“Dota’s a great testbed for artificial intelligence; it’s a very complicated game with a large competitive scene,” OpenAI co-founder and CTO Greg Brockman said in a video explaining the project (below). “The rules of Dota are so complicated [that] if you just think really hard about how the game works and try to write those rules down, you’re not even going to be able to reach the performance of a reasonable player.”
OpenAI has been hard at work on a bot capable of beating top professional players at Dota 1v1. On Friday, it was showcased at The International, a huge event hosted by Valve that draws 20,000 fans and players competing for $24 million in prizes.
The OpenAI bot went up against Danylo “Dendi” Ishutin in rather dramatic fashion, and handily beat the pro player twice before Ishutin bowed out.
“Bot is really fun and challenging to play against 🙂 I am sure it is possible to beat it. But it [has] no room for even slight mistakes,” Ishutin tweeted after the match. “Need much more tries for that 🙂 I think first few levels are most important to keep the lead. Still was amazing fun to play on stage!”
Ishutin added that it’s “awesome what awaits us in future with all those technologies. Excited!!!”
According to Brockman, OpenAI “trained entirely through self play. It starts out completely random with no knowledge of the world and simply plays against a copy of themself, which means it always has an evenly matched opponent. And it climbs this ladder of skill level until it’s able to the reach the performance of the best professional players in the world.”
OpenAI then invited pro players to try out the OpenAI bot. “Many pros wanted to keep playing the bot and talked about using it as part of their training routine,” Brockman said, though “amateur players enjoyed playing the bot as well.”
The International was OpenAI “introducing our Dota product to the world and see if they can reach the world of the top human professionals,” Brockman said.
Going forward, OpenAI wants to “mix AIs and humans on a single team and reach a level of performance that neither of them can reach on their own,” according to Brockman. “AI can be extremely beneficial to humanity and it’s going to require fundamental advances to see what it’s really capable of.”