OpenAI's AI that reached professional level in Dota 2 through massive-scale reinforcement learning.

Description

OpenAI Five is the artificial intelligence system that reached professional human level in Dota 2, one of the most competitively complex games. It features a team of 5 AI agents coordinating without explicit communication. Using Reinforcement Learning, it trained by playing the equivalent of 45,000 years of gameplay against itself every day. It handles 20,000+ observable situations and 1,000+ possible actions, demonstrating Long-Term Planning in matches exceeding 45 minutes. Teamwork emerged naturally without specific programming. It beat professional teams, including world champions. Public matches allowed anyone to play against the AI, demonstrating coordination, strategy, and adaptation. Limitations included a limited hero pool and item restrictions. Retired in 2019, it showcased the scalability of reinforcement learning and influenced game AI research. While there were controversies over whether its strategy was 'human-like', it remains a historic milestone in AI gaming. Public technical documentation is available.