FT社評 AlphaGo的一場發人深省的棋局

The development of computer programs that can beat humans at games has a long history — from the mastery of noughts and crosses in the 1950s to Deep Blue’s celebrated defeat of world chess champion Garry Kasparov in 1997.

能夠在遊戲中擊敗人類高手的計算機程序有着悠久的發展歷史——從上世紀50年代掌握“井字棋”制勝之道，到1997年“深藍”(Deep Blue；IBM研發的計算機——譯者注）擊敗國際象棋世界冠軍加里•卡斯帕羅夫(Garry Kasparov)。

In recent years, however, the pace of advance has quickened. Data-crunching devices routinely notch up previously unthinkable victories. Computers can triumph in quiz games, as IBM’s Watson proved when it won the TV show Jeopardy in 2011. They also mimic human aptitudes with ever greater facility. For instance, machines play arcade games simply by observing the movement of objects on the screen.

然而，近年來進步速度加快了。能夠運算海量數據的設備經常取得以往不可想象的勝利。計算機能夠在智力競賽中取勝，IBM的“沃森”(Watson)在2011年贏得電視節目《危險邊緣》(Jeopardy)就是例證。它們還能以越來越強大的“悟性”模仿人的天賦。例如，機器通過觀察屏幕上物體的運動，就能學會玩街機遊戲。

Even so, the triumph of the AlphaGo computer over the South Korean world champion Lee Se-dol in the first of a five-match series in the ancient Chinese board game of Go marks more than just a new notch on the computerised honours board. Mr Lee had been confident of victory and proclaimed himself “shocked” by his defeat.

即便如此，AlphaGo電腦在古老的中國棋盤遊戲——圍棋的對壘中擊敗韓國九段棋手李世石(Lee Sedol)，在五局“人機對戰”中首戰告捷，不僅標誌着電腦榮譽板上的一個新檔次。賽前對勝利信心滿滿的李世石，在落敗後坦承“震驚”。

Go is a little like a version of chess, only vastly more complicated. Indeed the possible moves within a game exceed the number of atoms within the universe. This is a challenge that would defeat traditional programmes. Indeed it can only be mastered by computers assembled into neural networks that teach themselves through observation and practice — abilities that remain at the frontiers of computer science.

圍棋有點像國際象棋的變體，只是複雜程度高得多。的確，其棋局的變數比宇宙中的原子數量還要多。這個挑戰會挫敗傳統的程序。事實上，只有多臺計算機組成神經網絡，通過觀察和實踐來“自學”（這些能力仍處於計算機科學的前沿），才能駕馭這種高難度挑戰。

Demis Hassabis and his team at DeepMind, the UK-based artificial intelligence (AI) arm of Alphabet, deserve credit for the speed at which they have mastered this undertaking. True, AlphaGo, a formidable piece of IT, could be described as a computerised sledgehammer aimed at a recreational nut. Its victory, however, is a reminder of how fast the world is overcoming the obstacles in the way of AI, and its deployment in the world about us.

傑米斯•哈薩比斯(Demis Hassabis)以及他在DeepMind（Alphabet旗下英國人工智能部門）的團隊以如此快的速度掌握圍棋制勝之道，這一點值得讚賞。沒錯，作爲一件具有強大能力的信息技術設備，AlphaGo可以被形容爲一把計算機化的大錘，其用途是敲開一個消遣的堅果。然而，它的勝利提醒世人，世界正在快速攻克人工智能及其實際部署所面臨的障礙。

That is largely due to the huge amount of cash being poured into AI research by US and Chinese companies. These are poaching some of the brightest computer scientists from universities, giving them the capacity and tools to pursue their heart’s desire.

這在很大程度上歸功於美國和中國企業對人工智能研究的巨大投入。這些企業從高校挖走一些最優秀的計算機科學家，並提供資源和工具，讓這些科學家從事內心渴望的研究。

According to a recent survey, half of the world’s AI experts believe human level machine intelligence will be achieved by 2040. This opens up huge possibilities for the enrichment of mankind, from tackling climate change and treating disease to labour-saving devices. It also raises ethical questions every bit as profound as those posed by genetics. AI experts talk about the possibility of the human brain being reverse-engineered. Physicist Stephen Hawking last year warned that unless we take care, board games might be the least of it: AI could ultimately “outsmart us all”.

根據最近的一項調查，全球半數人工智能專家相信，人類水平的機器智能到2040年就能成爲現實。這爲增進人類福祉開啓巨大可能性——從應對氣候變化、治療疾病，到節省勞動力的設備。這也引發種種道德問題，其深刻性絲毫不亞於遺傳學所構成的道德問題。人工智能專家談到人腦被“逆向工程”的可能性。物理學家史蒂芬•霍金(Stephen Hawking)去年曾警告，除非我們小心，否則棋盤遊戲可能是最無關緊要的問題：人工智能最終可能“比我們所有人更聰明”。

One does not have to believe in some future tech dystopia to believe that governments and wider society should take the implications of these developments seriously. Google, Facebook and other companies rushing into AI point out that they are establishing ethics panels to consider appropriate uses for these technologies. These are unlikely to be immune from commercial interests or indeed from the gung-ho enthusiasm of the researchers.

人們不一定非要相信未來將出現某種科技“敵託邦”纔會認爲，政府和整個社會應該認真對待這些發展的潛在影響。競相進軍人工智能領域的谷歌(Google)、Facebook等公司指出，他們正在成立倫理小組以考量這些技術的適當用途。這些小組不太可能對商業利益以及研究人員的熱忱無動於衷。

Some external scrutiny akin to that supplied in the case of genetics by the UK’s Human Fertilisation and Embryology Authority is needed to protect the public from developments that may threaten more than the amour-propre of a South Korean Go champion. Granted, there may yet be no evidence that computers will ever shrug off their human masters but we should still treat these developments with the humility and caution they deserve.

需要進行一些外部監督，類似於遺傳學領域的英國人類受精和胚胎學管理局(HFEA)，以保護公衆免受相關發展的威脅，這些威脅所牽涉的不只是韓國圍棋高手的自尊。當然，目前也許還沒有證據表明計算機有朝一日將踢開他們的人類主人，但我們仍應該對這些發展給予應有的謙卑和審慎。