Master of Go Board Game Is Walloped by Google Computer Program
By CHOE SANG-HUN and JOHN MARKOFF

SEOUL, South Korea — Computer, one. Human, zero.

A Google computer program stunned one of the world’s top players on Wednesday in a round of Go, which is believed to be the most complex board game ever created.

The match — between Google DeepMind’s AlphaGo and the South Korean Go master Lee Se-dol —was viewed as an important test of how far research into artificial intelligence has come in its quest to create machines smarter than humans.

“I am very surprised because I have never thought I would lose,” Mr. Lee said at a news conference in Seoul, South Korea. “I didn’t know that AlphaGo would play such a perfect Go.”

Mr. Lee acknowledged defeat after three and a half hours of play.

Demis Hassabis, the founder and chief executive of Google’s artificial intelligence team DeepMind, the creator of AlphaGo, called the program’s victory a “historic moment.”

The match, the first of five scheduled through Tuesday, took place at a Seoul hotel amid intense news media attention. Hundreds of reporters, many of them from China, Japan and South Korea, where Go has been played for centuries, were there to cover it. Tens of thousands of people watched the contest live on YouTube.

Go is a two-player game of strategy said to have originated in China 3,000 years ago. Players compete to win more territory by placing black and white “stones” on a grid measuring 19 squares by 19 squares.

The play is more complex than chess, with a far greater possible sequence of moves, and requires superlative instincts and evaluation skills. Because of that, many researchers believed that mastery of the game by a computer was still a decade away.

Before the match, Mr. Lee said he could win 5-0 or 4-1, predicting that computing power alone could not win a Go match. Victory takes “human intuition,” something AlphaGo has not yet mastered, he said.

But after reading more about the program he became less upbeat, saying that AlphaGo appeared able to imitate human intuition to a certain degree and predicting that artificial intelligence would eventually surpass humans in Go.

AlphaGo posed Mr. Lee a unique challenge. In a human-versus-human Go match, which typically lasts several hours, the players “feel” each other and evaluate styles and psychologies, he said.

“This time, it’s like playing the game alone,” Mr. Lee said on the eve of the match. “There are mistakes humans make because they are humans. If that happens to me, I can lose a match.”

To researchers who have been using games as platforms for testing artificial intelligence, Go has remained the great challenge since the I.B.M.-developed supercomputer Deep Blue beat the world chess champion Garry Kasparov in 1997.

“Really, the only game left after chess is Go,” Mr. Hassabis said on Wednesday.

AlphaGo made news when it routed the three-time European Go champion Fan Hui in October, 5-0.

But Mr. Lee, 33, is one of the world’s most accomplished professional Go players, with 18 international titles under his belt. He has called the European champion’s level in Go “near the top among amateurs.”

AlphaGo has become much stronger since its matches with Mr. Fan, its developers said. It challenged Mr. Lee because it was ready to take on someone “iconic,” “a legend of the game,” Mr. Hassabis said. Google offered Mr. Lee $1 million if he wins the best-of-five series.

Mr. Hassabis said AlphaGo does not try to consider all the possible moves in a match, as a traditional artificial intelligence machine like Deep Blue does. Rather, it narrows its options based on what it has learned from millions of matches played against itself and in 100,000 Go games available online.

Mr. Hassabis said that a central advantage of AlphaGo was that “it will never get tired, and it will not get intimidated either.”

Kim Sung-ryong, a South Korean Go master who provided commentary during Wednesday’s match, said AlphaGo made a clear mistake early on, but that unlike most human players, it did not lose its “cool.”

“It didn’t play Go as a human does,” he said. “It was a Go match with human emotional elements carved out.”

Mr. Lee said he knew he had lost the match after AlphaGo made a move so unexpected and unconventional that he thought “it was impossible to make such a move.”

Mr. Lee said he now thought his chances for victory in the five-match series were 50-50.

Some computer scientists said Wednesday that they had expected the outcome.

“I’m not surprised at all,” said Fei-Fei Li, a Stanford University computer scientist who is director of the Stanford Artificial Intelligence Laboratory. “How come we are not surprised that a car runs faster than the fastest human?”

On Tuesday, before the match began, Oren Etzioni, the director of the Allen Institute for Artificial Intelligence, a nonprofit research organization in Seattle, conducted a survey of the members of the Association for the Advancement of Artificial Intelligence.

Of 55 scientists, 69 percent believed that the program would win, and 31 percent believed that Mr. Lee would be victorious. Moreover, 60 percent believed that the achievement could be considered a milestone toward building human-level artificial intelligence software.

