In Brief
  • DeepMind and Blizzard are trying to create an AI agent that can play StarCraft II effectively, navigating the challenges of the game and engaging opponents in real-time.
  • This complex testing environment can mimic the "messiness of the real-world" and could therefore lead to innovations in deep reinforcement learning.

AI in Gaming

Artificial intelligence (AI) in games is often confused with programmed bots (or NPCs). Yes, these bots are “intelligent,” so to speak, because they interact with human players, but those interactions are limited by the bots’ programming — they don’t behave outside their coding. Conversely, AI applications in games can react to the behavior of human players with their own assessment of the situation — like in that historic game of Go — using an algorithm called deep learning.

At BlizzCon 2016 on Friday, Google and Blizzard Entertainment announced a partnership to bring Google’s deep-learning AI, DeepMind, to StarCraft II. DeepMind will use the real-time strategy game as a testing environment for AI research, using deep reinforcement learning to develop an AI agent that can play StarCraft II effectively.

In a typical game of StarCraft, players begin by choosing a race, each with its own unique abilities and preferred gameplay techniques. Players go through dictated in-game economy actions — gathering resources, building and expanding a base, constructing units, and engaging opponents in real-time —all in a partially revealed map.

Leveling Up

According DeepMind research scientist Oriol Vinyals, “StarCraft is an interesting testing environment for current AI research because 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.”

The skills he’s referring to are very different from those required of the typical programmed bot. This system will need to “demonstrate effective use of memory, an ability to plan over a long time, and the capacity to adapt plans based on new information,” says Vinyals. It will need to do so at the same speed as a human in terms of actions per minute as well.

While it’s still a long way before an AI system will be ready to challenge human StarCraft champions, DeepMind is confident that this collaboration will serve as a platform for further AI research, and they have plans to make the environment they construct with Blizzard open and available to other researchers in 2017.