Back in 2016, artificial intelligence researchers at Google's Deepmind turned their attention to StarCraft 2, an effort that picked up steam the following year when Blizzard released tools that enabled them to build bots capable of taking on human opponents. It also opened a data cache of 65,000 past matches to be used in training the bots, and said that it would add another half-million games to the cache each month.
Blizzard issued a fairly low-key "work is continuing" update at the most recent BlizzCon, although there was a certain ominous edge to it: Deepmind had developed a habit of immediately worker-rushing its opponents, for instance, a behavior Blizzard said was "amusing," before acknowledging that it also had a 50 percent success rate against StarCraft 2 AI opponents at the "Insane" difficulty.
And it was still learning: "After feeding the agent replays from real players, it started to execute standard macro-focused strategies, as well as defend against aggressive tactics such as cannon rushes," Blizzard said at the time.
Even though that was in November 2018—less than three months ago—it's already time for another update. "It’s only been a few months since BlizzCon but DeepMind is ready to share more information on their research," Blizzard said today, reminding us all that AIs learn at a geometric rate. "The StarCraft games have emerged as a 'grand challenge' for the AI community as they're the perfect environment for benchmarking progress against problems such as planning, dealing with uncertainty and spatial reasoning."