AI model outperforms top human players in Stratego

Researchers from several American universities, including MIT, Carnegie Mellon, NYU, and Stanford, have developed an artificial intelligence model that outperforms top human players in the strategy game Stratego. Published in the journal Nature, the study highlights the model's ability to excel in environments with imperfect information, where opponent identities remain hidden.

Unlike previous AI projects, this new system is noted for being highly efficient and cost-effective during its training phase. The researchers achieved these results by combining advanced algorithms with an innovative decision-time planning technique based on probabilities. The model also demonstrated strong generalization capabilities by succeeding in other strategy games with different rules and designs.

According to the researchers, the technology holds potential for real-world applications involving complex, imperfect information, such as commercial negotiations, cybersecurity, and military maneuvers.