I think they're hinting that the techniques used in AlphaGO are a promising path towards general AI. AlphaGo has NNs trained on Go boards, but champion go players will also have parts of their brain dedicated to their Go intuition that don't transfer to investment banking or playing golf. However the techniques of reinforcement learning can be used to tackle other problems... so we hope.

Also this reminded me of a comment I read last week: "Don't anthropomorphize computers. They hate that"