It seems that DeepMind is now moving on to a bigger challenge, going from an ancient board game into a modern strategy video game.
StarCraft II As An AI
At the ongoing BlizzCon 2016, Blizzard and Google announced a collaboration to open up StarCraft II as an AI and machine learning environment for researchers around the world.
StarCraft II is one of the most fiercely competitive video games that is played professionally, and according to the DeepMind team, it will provide an interesting testing environment for research in AI as it "provides a useful bridge to the messiness of the real-world."
"The skills required for an agent to progress through the environment and play StarCraft well could ultimately transfer to real-world tasks," said Oriol Vinyals, a research scientist for DeepMind who was once the top-ranked player of the game in Spain.
While the complexity of Go is the result of a combination of simple rules, there exists several layers of such complexity in StarCraft II. The game requires players to make a succession of strategic decisions while they farm resources, build structures, create units and engage their opponents in combat.
Also, unlike Go where all the information is available to players as they progress in the game, the map of StarCraft II is initially obscured, giving them less information to base their decisions on. As such, playing the strategy game requires both memory and adaptation, along with long-term planning.
What To Expect From DeepMind And Blizzard Collaboration
According to the DeepMind team, the environment that it will be working on with Blizzard for AI research will be made available to all researchers in the field by next year. The environment will simplify the visuals of StarCraft II to be better received by machine learning systems, with an API that will allow for similar functions to bots in the past which study game data. Blizzard will also be providing replays of StarCraft II matches to give the AI systems data to analyze.
Will DeepMind be able to create an AI that can dominate matches against humans in StarCraft II, similar to how AlphaGo performed exceeding well against a world champion? That is the ultimate goal of the project, but the DeepMind team said that such an achievement is still very much in the future.
However, it should be noted that AlphaGo's ability to win against a human opponent was developed earlier than expected.