In its quest to advance artificial intelligence research, Google’s DeepMind has created systems capable of mastering a range of Atari computer games, and besting humans at the ancient Chinese board game Go.
Now, the company is seeking to overcome an even bigger challenge — creating an AI system that can play, and perhaps master, “StarCraft II.”
The company announced Thursday that it was teaming up with Blizzard Entertainment — the maker of the real-time strategy game — to open up "StarCraft II" to AI and machine learning researchers around the world. This means that anyone with the inclination and skills to do so can test and train their AI systems using StarCraft’s complex gaming environment.
“DeepMind is on a scientific mission to push the boundaries of AI, developing programs that can learn to solve any complex problem without needing to be told how. Games are the perfect environment in which to do this, allowing us to develop and test smarter, more flexible AI algorithms quickly and efficiently, and also providing instant feedback on how we’re doing through scores,” the company said in a statement.
StarCraft players control one of the three warring races in the game — the humans, the insect-like Zerg, or the technologically advanced Protoss. The players’ goal is to survive the hazards of the perilous universe they are stranded in and expand their bases and territory.
“Players’ actions are governed by the in-game economy; minerals and gas must be gathered in order to produce new buildings and units. The opposing player builds up their base at the same time, but each player can only see parts of the map within range of their own units,” DeepMind said. “Thus, players must send units to scout unseen areas in order to gain information about their opponent, and then remember that information over a long period of time.”
It is this feature that makes StarCraft, in some ways, a more difficult game to master than Chess or Go — where the players can at least see the entire field of play. According to DeepMind, mastering StarCraft would require an AI agent to demonstrate effective use of memory, an ability to plan over a long time, and the capacity to change plans based on new information — skills that resemble those needed in real-world tasks.
As a result, if researchers do manage to create an AI that can learn to play StarCraft, it would have ramifications far beyond the creation of a flexible video game-playing system.
“StarCraft is an interesting testing environment for current AI research because it provides a useful bridge to the messiness of the real-world,” DeepMind said in the statement. “The skills required for an agent to progress through the environment and play StarCraft well could ultimately transfer to real-world tasks.”