Dota 2
OpenAI's Dota 2 bot beats world champions using pure self-play—no tree search, no imitation learning. Here's why that matters for real-world AI.

Why it matters
OpenAI demonstrated a major AI capability milestone: an agent learning complex, multi-agent competitive strategy from scratch through self-play alone. This proves reinforcement learning can scale to human-level performance in adversarial, real-time environments—a foundational step toward goal-oriented AI systems that operate in messy, unstructured domains.
The key facts
5 to knowBot defeats world-class Dota 2 professionals in 1v1 tournament-standard matches
Trained via self-play reinforcement learning without imitation learning or tree search
Published August 2017 by OpenAI
Capability demonstrated: multi-agent competitive reasoning in complex game environments
Framed as stepping stone toward real-world goal accomplishment in human-involved scenarios
Go to the source
OpenAI Blogopenai.com
Publisher excerpt: We’ve created a bot which beats the world’s top professionals at 1v1 matches of Dota 2 under standard tournament rules. The bot learned the game from scratch by self-play, and does not use imitation learning or tree search. This is a step towards building AI systems which accomplish well-defined…