That's exactly how it's trained, it competes against itself in a millions of games and updates it's neural network with the heuristics of the winning instance.