Researchers' Ataraxos AI Beats Top Stratego Player 15-1

Training Ataraxos was comparatively cheap. Ars Technica reported that the system took just 16 GPUs and a few thousand dollars to train. That contrasts with previous efforts: even DeepMind, with its exceptional budget, could not build a machine that reliably beat the best human Stratego players.

Stratego's difficulty comes from what players cannot see. Each player gets 40 pieces representing military ranks, from a marshal down to a spy, plus bombs and a flag. A player wins by capturing the opponent's flag. The opponent knows where the pieces are but not what they are. Identities are revealed only when two pieces collide in battle: the weaker piece is removed, and the identity of the winner is revealed.

That makes Stratego an imperfect-information game, like poker, which computers cracked years ago. The article notes that Deep Blue took down Garry Kasparov at chess in 1997, AlphaGo beat Lee Sedol at Go in 2016, and poker bots have been beating professionals for years.

Eugene Vinitsky, a researcher at NYU and co-author of the study, said in the article: "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."

Editor's Summary A research team from Carnegie Mellon, MIT, NYU and Stanford built Ataraxos, an AI that beat Pim Niemeijer 15-1 with four draws in Stratego. The system was trained with 16 GPUs and a few thousand dollars, according to Ars Technica. The result extends machine victories to another imperfect-information game that had resisted reliable machine play.