Editing Stratego (section)

== Stratego AI ==
In July 2022, [[DeepMind]] announced the development of DeepNash, a model-free [[multi-agent reinforcement learning]] system capable of playing ''Stratego'' at the level of a human expert.<ref>{{cite web |date=9 July 2022 |title=Deepmind AI Researchers Introduce 'DeepNash', An Autonomous Agent Trained With Model-Free Multiagent Reinforcement Learning That Learns To Play The Game Of Stratego At Expert Level |url=https://www.marktechpost.com/2022/07/09/deepmind-ai-researchers-introduce-deepnash-an-autonomous-agent-trained-with-model-free-multiagent-reinforcement-learning-that-learns-to-play-the-game-of-stratego-at-expert-level/ |website=MarkTechPost}}</ref> ''Stratego'' has been difficult to model well because the opponent's pieces are hidden, making it a game of [[perfect information|imperfect information]], the initial setup has more than 10<sup>66</sup> possible states, and the overall game tree has 10<sup>535</sup> possible states. DeepNash was able to win 84% of 50 ranked matches in online matches hosted by Gravon over a period of two weeks in April 2022 against human players, and won at a minimum rate of 97% over hundreds of matches against previously-developed ''Stratego''-playing programs including Probe, Master of the Flag, Demon of Ignorance, Asmodeus, Celsius, PeternLewis, and Vixen.<ref>{{cite journal |doi=10.1126/science.add4679 |author1=Perolat, Julien |author2=De Vylder, Bart |author3=Hennes, Daniel |author4=Trassov, Eugene |author5=Strub, Florian |author6=De Boer, Vincent |author7=Muller, Paul |author8=Connor, Jerome T. |title=Mastering the game of Stratego with model-free multiagent reinforcement learning |date=1 December 2022 |volume=378 |issue=6623 |pages=990–996 |journal=Science|pmid=36454847 |arxiv=2206.15378 |bibcode=2022Sci...378..990P |s2cid=250144392 }}<!--author-token to entire article: https://www.science.org/stoken/author-tokens/ST-887/full --></ref>