David Silver explains how self-improvement can lead to optimal possible behavior in single agent games, and how the system AlphaZero was able to beat AlphaGo Zero, AlphaGo, and even the world's strongest computer chess program, using its own principles.