OpenAI Five defeated professional Dota 2 players in a 2019 best-of-three exhibition match.
OpenAI Five was a team of neural networks trained to play the multiplayer online battle arena game Dota 2. Unlike a chess system, it had to coordinate five separate agents, act under incomplete information, and make rapid decisions in a changing environment.
In April 2019, OpenAI Five defeated a team of professional Dota 2 players in a best-of-three demonstration in San Francisco. The event followed earlier matches in which the system had defeated individual professionals and improved through large-scale self-play.
OpenAI Five is sometimes confused with AlphaStar, DeepMind’s system for StarCraft II. Both systems used reinforcement learning and played complex strategy games, but AlphaStar focused on StarCraft II, while OpenAI Five focused on Dota 2.