More on Dota 2
OpenAI Blog
A machine learning system trained through self-play improved from matching high-ranked Dota 2 players to defeating top professionals in approximately one month. The system achieved superhuman performance through continuous self-improvement, where the training data quality automatically increased as the agent's skill level rose. Self-play training eliminates the performance ceiling imposed by supervised learning approaches, which are limited to the quality of their static training datasets.
Why it matters
Our Dota 2 result shows that self-play can catapult the performance of machine learning systems from far below human level to superhuman, given sufficient compute. In the span of a month, our system went from barely matching a high-ranked player to beating the top pros and has continued to improve since then. Supervised deep learning systems can only be as good as their training datasets, but in self-play systems, the available data improves automatically as the agent gets better.
