Forgot your password?
Close
typodupeerror
AI Games

AI Has Finally Learned To Play Stratego (arstechnica.com) 40

Posted by BeauHD from the tough-challenges dept.
Researchers from Carnegie Mellon, MIT, NYU, and Stanford have built an AI called Ataraxos that finally cracked Stratego, beating four-time world champion Pim Niemeijer 15 games to one with four draws. Its key advantage was a second neural network that estimates the identities of hidden enemy pieces, letting the system search plausible game states rather than brute-force a massive hidden-information space. Ars Technica reports: Just like [DeepMind's DeepNash, introduced in 2022], Ataraxos learned by playing against itself -- 163 million games in total. In these self-play sessions, moves that led to wins were reinforced and played more often in future matches, while moves that led to losses were played less, which was the same simple training idea. The difference was in how much Ataraxos adjusted after each game, because hidden information tends to send self-play learning algorithms around in circles. The team addressed this by making big, bold changes in strategy early in training and small, careful ones later.

The even bigger innovation was something DeepNash never had: thinking ahead before each move. AIs like AlphaGo refine their general strategy with a search just before acting. DeepMind couldn't make that work in Stratego because the search space was too large, leaving it an open question whether it was worth trying. "This is one of the things that we did figure out how to do," Farina said. The solution was a second neural network, a belief model, trained to guess the opponent's hidden pieces based on how they had been moving. This way, instead of iterating through every possible arrangement, Ataraxos samples plausible ones, plays out candidate moves in each, and picks based on how they turned out. And it shows in its playstyle.

[...] DeepNash was trained for two to three months on 1,024 of Google's specialized chips, a run the Ataraxos team estimates would cost $3 million to $4.5 million at 2025 prices. Ataraxos, in contrast, needed 16 GPUs for a week, plus an additional four GPUs for four days to train the belief model. Farina and lead author Samuel Sokota achieved this efficiency by writing a simulator that runs millions of moves per second on graphics cards. "At the scale that we are in academia, we don't really have access to an entire field of GPUs," Farina said. The algorithm also learned far faster -- it played about 34 times fewer games than DeepNash, and still ended up much stronger. The findings have been published in the journal Nature.
This discussion has been archived. No new comments can be posted.

AI Has Finally Learned To Play Stratego More

AI Has Finally Learned To Play Stratego

Comments Filter:

  • Fascinating... (Score:3)

    by ffkom ( 3519199 ) on Friday October 02, 2026 @07:16PM (#66360510)
    This is the first time I saw "Stratego" mentioned for 40 years or so... never knew this game received attention among those aged above 12 or so. I can vaguely remember playing it a few times in childhood... are there specific countries where this is popular among adults?

    • Same here. And I probably expended a lot less energy learning how to play it than AI did. But I was certainly not world champion level. I might be able to beat my cousin once in a while...

    • Re: (Score:3)

      by Pollux ( 102520 )

      I own a Stratego board and play it regularly with my son. We also play it with close family from time to time. It's one of my favorite games. It allows for the right balance of strategy, skill, and adaptation, with just the right amount of luck mixed in. Additionally, even as one may be on the verge of losing, there still exist opportunities to win the game. And because arrangement of pieces can be different every game, and because your pieces are largely unknown to the other player until an attack is

      • What would you say is a good age for kids to start playing?

        • Re: (Score:2)

          by _merlin ( 160982 )

          Back in my day, it was popular with boys about 9 and up. But it was a bit of a niche game - it had its die-hard fans, and a whole lot of people who never played it.

    • That's the point, it's getting hard to find specific thinking tasks / games at which AI doesn't dominate people. Now there is one fewer known example.

    • Germany comes to mind.
      But probably not on a much higher level than Go.

      I did play Stratego since ages.

  • I would LOVE to know how AI would fare with the classic board game "Diplomacy"!
    For the ones that don't know it - it's a bit like "Risk", but there is no luck anywhere involved. Between each turn, all the players engage in e.g. 15 minutes of "Diplomacy" and literally anything is allowed (except moving the pieces). You can negotiate alliances, bribe and blackmail people, whatever. Many friendships have been lost to this game, because sooner or later you will have to backstab allies and betray friends.
    It's g
    • Turns out, making AI's play Diplomacy agains each other has been done quite often already! But I would like humans to play against the popular AIs.
  • The headline makes it sounds like the AI has just managed to play the game. But that's not what TFS and TFA are talking about. This is about the system beating the best human player at the game. In the meantime, in a very different game direction, there's a new Starcraft AI which learned the game playing against itself (similar to how AlphaZero self-trained) and is apparently beating almost all the humans even as it does some really weird stuff, some of which looks decidedly suboptimal https://relog.gg/razn [relog.gg]
    • It's not hard to make a computer that is unbeatable in Starcraft, because the computer can click perfectly every time. According to your link:

      the unauthorized version of Pluto was additionally modified to use a maphack...During some games, Pluto reached several thousand actions per minute... the bot used its micro to surround and neutralize Mutalisks... the bot simultaneously fought in multiple areas of the map, precisely micro-managed its units

      So yeah, that kind of clicking (and maybe perfect information, or maybe not) gives the computer a massive advantage, it doesn't even need a neural network to win.

    • learned the game playing against itself

      So did Joshua/WOPR and that was over 40 years ago. Stephen Falken was a fucking visionary and a genius!

  • They tried teaching it Risk (Score:3)

    by rsilvergun ( 571051 ) on Friday October 02, 2026 @09:26PM (#66360600)
    But after 10 million compute hours it's still setting up the board.

  • does anyone not think computers will always beat humans at any game so long as that game is a matter of computation - as long as you can write a good program for the computer. Generative ai is just that last bit.

    What people seem to have forgotten is that few games were made to be simply a contest of who can compute best, mostly they're about spending time with friends and making memories. Did the people who played stratego with the AI have as much fun as they did when they were 9 years old on a snow day lay

  • Perhaps after 165 million games played there is little substitute for the real thing. But given how AI (and humans) often learn a game to the level of building strategy, is there a reason these AIs weren’t merely trained by first reading every Stratego strategy book?

    I’ve not heard of a single Grandmaster that has ever reached that level by simply playing the game of chess. They read about it. A lot.

  • So now it can learn to kill us all more 'strategically'. :-)

  • The solution was a second neural network, a belief model, trained to guess the opponent's hidden pieces based on how they had been moving.

    Half learning, half having the logic augmented. That's a remodel.

  • If I had an AI company, I would have my AI post positive things here, wouldn't you?

Slashdot Top Deals

Always look over your shoulder because everyone is watching and plotting against you.

Close