DeepMind’s learning algorithms are moving from the ancient Chinese strategy board game to protecting Earth from alien races. In partnership with Blizzard, programmable A.I. agents are being implemented into the Starcraft II API.
While a game like Go — despite its monolithic complexity — has only about ten simple actions that can be taken during the course of the game, Starcraft II starts with 300. Mastering Go is a question of understanding its more than a googol’s worth of variations. Starcraft II, while on the surface a more simple game, has so many variances in information from moment to moment that it is a far greater challenge for any A.I. that hasn’t been programmed specifically to play it within certain parameters.
Adding to the complexity of the task is the “fog of war” that obscures an opponent’s activity. This means that not only does the A.I. need to learn how to play the game itself, it needs to recognize general strategy, map variation, and predictive tactical reasoning. It cannot be overstated how much of a challenge it would be for anyone outside of Blizzard to program an A.I. specifically for this game, let alone an A.I. capable of teaching itself to play.
Any hopeful researcher can customize a Starcraft II Deepmind A.I. agent for themselves, given the basic framework provided. The artificial intelligence can be educated using some of the thousands of in-game replays already made available by Blizzard, with hundreds of thousands more on the way.
Follow Nate Church @Get2Church on Twitter for the latest news in gaming and technology, and snarky opinions on both.