Nonprofit research company OpenAI wants to pit its team of five neural networks, called OpenAI Five, against a team of top professional “Dota 2” players at The International esports tournament in August.
OpenAI Five has already defeated amateur human teams using a number of restrictions, the company said. The bots only played mirror matches with limited heroes and there was no warding, Roshan, invisibility, or summons.
“The hero set restriction makes the game very different from how ‘Dota’ is played at world-elite level,” OpenAI said. “However, the difference from regular ‘public’ games is smaller.”
The bots are reportedly playing 180 years worth of matches against themselves every day to learn the game. They train using a system that runs on 256 GPUs and 128,000 CPU cores. Currently, they’re bad at last-hitting, OpenAI said. It’s a technique “Dota 2” players use to grind gold during a match. But, the bots’ objective prioritization matches a common professional strategy.
“Gaining long-term rewards such as strategic map control often requires sacrificing short-term rewards such as gold gained from farming, since grouping up to attack towers takes time,” OpenAI said. “This observation reinforces our belief that the system is truly optimizing over a long horizon.”
The International 2018 takes place Aug. 20-25 at the Rogers Arena in Vancouver, Canada.
“Our team is focused on making our August goal. We don’t know if it will be achievable, but we believe that with hard work (and some luck) we have a real shot,” the nonprofit said.