Researchers from MIT, Carnegie Mellon University, New York University, and Stanford University created a new AI system for Stratego.
Stratego is a game with hidden information, where players do not see all opponent pieces. The game has over 10 to the 66th possible configurations, making it very hard for AI.
Previous AI models, like those from Google’s DeepMind, used costly operations and still could not beat top human players. They required millions of dollars to train.
The new system, called Ataraxos, uses a technique called self-play reinforcement learning. It plays against itself many times to learn strong strategies.
Ataraxos trains faster and more efficiently than past models. It needs fewer examples and less training time, lowering costs.
The AI achieved higher performance than DeepNash, a DeepMind system, with much less data and training.
This development can help in real-world problems like military planning, business negotiations, and cybersecurity.
Why it matters
This AI can handle complex situations with hidden information more efficiently, saving time and resources.
What to do
Consider testing this approach for your strategic decision tasks. Focus on models that use self-play reinforcement learning.
Source: https://news.mit.edu/2026/game-playing-ai-stratego-new-champ-0930