DeepMind has mastered StarCraft

Alphabet’s DeepMind has beaten a top pro at StarCraft II. Wired reports:

After months of training, the Alphabet-owned AI firm’s AlphaStar program is now capable of playing a full game of StarCraft II against a professional human player – and winning. It might sound frivolous, but mastering a game as complex as StarCraft is a major technological leap for DeepMind’s AI brains.

Now, my initial reaction to this was: so what? We’ve seen DeepMind create an AI that beat top players at games before, a journey that started with it learning how to play, and get unbeatably good at, Pong.

Thankfully, James Temperton at Wired explains why exactly this latest bit of gaming matters:

While conquering Pong required understanding a relatively small number of basic actions, mastering StarCraft was many orders more complex. Around 300 basic in-game actions branch out into millions of possible counter-actions, all of which the AI has to understand in the blink of an eye.

Now, 18 months after work started, DeepMind has mastered StarCraft. So how did it do it? The DeepMind team has claimed a number of algorithmic and engineering breakthroughs. From game theory to working with imperfect information and mastering long term planning, the AlphaStar system had to grapple with a complex and unpredictable world.

I strongly recommend reading the full piece.