South Korean Gets ‘Priceless’ Victory Over Computer in Go Match
By CHOE SANG-HUN

SEOUL, South Korea — After three straight losses, a South Korean expert rallied on Sunday for his first victory against a Google computer program playing Go, an ancient board game known as the most complex ever invented.

Lee Se-dol, 33, a boyish South Korean Go master, was all smiles after a brilliant move forced the Google program, AlphaGo, to surrender the match in the middle of the contest.

Hundreds of local Go enthusiasts and reporters who were gathered at the Four Seasons Hotel in downtown Seoul burst into applause over the human Go master’s dramatic comeback against the machine.

“You know, I have played many, many Go games, but I don’t think I have ever been as happy with one single victory as with this one,” Mr. Lee said. “This is priceless.”

Demis Hassabis, the chief executive of Google DeepMind, Google’s artificial intelligence company, said Mr. Lee’s victory was a reminder that AlphaGo still had room for improvement. A “creative genius” like Mr. Lee tests the limits of the machine, he said.

Go has been seen as the last great challenge in computer programmers’ efforts to create software that can outwit humans in board games. Go is such a complex game, with an almost infinite possible sequence of moves, that artificial-intelligence experts had predicted that computer programs needed more than 10 more years before they would be able to beat Go legends like Mr. Lee, who has 18 international titles.

Before the best-of-five series began on Wednesday, Mr. Lee had been upbeat. But the mood quickly sank after he lost the first three matches.

After losing his third match — and $1 million in prize money — on Saturday, Mr. Lee admitted that the psychological pressure he felt in facing a nonhuman foe was a big handicap.

But Mr. Lee also said AlphaGo was not perfect.

“It was Lee Se-dol, not humans, who lost the matches,” he said defiantly on Saturday, adding that he would do his best to beat the machine in the remaining two games.

On Sunday, Mr. Lee delivered.

Go is a two-person strategy board game said to have been created in China more than 3,000 years ago. The players compete for territory by placing black and white stones on intersections of a board of 19 horizontal and 19 vertical lines. The first player in a match uses black stones; the other, white. On Sunday, AlphaGo held the black stone.

During the post-match news conference on Sunday, Mr. Lee said he had found some weaknesses of AlphaGo.

When he made a surprise move, AlphaGo acted as if it had “a bug,” he said. He also said AlphaGo seemed to do better when it held the white stone.

Mr. Lee then offered to let AlphaGo play the white stone in the last match, scheduled for Tuesday.

“I beat it with the white stone,” Mr. Lee said. “Now, I want to see if I can beat it with black.”

Google’s DeepMind team accepted the offer.

笑開了！奇招打亂AlphaGo 南韓棋王逆轉勝

南韓圍棋王李世乭與谷歌（Google）人工智慧超級電腦系統AlphaGo十三日在首爾四季飯店進行第四輪對弈，雙方鬥智近五個小時，李世乭終於在三連敗後拿下首勝。

卅三歲的李世乭在賽後記者會上難掩興奮之情，但仍謙虛地說：「這是我第一次只贏一場比賽，就受到人們如此盛情恭賀。」他說：「如果我以三比○領先輸掉一局，可能很痛苦，但連輸三局後取得首勝，我高興極了。這是無價的勝利。」

研發AlphaGo的谷歌DeepMind執行長哈薩比斯恭喜李世乭得勝，他說：「李世乭證明了自己為何如此傳奇，因為他的精湛棋藝讓AlphaGo感受到壓力而犯錯。」李世乭贏過四十七項圍棋賽事，十八度榮獲世界棋王銜頭，他是目前職業圍棋世界排名第一維持時間最久的紀錄保持人。

在場觀戰的谷歌共同創辦人布林說：「我們對這場比賽印象深刻也很興奮，恭喜李世乭。我很期待觀賞最終盤比賽（十五日），儘管屆時我會在美國加州。」

法新社報導，李世乭此戰一開始原本居下風，尾盤才逆轉勝，是場「令人驚喜」的勝利。韓聯社報導，AlphaGo此局一開始的行棋套路與第二局相似，一度讓圍棋專家懷疑電腦有固定的下棋策略和手法。李世乭在第十二手改變招式後，AlphaGo也使出不同招數應對，一度下出讓人匪夷所思的棋路。

李世乭在第七十八子出奇招下出「挖」的棋形，AlphaGo似乎大為意外應對失當，下了幾步被李世 乭形容為「好像程式出現漏洞」的棋。人機對弈三個多小時後，李世乭終將獲勝的情勢逐漸明朗，南韓圍棋評論員難掩喜悅興奮之情。但AlphaGo仍試圖翻盤，在李世乭的奇招逆轉局勢後仍苦撐卅多步，才認輸敗陣。

哈薩比斯說：「AlphaGo自行計算的勝率，在第七十九手時約七成，但到了第八十七手時僅剩不到五成。」

李世乭說，這一局前半他原本以為又要輸了，第七十八子是在當時局面下「別無選擇的一手」，卻逆轉局勢受到讚揚，「真讓我摸不著頭腦」。

美聯社報導，AlphaGo落敗，顯示電腦也有弱點，並不完美且仍有進步空間。李世乭曾說找不到AlphaGo策略弱點是輸棋主因。

原文參照：
http://www.nytimes.com/2016/03/14/world/asia/south-korean-gets-priceless-victory-over-computer-in-go-match.html

2016-03-14．聯合報．A13．國際．編譯田思怡