Google’s AlphaGo defeats Go Master

Have you ever imagined a future where robots can play your favorite game with or against you? Now, imagine that you were a world champion at this game. If this were the case, you might think that no computer would be able to defeat you, which is exactly what the artificial intelligence (AI) community thought of the game Go.

Go, a game that originated in ancient China, is a complex strategy-heavy board game for two players who take turns putting stones on a 19x19 board. The purpose of the game is to control more territory than one’s opponent.

Despite Go’s seemingly simple rules, the game itself is very complex since there are approximately 10,170 unique games possible (compared to 10,120 unique games for chess), which is why some experts thought that it would take another 10 years for a Go computer program to defeat a human Go champion.

In 2014, Google Deepmind in London developed a program called AlphaGo that plays a unique style of Go, which has been found to confuse many top players. Each move that the computer makes is designed to raise the probability of winning, even if it makes minor sacrifices along the way. The program had previously played against the European Go champion, Fan Hui, and won 5-0 in October of 2015.

Recently, a series of five Go matches was played from March 9-15 between Google’s AlphaGo program and Lee Se-dol, a top-tier Go champion. Many commentators said that there was a massive difference in skill level between Lee and Fan, the European champion. Before the series, many experts predicted that Lee would win against the computer. Lee himself was also confident of a victory.

However, to the surprise of these experts, AlphaGo won the first game of the series. While Lee led for most of the game, AlphaGo took advantage in the last 20 minutes after Lee made increasingly more mistakes as the game went on, showing that the machine’s consistency was able to best human creativity.

The second game was much more lopsided than the first, with AlphaGo in command from the beginning. Lee commented that AlphaGo played a “nearly perfect game,” and AlphaGo was certain of victory midway through the game even though many top-ranked players could not tell who was ahead at that point.

Although AlphaGo won the first two games, many still doubted whether it was truly a strong player. These doubts were erased in the third game, which AlphaGo reportedly won in a more convincing fashion.

Although the series was decided, games four and five were still played out. Lee finally broke through in game four, capitalizing on a rare mistake made by AlphaGo mid-game. Lee took a risky approach from the beginning in an attempt to counter AlphaGo’s calculated approach.

This extremely calculated approach is something humans are able to harness after a large deal of practice but on a far more limited level than the one AlphaGo operates on.

AlphaGo “played itself, different versions of itself, millions and millions of times and each time got incrementally slightly better — it learns from its mistakes,” Demis Hassabis, chief executive of DeepMind, told the BBC before the matches started.

This learning tendency was evident in the fifth game when Lee adopted an all-out approach similar to his strategy in game four. AlphaGo still made some mistakes but it was able to adapt and learn from its mistakes in the previous game. In the end, the Go-playing computer fended off Lee’s attacks in a very close contest. This ended the series, with AlphaGo winning four times and losing only once.

How did AlphaGo beat one of the best Go players in the world? Since there are so many possibilities in the game, the first step for the AI is to make the number of available moves smaller. AlphaGo starts by choosing from 30 million high-level moves, and it plays out many different scenarios using its AI. It then chooses the move that maximizes its chances of winning.

Humans cannot calculate the probability of winning the way a computer can. They often choose the best move for a given situation, going off their instincts and past experiences.

However, humans can have creativity and passion, which robots may never be able to replicate.

“I think the AI will ultimately defeat humans in Go, and it is an inevitable consequence of changing times,” Lee told reporters. “But robots will never understand the beauty of the game the same way that we humans do.”