A university research team just built an AI that beats the best Stratego player alive - something DeepMind, despite its budget, never managed.
Researchers from Carnegie Mellon, MIT, NYU, and Stanford built an AI called Ataraxos that took on Pim Niemeijer, widely considered the greatest Stratego player in history, and won 15 games to 1, with 4 draws. Stratego hides each side's 40 pieces from the opponent; identities are revealed only when two pieces collide in battle, making it a far messier information problem than chess or Go. Training Ataraxos took just 16 GPUs and a few thousand dollars. DeepMind, which has vastly larger budgets, had already tried and failed to build a Stratego bot that could reliably beat top humans.
Stratego's hidden information unfolds over many more moves than poker's hands or Go's fully visible board, which is why it resisted AI for so long. NYU researcher Eugene Vinitsky, a co-author of the study, said: "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." That Ataraxos cracked it on a budget suggests the holdup was never raw compute - it was finding the right approach to that scale of hidden information.
Chess fell in 1997, Go in 2016, poker years ago - Stratego just lost its spot as the board game AI couldn't handle, and it didn't take a DeepMind-sized budget to do it.