If you look at DeepMind's numbers, AlphaGo is meant to win more than 50% of the time against human players (expert). Right now, my opinion is that its actual capability in wins is less than 50%.