First AI to beat a 9-dan professional Go player

Between 9–15 March 2016, a largely self-taught Go-playing AI called AlphaGo (developed by British AI developers DeepMind) defeated Lee Sedol (KOR) 4–1 in a five-game series. This is the first time a 9-dan professional (the highest rank in Go) has been defeated by a computer program of any kind.

Before AlphaGo, it was generally assumed that it would be a decade before computers could challenge skilled Go players, let alone beat 9 dan professionals. This is because Go is resistant to the sort of "brute force" analysis that had been employed by computers against games like tic-tac-toe and, most famously, chess.

A brute-force game-playing program works by calculating every possible move in a given situation, then every possible response to it, then every possible response to that, and so on – often looking many moves ahead of the current state of play. It then applies a few relatively simple rules (considerations about the value of pieces or the importance of different sectors of the board, for example) to decide which of these possibilities to pick.

This is not a practical approach for Go because there are so many possibilities in any given situation – looking more than a move or two ahead would require the computer to calculate and assess an almost infinite number of potential outcomes.

The developers at DeepMind used a totally different approach called Deep Reinforcement Learning to create their Go-playing AI. AlphaGo was shown information about tens of thousands of past Go matches, which it then analysed using neural networks. One neural network decided where on the board to play based on moves used in winning games and the other evaluated how likely it would be to win based on that move. This meant that AlphaGo developed strategies it hadn’t seen in human games, performing moves that went against centuries of established Go wisdom. AlphaGo’s capacity for intuitive and creative play stunned many commentators. Speaking after his own defeat in May 2016, Ke Jie (the world's top ranked Go player at the time) remarked, "After humanity spent thousands of years improving our tactics, computers tell us that humans are completely wrong... I would go as far as to say not a single human has touched the edge of the truth of Go."