A research team from Carnegie Mellon, MIT, New York University, and Stanford University has developed an artificial intelligence system that defeated Pim Niemeijer, considered the best Stratego player of all time. The AI, named Ataraxos, won 15 games to one, with four draws, marking the first time a machine has reliably beaten elite human players in the classic board game. Previous attempts by major entities, including DeepMind, had failed to achieve this milestone despite significant resources.
Stratego presents a unique challenge for machine learning because it is an imperfect-information game. Each player commands 40 pieces representing military ranks, from marshal to spy, along with bombs and a flag. The objective is to capture the opponent's flag. While players know the positions of their opponent's pieces, they do not know their identities. Identities are only revealed when two pieces collide in battle; the weaker piece is removed, and the winner's rank is disclosed. This dynamic creates a massive amount of hidden information that unfolds over a long time scale.
Eugene Vinitsky, a researcher at NYU and co-author of the study, noted that this specific characteristic distinguishes Stratego from other games like poker, which computers have already mastered. The success of Ataraxos was achieved with a relatively modest compute setup. The team trained the model using just 16 GPUs and a budget of a few thousand dollars. This result follows a long history of AI victories in other domains, including Deep Blue defeating Garry Kasparov at chess in 1997 and AlphaGo beating Lee Sedol at Go in 2016.