That suggests a potentially more accessible route to strong performance in this particular game. It does not, by itself, show that the methods will transfer to other settings, or establish a general advance in AI beyond the reported match.
The result is also a benchmark against an accomplished human opponent, not proof that every Stratego player would lose to the system or that it wins in every game. The published score provides a concrete measure of performance in the reported contest, while the size and conditions of that evaluation remain relevant to interpreting it.
As an affiliate, we earn on qualifying purchases.
Why Stratego Resisted Earlier Systems
Computers have achieved high-profile results in games with very different information structures: Deep Blue defeated chess champion Garry Kasparov in 1997, and AlphaGo beat Lee Sedol at Go in 2016. Poker systems have also surpassed professional players in some settings. Stratego presented another kind of obstacle because the board is visible but the strength and identity of most pieces are concealed.
NYU researcher and co-author Eugene Vinitsky described the game as distinctive for its large amount of hidden information unfolding over a long time scale. Farina contrasted it with Texas Hold’em, where a player’s two private cards create a much smaller set of possibilities. Stratego also rewards bluffing: a weak piece can be moved as though it were a powerful one, but repeated or absent bluffing can make a player predictable.
The report says DeepMind’s DeepNash, introduced in 2022, had not reliably beaten the best human players. The Ataraxos result therefore marks a reported step beyond that earlier effort, although the available source does not provide a direct head-to-head comparison between the two systems.
“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.
Limits of the Reported Match
The source reports the match score and training resources but does not provide the full evaluation protocol, including how the games were selected, whether the players used particular rules or time limits, or how the draws were handled. It also does not give enough detail to independently assess the claimed training cost or reproduce the result.
It remains unclear how Ataraxos performs against a wider group of expert players, whether its strength holds across different match conditions, and how consistently it can beat top human opponents. The result supports a claim about this reported contest; it does not settle those broader questions.
strategy game pieces and accessories
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Testing Ataraxos Beyond Niemeijer
The next useful evidence would be evaluations against additional high-level Stratego players and a fuller account of the system’s training and match conditions. Those details would help establish whether the 15-1 score reflects durable strength across opponents or a result specific to this contest.
The available report does not specify a date for further matches, a release of Ataraxos, or planned follow-up research. Until such information is provided, the outcome should be treated as a significant reported result in Stratego rather than a final measure of the AI’s performance across the game.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Ataraxos?
Ataraxos is an AI developed by researchers from Carnegie Mellon, MIT, NYU and Stanford to play Stratego.
What was the reported match score?
The researchers report that Ataraxos beat Pim Niemeijer 15 games to one, with four draws.
Why is Stratego difficult for AI?
Players cannot see the identities of an opponent’s pieces, and must make decisions amid bluffing and long games. The researchers say a match can reach about 2,000 moves.
How much computing did the team say it used?
The team said Ataraxos was trained using 16 GPUs and a few thousand dollars. Further details about the cost and training setup are not provided in the source material.
Does the result prove Ataraxos beats every top player?
No. The reported score is from a match against Pim Niemeijer. Performance against a broader field of elite players has not been established by the information available.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
