The working principle of Alpha Go program is a multi-layer neural network based on deep learning. Neural network deals with the positioning and situation judgment of Go board, and makes decisions by heuristic two-dimensional filtering, classification and logical reasoning. The network improves its performance through repeated training and parameter adjustment, and gradually evolves to a better state. The specific thinking process of the network is random, and through a lot of training, you can show strong chess power in Go.