Nov 4th, 2016

Google’s Deepmind AI is an advanced artificial intelligence that the gaming world should pay attention to with its latest announcement. Earlier this year, the AI managed to beat the world’s best Go player in what is probably the most astounding victory for an AI in the short history that it’s been technologically possible.

Now, Google is partnering with Blizzard to turn Deepmind to Starcraft II. In today’s current landscape, many of the world’s best human players can easily beat current AI opponents, which means Deepmind should be up for quite a challenge. While Go itself is a mathematically calculable game, Starcraft II has lots of hidden information thanks to the game’s “fog of war”, which keeps maps hidden from players unless they have a unit within line of sight.

The challenge here is putting an AI against a human in a game that is ever-changing. The map of a Starcraft II game doesn’t present the computer with a perfect presentation of all possibilities at that moment because the map changes so much if it’s not monitored with quick efficiency. But what happens if Google’s Deepmind can figure this out?

Tackling imperfect games with perfect AI

While the ability to adapt to ever-changing information is in the human players favor, one thing that has always dominated the competitive scene of Starcraft II is the number of actions per minute (APM) that a player can perform. That simply means the number of times a person can click each minute. This is where Deepmind has the advantage, because it can naturally outclick a human player.

While there have been attempts at creating an AI that’s better than humans at playing the game, those attempts were programmed with a set of actions that need to be performed at the start of any Starcraft II game, so technically this AI was an extension of its creator. Google’s Deepmind AI goal is machine learning, which means the machine teaches itself to play the game, learning as it goes.

Of course, Google isn’t without help when it comes to this venture. Blizzard is working on a special AI API that can be used to help teach bots fundamental aspects, like the layers of the game screen versus the mini-map. Blizzard said their goal in helping with the project is hopefully finding ways to improve their own games.

“Is there a world where an AI can be more sophisticated, and maybe even tailored to the player? Can we do coaching for an individual, based on how we teach the AI? There’s a lot of speculation on our side about what this will mean, but we’re sure it will help improve the game.”

In the future game developers could use AI input from projects like Deepmind to more thoroughly beta test their games, without requiring human testers that can find bugs and work out problems. This is especially helpful with the release model that Blizzard uses for its current games, where they’re added onto with expansion packs that brings entirely new content and new ways to play.