On the company blog, OpenAI explains their AI agent learned Dota 2 from scratch by self-play. The bot learned to predict where other Dota 2 players will move, to improvise in unfamiliar situations, and how to influence the other player's allied units to help it succeed.
And while DeepMind intends to use imitation learning to improve their StarCraft II bots by analyzing hundreds of thousands of replays of human matches, OpenAI stresses that their AI system did not need imitation learning nor tree search to excel at Dota 2.
Today we played Dendi on mainstage at The International, winning a best-of-three match. Over the past week, our bot was undefeated against many top professionals including SumaiL (top 1v1 player in the world) and Arteezy (top overall player in the world).
Dota 1v1 is a complex game with hidden information. Agents must learn to plan, attack, trick, and deceive their opponents. The correlation between player skill and actions-per-minute is not strong, and in fact, our AI’s actions-per-minute are comparable to that of an average human player.
Elon Musk thanks Microsoft for the providing access to the Azure cloud computing platform, which delivered the necessary processing power for the AI. The next step is to create a team of five Dota 2 bots to take on the top human teams.