Sections

Machine learning bots beat DOTA 2 world champions in best-of-three

After millions of learning matches, OpenAI Five readily best world champs

The DOTA 2 world champions were bested over the weekend in unprecedented fashion by a team of AI agents.

Created by artificial intelligence firm OpenAI, the machine learning bots ripped through reigning champs Team OG in a decisive 2-0 sweep that was live streamed on Twitch.

Dubbed the OpenAI Five, the bots were trained by playing millions of matches against themselves, and were subject to certain restrictions.

Not only were they were limited to realistic reaction times, but also a predetermined set of characters and items. Additionally, they were isolated and unable to communicate with each other.

"We started OpenAI Five in order to work on a problem that felt outside of the reach of existing deep reinforcement learning algorithms," said OpenAI in a blog post.

"We hoped that by working on a problem that was unsolvable by current methods, we'd need to make a big increase in the capability of our tools.

"We were expecting to need sophisticated algorithmic ideas, such as hierarchical reinforcement learning, but we were surprised by what we found: the fundamental improvement we needed for this problem was scale. Achieving and utilizing that scale wasn't easy and was the bulk of our research effort."

In a write-up of the match, Ars Technica noted that despite the success of the OpenAI Five, the bots lacked understanding of certain MOBA strategy nuances. This ultimately didn't prove a problem for the bots, however.

For a limited time, DOTA 2 players can now sign up to take on the bots themselves.

Related stories

FanAI lands $8m in Series A funding

Esports audience monetization platform has raised a total of $12.5m to date

By Brendan Sinclair

Matthew "Nadeshot" Haag to be honoured at Esports Bar Cannes

100 Thieves founder will be named Shaker of the Year at annual awards

By Matthew Handrahan

Latest comments

Sign in to contribute

Need an account? Register now.