AlphaGo's first part of training is supervised learning i.e. it's learning from experts. By playing against alphago, experts are learning new strategies. This c

AlphaGo's first part of training is supervised learning i.e. it's learning from experts. By playing against alphago, experts are learning new strategies. This can then further boost alphago's performance. Since it's a 3000 year old game, some thought that the game didn't hold a lot of unexplored moves even though the known popular move sets are complex. Maybe this is the type of symbiosis Musk is always talking about.