‘I’m in shock!’ How an AI beat the world’s best human at Go Source: Mark Zastrow
History in the making
Humanity’s machines just outstripped us, again. A few hours ago in Seoul, South Korea, an artificial intelligence built by Google defeated top player Lee Sedol in one of our most complex games, Go.
“This is history, you saw it folks,” said Chris Garlock of the American Go E-Journal, one of the match’s commentators. The match was the first of a five-game series in Seoul.
As the game reached its conclusion, the reality of Lee’s defeat set in slowly across the venue, prompting quiet gasps of shock. This was stark contrast to Tuesday, when some Korean journalists had openly cheered Lee at a press conference. It was hard not to feel sympathy with Lee as I watched this opening defeat. He carries the hopes of a nation – not to mention a species – on his shoulders.
The news, that artificial intelligence has defeated humanity’s best Go player, has sent shock waves through the international Go community. “I felt emotional and dizzy, and stepped outside for a minute,” said Ben Lockhart, one of the top US amateur players, watching on in the press room.
AlphaGo made headlines in January when DeepMind – an artificial intelligence company that Google bought in 2014 – announced that its AI had defeated reigning European champion Fan Hui 5–0. Most observers had thought such an AI was a decade away. That prompted DeepMind to challenge Lee, considered the game’s dominant force of the last decade.
Putting on a show
The match against Fan was played in secret. Not this one. Google’s AlphaGo is facing off against Lee Sedol in the swanky Four Seasons hotel in the heart of downtown Seoul. The match has captured public imagination in Korea, with dozens of cameras and hundreds of reporters descending on the match site. The press filled two separate conference rooms – one with English commentary and one with Korean.
The match itself took place in a room that organisers wouldn’t disclose. There, isolated from outside noise, Lee sat across the board from Aja Huang, one of AlphaGo’s lead programmers, who referred to a monitor plugged into AlphaGo. A sign gave their names and flags – a Union Jack for Google DeepMind’s London headquarters.
Watch the best human player take on the mighty AlphaGo AI at Go
Go is riding a recent wave of pop culture relevance here in Korea, thanks to a couple of recent TV dramas featuring Go players. The game was televised live across Korea, China and Japan, where the game is most popular, and livestreamed on YouTube. On Naver, the top Korean search engine, combinations of “Lee Sedol” or “AlphaGo” were the top three search terms on the day of the match.
Human nature
Lee Sedol is a national hero in his native South Korea, known for his unconventional and creative play, as well as his brashness. He flashed his swagger at a press conference last month, predicting he would win in a “landslide”. “Of course, there would have been many updates in the last four or five months, but that isn’t enough time to challenge me,” he said.
But after watching DeepMind explain the algorithm at a press conference on Tuesday, Lee admitted to being “quite nervous” and backed off from his 5–0 prediction.
Lee had been in fine form, with a runner-up finish in the world championship and victory in the Korean national championship in January, where it is known as baduk. Most professional players thought that based on AlphaGo’s play in October, Lee would win handily. But no one knew how much progress it had made, or how it would respond to tougher competition.
Play-by-play
Lee showed his opening gambit in the game’s seventh move, with an unusual play in the middle right of the board, allowing him to stake out territory along the right side of the board. But AlphaGo continued to push, playing unconventionally for a human. “It’s unheard of! It’s crazy,” said Jackson.
AlphaGo fell further behind in the midgame with a bizarre stretch of moves that had observers shaking their heads. But as it fought back over the rest of the match, Lee grew agitated, dabbing at his chin with his fist and tapping his finger on the side of the board. “I’ve never seen him so nervous before,” said Lockhart.
The match ended in what Garlock called a “slugfest”, with AlphaGo’s performance impressing. “It’s all that I hoped for or expected,” said Michael Redmond, one of other commentators and the only Western Go pro to have reached 9-dan, the game’s highest level.
As the endgame played out, Lee appeared downbeat, his hands shaking at every move. He flashed a wan smile as he resigned.
The question is whether Lee can recover. In AlphaGo’s previous match, experts say Hui Fan’s moves grew erratic after dropping the first game.
“It’s so shocking. I expected AlphaGo to win one game, but I didn’t expect it to be the first one,” said Myung-wan Kim, a Korean 9-dan professional living in Los Angeles and commenting with Jackson.
“I am in shock, I admit that,” said Lee. But he said he had no regrets about accepting the challenge and planned to come back strong in the rest of the matches. “I didn’t think AlphaGo would play the game in such a perfect manner.”
| }
|