AI Bots Seek to Beat Top Human Players at Dota 2 | Tech News
OpenAI, a research group backed by Elon Musk, has developed a team of five AI bots that are teaching themselves to beat top-ranked human players of the hit multiplayer strategy game.
To practice, the bots have a training regime only possible for machines: they play 180 years worth of Dota 2 games against themselves every day.
That training appears to be working. On Monday, OpenAI said the bots have so far managed to beat teams of amateur human players in 5v5 matches. Now they’re slated to take on professional Dota 2 gamers in a competition this August.
“This year, we want to beat the best pro teams at the full game,” Christy Dennison, a machine-learning engineer at OpenAI, wrote in a blog post.
To do so, the research group developed not one, but five bots, using a machine-learning technique called “reinforcement learning.” This involves the bots playing the game over and over against themselves, with the goal of optimizing their strategy and teamwork. All that training is done over 128,000 CPU cores, which can let the AI play the game at an accelerated pace.
So far, the bots are only programmed to play Dota 2 under limited restrictions that remove certain aspects to the game. Nevertheless, the training has taught the bots to intuitively use strategies that can take a human eight years to develop, said William Lee, a commenter and ex-professional Dota 2 player.
“The teamwork aspect of the bot was just overwhelming. It feels like five selfless players that know a good general strategy,” Lee said in the blog post.
Time will tell if the training is enough to beat the top-ranked players in the Dota 2 competition this August. But before then, the AI bots will go up against a team of top-ranked human players on July 28, which will be streamed online.
Pitting an AI against human players at Dota 2 may sound trivial. But if bots can master a complex game, then presumably, they can also be used to solve other difficult problems, OpenAI said. Experts have noted that AI might excel at addressing major problems such as climate change or finding cures for diseases.
Musk, meanwhile, left the OpenAI board in February to avoid conflicts of interest with Tesla, but he remains an adviser and donor.