When Blizzard announced that it would be working with Deepmind to teach an artificial intelligence to play Starcraft II during Blizzcon 2016, gamers scratched their heads.
In truth the announcement made sense to us because unlike Go, or the Atari games which Deepmind’s AI has already mastered, Starcraft is tough. This toughness comes for the multiple layers that exist in a Starcraft game. Sure the objective is destroy the enemy but there’s a lot more going on.
“StarCraft’s action space presents a challenge with a choice of more than 300 basic actions that can be taken. Contrast this with Atari games, which only have about 10 (e.g. up, down, left, right etc). On top of this, actions in StarCraft are hierarchical, can be modified and augmented, with many of them requiring a point on the screen. Even assuming a small screen size of 84×84 there are roughly 100 million possible actions available,” Deepmind wrote in a blog post.
This difficulty has created a roadblock for Deepmind so now its calling for help. More folks working on the project could lead to greater advancements.
The SC2LE has been released to the public and contains tools that will help AI researchers teach a machine to play Starcraft II and accelerate the research.
The tools include the Machine Learning API developed by Blizz for Starcraft II, a set of 65 000 game replays, PySC2 which allows researchers to use Blizzard’s API with their AI agents and a white paper that contains baseline results for mini-games as well as the results of the 1v1 mini-game against the game’s built-in AI.
Of course the eternal question is whether Deepmind’s AI prefers Zerg, Terran or Protoss? The answer to that question however, is actually rather interesting.
Incidentally, we spoke to Starcaft II’s Executive Producer Chris Sigaty at BlizzCon last year. We wanted to ask him if the AI preferred Zerg, Terran or Protoss – which at that stage he couldn’t answer – but he did reveal something rather interesting.
“One of the things that could come out of this project that would be incredibly beneficial to us and to other games as well is that we could have an inherent balance problem,” said Sigaty.
“We pay attention to this very closely – we want the game to be fair through all three races and we keep that balance over time. But maybe there is something that gets revealed that shows there’s an abusable aspect in the game and that would help us find an issue we’d need to fix,” he added.
So if your AI agent starts consistently picking one race over another you might want to contact Blizzard and let it know its game’s balance is out of whack. The more you know.
[Source – Deepmind]