Google’s DeepMind Artificial Intelligence Could Next Take On StarCraft

Having beaten the world’s No. 1 Go player in the opening two games of a five-game series this week, the people behind Google’s artificial intelligence division DeepMind are already looking for its next challenge — and that could be the complex real-time strategy game StarCraft.

Speaking after DeepMind’s AlphaGo program defeated Lee Sedol for the first time on Wednesday, co-founder Demis Hassabis was asked what was next for his artificial intelligence algorithm and he indicated that playing the multiplayer game was a possibility.

“I think for perfect information games, Go is the pinnacle,” Hassabis told the Verge. “Certainly there are still other top Go players to play. There are other games — no-limit poker is very difficult, multiplayer has its challenges because it’s an imperfect information game. And then there are obviously all sorts of video games that humans play way better than computers, like StarCraft is another big game in Korea as well. ”

Hassabis’ comments were backed up by Google Senior Fellow Jeff Dean who was asked what’s next for DeepMind. “StarCraft, I think, is our likely next target,” he said at the Structure Data event in San Francisco on Thursday.

StarCraft would seem like a perfect fit for DeepMind and Hassabis, given he started out his career creating video games, including co-creating the celebrated Theme Park when he was just 17 with industry icon Peter Molyneux.

However, Hassabis said that he would only look at taking on StarCraft if it fit in with the development of AlphaGo. “We’re only interested in things to the extent that they are on the main track of our research program. So the aim of DeepMind is not just to beat games, fun and exciting though that is.”

This would not be the first time that StarCraft has been played by AI programs. At the annual AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment last year, 22 programs played against each other continuously for two weeks on 12 virtual machines, which amounts to over 1,800 games each. However, despite the number of AI programs taking part, human players still triumphed.

“It is obvious from the results of the human vs. machine match that StarCraft AI bots still have a long way to go before beating the best expert human players,” the organizers said.

While games, including Go, and video games like StarCraft are an ideal test bed for programs like AlphaGo, the real test for any AI platform will be transitioning to the real world, something IBM’s Watson has found to be a struggle to date. DeepMind’s first attempt to solve real-world problems will be in conjunction with the U.K.’s health service, the NHS.

Hassabis said what DeepMind will be able to do in the health area is a lot different to what IBM is doing with Watson in this area, and will include things such as “the medical diagnosis of images and then maybe longitudinal tracking of vital signs or quantified self over time.”