I think Deepmind definitively showed the agents are learning high-level play, which was great to see. I didn't really pay attention to AlphaGo, but did skim the

I think Deepmind definitively showed the agents are learning high-level play, which was great to see. I didn't really pay attention to AlphaGo, but did skim the AlphaZero paper, and I'm not really left with any doubts about how good RL/LSTMs can get against other AI or humans given enough time to train. That said, it's an open question whether given all the constraints of the last live match and what was mentioned during the talk (that newer strategies keep getting discovered by humans and agents), whether humans could even win 50% of the time against an agent.