The AI makes some awful decisions, such as building five observers. This calls into question its "understanding" of the game. It looks like a lot of its ability

The AI makes some awful decisions, such as building five observers. This calls into question its "understanding" of the game. It looks like a lot of its ability comes from micro, which it's unsurprising a computer can do better than a human. This is impressive, but with a lot of caveats. DeepMind's work on chess and go was impressive with no caveats whatsoever.