AlphaZero, successor of AlphaGo, was a model developed by DeepMind which served as a generalized algorithm which could master multiple games without prior knowledge of the game. Once again, it was trained with Reinforcement Learning (RL) and played millions of games of Chess, Go, and Shogi.
Besides the policy and value heads it inherited from AlphaGo, AlphaZero also uses Monte Carlo Tree Search (MCTS) (a form of Alpha-Beta Pruning) to look ahead at possible moves and evaluate the best possible outcomes.
AlphaZero was trained entirely with self-play, and did not require any human data to achieve itβs performance.