DeepMind AI goes 10-1 against pro StarCraft 2 players TLO and MaNa

Today, the AI developers at DeepMind showcased AlphaStar, the company’s attempt at bringing machine learning to StarCraft II. The results are not good for us human players of the best strategy games. AlphaStar took down two separate StarCraft pros - TLO and MaNa - with a whopping 10-1 record. The AI’s technique is occasionally wildly off meta, and sometimes just what you’d expect from standard pro play, but either way it’s incredibly effective. The exhibition broadcast shows the results of two separate series, all Protoss versus Protoss, held over the course of a few weeks. The first five games pitted AlphaStar against TLO, and in those games we see some very non-standard plays from the AI - repeatedly pushing up ramps, failing to wall off its base early on, building way more probes than usual, and favouring unusual army configurations like tonnes of Disruptors. While none of that’s standard play for Protoss mirrors, AlphaStar still went 5-0 in that first set. TLO is a Zerg main, so the next round was more of a like versus like matchup against Protoss player MaNa. The AI got an extra week of training in the meantime, and it played a bit different - but still won 5-0 in a more challenging match.

from PCGamesN http://bit.ly/2ReaZo0