DeepMind’s AlphaStar defeated professional players in the real-time strategy game StarCraft II in 2019.
DeepMind announced AlphaStar in January 2019 after it defeated two professional StarCraft II players, Grzegorz “MaNa” Komincz and Dario “TLO” Wünsch, in exhibition matches. The system played the Protoss faction and used deep neural networks trained with imitation learning and reinforcement learning.
AlphaStar’s achievement was notable because StarCraft II requires long-term planning, rapid decisions, resource management, scouting, and adaptation to an opponent whose strategy is only partly visible. Unlike a board game, it also presents a large action space and unfolds in real time.
In October 2019, DeepMind reported that AlphaStar had reached Grandmaster level on the European ladder, placing it among the top 0.2% of ranked players. AlphaStar is distinct from AlphaGo, which was developed for the board game Go, and AlphaZero, a more general game-playing system.