Playing go cooperatively is very different from chess or most other turn-based tactical games in that combinatorics make the gamespace astronomically large. Most pro players understand what impact this has on gameplay and how to adapt their risk management strategies to the game. Way back in the day, popular man v cpu strategies took into account the limitations of monte carlo simulations and would intentionally make moves that change the game state as much as possible to exploit how hard the computer would need to work in order to begin finding optimal moves.
I think alpha go has advanced to the point where its unlikely human players can reliably defeat it, but I think there are still opportunities for better algorithmic players to defeat it.
Anyways, certain moves can cut off game states by the quadrillions, so it's often pretty intuitive what an optimal response to a move is, given certain context of the board and sometimes the player. In that respect, I'm curious how pro level go is going to change after these alpha go games are studied, because it has a very peculiar, decidedly calculatorish style of play, but it's obviously very effective regardless. Go has prospered for so long because there's so much room to express yourself in a move, pro players really play with their whole soul, but computers are just taking advantage of the pure mathematical angles of the game
It's fascinating stuff. I really want to see the alpha go team write a starcraft ai or something like that.