With most information hidden, the game Stratego had stumped AI–until now (arstechnica.com)

A groundbreaking achievement in AI has emerged as a team of researchers from Carnegie Mellon, MIT, NYU, and Stanford developed Ataraxos, an AI that mastered the strategic board game Stratego, defeating the world's top player, Pim Niemeijer, in a staggering 15 to 1 match record. Unlike previous attempts by DeepMind, which struggled to outperform human players in this complex game, Ataraxos utilized just 16 GPUs and a modest budget to effectively navigate the vast array of strategic possibilities presented by Stratego's unique dynamic of hidden information. Stratego poses a significant challenge for AI due to its large number of variables, with players managing 40 pieces and a plethora of possible configurations—over a decillion setups. The game’s intricacies, including the lengthy gameplay and elements of bluffing, further complicate computational strategies. While other games like poker have specific limitations in their hidden information, Stratego's depth and the long duration of potential matches had previously stumped AI models. Ataraxos’s success not only showcases advancements in AI gaming strategies but also signifies a leap forward in developing algorithms capable of handling complex decision-making in imperfect-information scenarios, offering valuable insights for the AI/ML community.