According to a post on Google’s DeepMind website, “Testing our agents in games that are not specifically designed for AI research, and where humans play well, is crucial to benchmark agent performance. That is why we, along with our partner Blizzard Entertainment, are excited to announce the release of SC2LE, a set of tools that we hope will accelerate AI research in the real-time strategy game StarCraft II.”
The StarCraft franchise from Blizzard is one of eSports’ most played competitive games. It even spawned an entire competitive community and culture in countries like South Korea. It might not seem like a particularly complex game on the surface, but it does involve quite a lot of action from gamers where they will need to successfully manage their units on a micro level, all the while building up their base, gathering resources, and upgrading. This does seem like a perfect training ground for the DeepMind AI.
As to why they chose StarCraft 2, Google writes, “Even StarCraft’s action space presents a challenge with a choice of more than 300 basic actions that can be taken. Contrast this with Atari games, which only have about 10 (e.g. up, down, left, right etc). On top of this, actions in StarCraft are hierarchical, can be modified and augmented, with many of them requiring a point on the screen. Even assuming a small screen size of 84×84 there are roughly 100 million possible actions available.”