We've already reported on the DeepMind AI team, whose demonstration included pitting their AI called AlphaStar against seasoned StarCraft 2 pros and as it turns out, their digital opponent not only won, it did so by 5:0 in two matches.
The matches in question were held on 19 December 2018, when AlphaStar took on Team Liquid's Grzegorz MaNa Komincz, following a test match against Dario TLO Wunsch.
DeepMind's team are quite delighted since this is apparently the first time an AI beat a professional StarCraft 2. Apparently, StarCraft 2 has proven to be a more complex game than the likes of Quake 3 Arena or Dota 2, which were just some of the games DeepMind tackled so far.
No AI came even close to challenging actual pros and not even unfair advantages helped, as DeepMind threw in specifically tailored rules, superhuman abilities for the AI, and so on. "In contrast, AlphaStar plays the full game of StarCraft II, using a deep neural network that is trained directly from raw game data by supervised learning and reinforcement learning", they said.
Finally cracking StarCraft 2's multi-layered nature was no easy task for DeepMind, as this is obviously not their first take. Their progress was shown off on last year's BlizzCon but knowing all that went down with Diablo Immortal, nobody can be blamed for missing it.
Apparently, AlphaStar's baby steps quickly turned into cheese, as it quickly advanced to attempt the worker rush. While this isn't the sort of strategy you want to be trying against pro players, DeepMind's AI was quite successful against the CPU at Insane difficulty, hitting a 50 per cent win rate.
The main reason for AlphaStar's success, however, can be attributed to its learning algorithms and supervised studying of human games, where it learned to analyze and emulate winning strategies. Moreover, DeepMind's team set up a sort of a league, where AIs that learned from different players fought amongst themselves, perfecting different strategies like any players in a league would.