Facebook develops AI for playing Go

The Japanese game of Go has many thousands of years. Despite the fact that the rules of the game are quite simple, it is not yet possible to create a program that could beat an experienced human player. Yes, there are programs, and they are very good, but professionals are not able to win such a program yet. Now, Facebook has decided to take matters into its own hands and create a self-learning system with AI elements that can defeat any person.

Guo’s rules are very simple, yes, but it’s difficult for the car to play this game, since there are a huge number of options, and so far no computer is able to calculate everything for sane time. But the man, with his seemingly limited ability to perform complex calculations for a certain time, turned out to be unexpectedly good at playing Go. According to company representatives, most good players use visual patterns to identify good or bad moves, and intuition plays a huge role here.

The developers of AI for playing Go are going to create a system that combines the elements of a regular computer program that calculates the moves in a huge number of options, and a system that recognizes visual patterns on the board, determining the optimal move option.

This task itself is part of a larger project whose purpose is to study the issuedeep learning. The term “depth” in this case refers to the depth of the graph of model calculations — the maximum length between the input and output nodes of a particular architecture. In the case of, for example, a simple direct distribution neural network, the depth corresponds to the number of network layers. The term 'deep learning' focuses on the complexity of learning the inner (deep) layers of a multilayer network that are difficult to classical teaching methods, such as the method of back propagation of error.

In order to defeat a person in Go, a computer system will be taught to “think” like a person.

In general, this is done not for the pleasure or satisfaction of the pure curiosity of researchers (although this too), but for the creation of computer systems that are able to work on approximately the same principle that the human brain works. In the future, such technologies will be used by the social network for their own purposes. For example, now the company is creating a version of a social network for people with any vision problems. So, to the question "what is a person holding in his hand", the system will be able to answer quite extensively and clearly for the questioner.

Facebook’s virtual assistant, M, will also be based on this technology, analyzing and studying user requests, in order to improve their work and respond in the way a person would do.