Gaming is a rich arena for artificial intelligence (A.I.) research. What better way to test a platform’s ability to adapt and improvise than to put it in a competitive situation, especially against a human opponent? Back in March, the A.I. platform AlphaGo conquered a human champion at Go, a notoriously complicated game. A few months later, Google announced that DeepMind, its A.I. subsidiary, had learned to beat the side-scrolling PC adventure “Montezuma’s Revenge.” Now comes the next step in the machines’ rise to (gaming) domination: “StarCraft II,” the immensely popular real-time strategy game, is becoming a sandbox for A.I. researchers and scientists. The initiative is the result of a partnership between DeepMind and Blizzard, which created the game. “StarCraft is an interesting testing environment for current AI research because it provides a useful bridge to the messiness of the real-world,” read a new note on DeepMind’s blog. “The skills required for an agent to progress through the environment and play StarCraft well could ultimately transfer to real-world tasks.” For those unfamiliar with the ‘StarCraft’ franchise, the games allow you to play as three intergalactic races: human, Zerg, or Protoss. Once you choose your race, you must build bases on alien planets, while mining minerals and fending off attacks. Because so much of every gameplay map is hidden from the player’s view, you must make decisions without full information, and respond quickly to unexpected situations. It’s difficult for a human game-player to handle, much less a machine. “StarCraft’s high-dimensional action space is quite different from those previously investigated in reinforcement learning research,” the blog continued. “To execute something as simple as ‘expand your base to some location,’ one must coordinate mouse clicks, camera, and available resources.” In order to beat the game, you must master hierarchical actions and planning, a difficult task for A.I. platforms, which have grown slowly better at Deep Reinforcement Learning. If you’re a researcher interested in machine learning, or simply a tech professional who wants to play around with the technology, be sure to check out the API developed by the StarCraft team once it’s released. DeepMind will work with Blizzard to develop scenarios so researchers can “get an agent up and running, and benchmark different algorithms and advances.” The API will arrive in early 2017. As DeepMind stated, the knowledge from playing a digital strategy game will help A.I. better strategize how to solve real-world problems. 'StarCraft' is the Latest A.I. Testbed
Gaming is a rich arena for artificial intelligence (A.I.) research. What better way to test a platform’s ability to adapt and improvise than to put it in a competitive situation, especially against a human opponent? Back in March, the A.I. platform AlphaGo conquered a human champion at Go, a notoriously complicated game. A few months later, Google announced that DeepMind, its A.I. subsidiary, had learned to beat the side-scrolling PC adventure “Montezuma’s Revenge.” Now comes the next step in the machines’ rise to (gaming) domination: “StarCraft II,” the immensely popular real-time strategy game, is becoming a sandbox for A.I. researchers and scientists. The initiative is the result of a partnership between DeepMind and Blizzard, which created the game. “StarCraft is an interesting testing environment for current AI research because it provides a useful bridge to the messiness of the real-world,” read a new note on DeepMind’s blog. “The skills required for an agent to progress through the environment and play StarCraft well could ultimately transfer to real-world tasks.” For those unfamiliar with the ‘StarCraft’ franchise, the games allow you to play as three intergalactic races: human, Zerg, or Protoss. Once you choose your race, you must build bases on alien planets, while mining minerals and fending off attacks. Because so much of every gameplay map is hidden from the player’s view, you must make decisions without full information, and respond quickly to unexpected situations. It’s difficult for a human game-player to handle, much less a machine. “StarCraft’s high-dimensional action space is quite different from those previously investigated in reinforcement learning research,” the blog continued. “To execute something as simple as ‘expand your base to some location,’ one must coordinate mouse clicks, camera, and available resources.” In order to beat the game, you must master hierarchical actions and planning, a difficult task for A.I. platforms, which have grown slowly better at Deep Reinforcement Learning. If you’re a researcher interested in machine learning, or simply a tech professional who wants to play around with the technology, be sure to check out the API developed by the StarCraft team once it’s released. DeepMind will work with Blizzard to develop scenarios so researchers can “get an agent up and running, and benchmark different algorithms and advances.” The API will arrive in early 2017. As DeepMind stated, the knowledge from playing a digital strategy game will help A.I. better strategize how to solve real-world problems. 
