DeepMind AlphaGo beats Go world champion Lee Sedol 3-0 in historic AI victory

Lee Sedol loses — The world's top Go player Lee Sedol lays the first stone against Google's artificial intelligence program AlphaGo during the third match of Google DeepMind Challenge Match in Seoul, South Korea REUTERS/Google/Yonhap

Google DeepMind's software AlphaGo has defeated Go world champion Lee Sedol 3-0 in a best of five match in a landmark victory for artificial intelligence research (AI). This week Lee has been playing the revolutionary computer programme in Seoul, South Korea, for $1m (£706,000) in prize money.

The Game of Go

The game of Go originated in China more than 2,500 years ago.
Two players take turns putting black or white stones on a 19-by-19 grid.
The aim is to place black or white stones on the board trying to capture the opponent's stones by surrounding them or surround empty space to make points from territory.
There are more possible positions in Go than there are atoms in the universe, making Go a googol times more complex than chess.
In 1950, Claude Shannon calculated there were one quadragintillion possible games of chess – 10<sup>123 or one followed by 40 zeros. Now mathematicians estimate the number of possible unique Go games exceeds 2.082 × 10^170 – two followed by 170 zeroes.
It can be very difficult to determine who is winning in competitions between skilled Go players, and many of the top human players rely on instinct over chess players' logic.

Google has decided to donate the winnings to charity after the match. The AI's victory follows a number of computer programmes that have beaten humans in games such as chess, jeopardy, chequers and scrabble.

The Chinese board game, which has been played in the east for at least 2,500 years, is considered to be a much more complex challenge for a computer than chess, because of the number of possible moves per turn. In chess, there are around 20 moves, whilst in Go, it can be up to 200. The increase in the number of possible moves means the total number of potential games of Go far exceed the number of games of chess.

AlphaGo, which was developed by British computer company DeepMind before it was bought by Google in 2014, studied thousands of previous Go matches and played itself millions of times before being unleashed on the world champion. Each time the machine learned more and more about the patterns of play and now the AlphaGo has usurped Lee for the top spot.

Before the match up Lee had been confident he would secure a win for humankind suggesting he would triumph either 5-0 or at worst 4-1 in the series. He said a machine could not match human intuition. But the former champ will now have to get used to second place as he prepares for the fourth and fifth matches in the series.

"AlphaGo played consistently from beginning to the end while Lee, as he is only human, showed some mental vulnerability," one of Lee's former coaches, Kwon Kap-Yong, told the AFP.

The genius machine had already beaten the European champion 5-0 and in the first game of the series, AlphaGo won by a slender margin after Lee was in the lead. In his second loss, Lee conceded that the programme had played a "nearly perfect game".

The programme was trained on 30 million expert moves before it began learning by playing itself. It is a far cry from Arthur Samuel's draughts-playing program developed in 1959. In another landmark victory for AI IBM's Deep Blue supercomputer beat world champion Garry Kasparov in 1997.