That question remains one of the most hotly debated within the field of artificial intelligence. Machines have had increasing success in the past half-decade at narrow humanlike capabilities, like understanding speech and vision.

However, the goal of “strong A.I.” — defined as a machine with an intellectual capability equal to that of a human — remains elusive.

Other artificial-intelligence scientists said that humans might still find refuge if the goal posts for the competition were moved.

“I wonder what would happen if they played on a 29 by 29 grid?” wondered Rodney Brooks, a pioneering artificial intelligence researcher who is the founder of a robot company, Rethink Robotics. By enlarging the playing space, humans might once again escape the machine’s computing power.

圍棋九段李世乭對弈AlphaGo 人腦首次輸電腦

「哈哈！沒想過會輸，太驚訝了。」代表人類與谷歌人工智慧系統AlphaGo進行世紀對決的南韓圍棋九段李世乭，9日輸掉首戰。這是人工智慧首度在圍棋領域擊敗人類頂尖高手，透過網路直播觀戰的10萬人，見證此一歷史時刻。

IBM的超級電腦「深藍」1997年擊敗世界西洋棋王卡斯波洛夫，轟動全球。各界預期人工智慧至少得花10年才能擊敗布局更複雜的圍棋高手，在李世乭吞敗後，南韓現代研究所研究員張祐碩表示：「這是象徵性的一勝，也是人工智慧的大躍進。」

最近10年職業圍棋世界排名第一時間最久的李世乭，與谷歌人工智慧系統AlphaGo自9日起在南韓首爾四季酒店進行5場對決。執黑子的李世乭先攻，前半局掌握主導權，但在後半局出現失誤，被AlphaGo找到破綻逆轉。李世乭最終在比賽進行3個半小時、第186手，剩下28分28秒時棄子，AlphaGo還剩5分30秒。

AlphaGo研發計畫Deep Mind負責人哈薩比斯用「人工智慧的珠穆朗瑪峰」形容圍棋的難度與高度，他對首戰表示滿意。幫AlphaGo下棋的棋手，是台灣師範大學資訊工程博士黃士傑。

李世乭誇讚AlphaGo的開局非常漂亮，甚至有幾手不可能出自人類棋士，讓他難以招架。他笑稱：「沒想到AlphaGo狀況如此完美，雖然結果讓我非常驚訝，但過程真的很有趣，我期待接下來的比賽。」

觀戰者可就沒這麼淡定。現場解說的金成龍九段表示：「我跟圍棋界的人都非常震驚，沒有人料想到居然就這麼敗了。」

中國大陸世界冠軍柯潔九段接受新華社記者電話採訪時，用「震驚得說不出話」來形容自己的感受。

柯潔認為，李世乭在右下的局部處理過於簡單：「其實這也不算大勺子（指嚴重漏算），就是一連串的小失誤，但是被對手抓住了，可見AlphaGo已經有了和李世乭匹敵的能力，抓住機會的逆轉能力非常強，這才是最可怕的地方。」

人機大戰第二場訂於台灣時間10日中午12時登場，後三場在12、13、15日舉行。

閱讀秘書/AlphaGo 去年完勝陸職業棋士

擊敗南韓世界棋王李世乭的谷歌人工智慧系統AlphaGo，由希臘語字母表中第一個字母Alpha和圍棋的日文「碁」的發音Go組成，Alpha也有考試成績優等的意思。大陸圍棋迷稱為「阿爾法圍棋」，暱稱「阿爾法狗」或「阿狗」。2010年研發出AlphaGo軟體的公司2014年被谷歌收購，去年10月以5比0完勝歐洲圍棋冠軍、前中國職業棋手樊麾，成為第一個擊敗職業圍棋手的電腦程式。

AlphaGo與李世乭共比5場，贏家可得100美元，李世乭出場費每場3萬美元，每贏一場另有2萬美元獎金。如果5場全勝，將獲125萬美元。如果AlphaGo獲勝，獎金將捐給聯合國兒童基金會和圍棋慈善團體。即使一方先取得3勝，也會下滿5場。比賽採用中國規則，執黑子一方貼7目半，各方用時為兩小時，3次60秒的讀秒。第一場由李世乭執黑子先下，下一場輪到AlphaGo先下。

原文參照：
http://www.nytimes.com/2016/03/10/world/asia/google-alphago-lee-se-dol.html

Video：Lee Se-dol, the world’s top player of the boardgame Go, lost the first of five matches to a computer program, AlphaGo, designed by Google DeepMind.
http://nyti.ms/1QM1c1N

2016-03-10．聯合報．A1．要聞．編譯張佑生