Abstract:

Early AI researchers focused on games like chess as a proving ground for machine intelligence. In this talk we’ll trace the evolution of chess robots from expert-encoded knowledge to AlphaZero in 2017, which learned chess from scratch in 9 hours of self-play and crushed the best traditional engines. We’ll walk through the key concepts of computer chess including minimax search, alpha-beta pruning, Monte Carlo tree search, and neural networks. We’ll wrap up with how these methods have since been applied to more complex games like poker, Go, and StarCraft.