That's not what Monte Carlo Tree search is. The new version is still one neural network + MCTS. There's no way to store enough information to judge the efficiency of every possible move in a neural network, therefore a second algorithm to simulate outcomes is necessary.
