AI finally cracks Stratego, a game of hidden information

A new research paper describes how AI systems have finally achieved strong performance at Stratego, a board game long considered difficult for AI because players can't see their opponent's pieces. The breakthrough came from adding a second neural network specifically designed to guess the identity of hidden enemy pieces, working alongside the main decision-making model.

The work, published in Nature and posted on arXiv, builds on techniques used in other AI game-playing milestones like chess and Go, but adapts them for games with incomplete information. Stratego has been a benchmark challenge in AI research because strategies must account for uncertainty about what pieces opponents control.