人工智能在閱讀測試中勝過人類

They have beaten us at chess, Go and co-written a Europop album. Now computers are taking a step into a very human territory: the reading comprehension test, scourge of schoolchildren everywhere.

它們已經(jīng)在國際象棋和圍棋上打敗我們，合寫了一張歐洲流行樂(Europop)專輯。現(xiàn)在，電腦又進(jìn)軍一個非常人性的領(lǐng)域：閱讀理解測試，世界各地學(xué)生的苦難源頭。

Alibaba on Monday said its artificial research outperformed mere mortals in a global reading comprehension test that seeks answers to such pressing questions as “what was Nikola Tesla’s ethnicity?” and “how big is the Amazon rainforest?”

阿里巴巴(Alibaba)周一宣布，其人工智能在一項全球閱讀理解測試中表現(xiàn)優(yōu)于人類。該項測試提出的艱深問題包括：“尼古拉•特斯拉(Nikola Tesla)是什么種族的人”，以及“亞馬遜雨林有多大?”

Luo Si, chief scientist of natural language processing at Alibaba’s research arm, the Institute of Data Science of Technologies, dubbed the machines’ victory “a milestone”. He said the technology has many uses, from customer service to museum tutorials to medical enquires — some of which are already being handled by chatbots globally.

司羅是阿里巴巴數(shù)據(jù)科學(xué)研究院(iDST)的自然語言處理首席科學(xué)家，他把機(jī)器取得的這個勝利稱為一個里程碑，稱該技術(shù)有很多用途，從客戶服務(wù)、博物館講解，到醫(yī)療查詢，其中一些在全球范圍已經(jīng)由聊天機(jī)器人處理。

In the test administered last week, companies subjected their artificial intelligence systems to queries from the Stanford Question Answering Dataset, which assessed reading comprehension. The computers’ answers were compared against average human responses and ranked.

在上周進(jìn)行的測試中，參賽公司讓各自的人工智能系統(tǒng)解答斯坦福問答數(shù)據(jù)集(Stanford Question Answering Dataset)的提問，該數(shù)據(jù)集評估閱讀理解能力。計算機(jī)的答案被與普通人的答復(fù)進(jìn)行比較，然后據(jù)此排名。

Tying in top place were Microsoft, the US software giant, and Alibaba, the Chinese tech group that began with ecommerce and is — like its peers at home and abroad — ploughing more funds into AI, much of it via its research arm.

美國軟件巨擘微軟(Microsoft)與阿里巴巴并列首位，后者是一家從電子商務(wù)起家的中國科技集團(tuán)，如今與國內(nèi)外同行一樣，加大對人工智能的研究投入，其中大部分是通過其研究機(jī)構(gòu)投入。

While Microsoft and Alibaba won by the slimmest of margins — at accuracy levels a few basis points above humans’ 82.3 per cent in providing exact matches to questions — the tie provided a fitting symbol of the AI arms race being waged by the US and China.

盡管微軟和阿里巴巴以微小優(yōu)勢勝出，在回答的精確匹配度上僅比人類的82.3%高出幾個基點，但它們并列首位為美國和中國正在進(jìn)行的人工智能軍備競賽提供了合適的象征。

China aims to parlay its vast troves of data, collated from its 1.4bn population including 730m who are online, and deep pockets to overtake the US in creating a $150bn industry that is seen as the next industrial revolution. AI is already being harnessed in everything from driverless cars to customising newsfeeds.

中國力求利用來自其14億人口(其中有7.3億上網(wǎng))的海量數(shù)據(jù)和雄厚財力，在創(chuàng)建被視為下一場工業(yè)革命的1500億美元產(chǎn)業(yè)方面超越美國。人工智能已經(jīng)廣泛應(yīng)用于各個領(lǐng)域，從無人駕駛汽車到定制新聞推送。

While China’s drive is being spearheaded by government, it is cultivating homegrown tech giants including Baidu, Alibaba and Tencent to leap ahead of the US.

盡管中國的努力是由政府帶領(lǐng)的，它也在培育本土科技巨頭，包括百度(Baidu)、阿里巴巴和騰訊(Tencent)，以求超越美國。

Baidu made an early bet on AI, and is leading its rivals in autonomous driving, some of which was on show at last week’s Consumer Electronics Show in Las Vegas. Tencent, which boasts just shy of 1bn monthly active users on its social media WeChat app, and Alibaba have adopted AI in operations such as customising news and ads, and are investing in new fields.

百度更早投入人工智能的研究，目前在自主駕駛方面領(lǐng)先于競爭對手，其中一些成果在上周于拉斯維加斯舉行的消費電子展(CES)上亮相。騰訊旗下的社交媒體應(yīng)用微信(WeChat)號稱擁有近10億月活躍用戶，該公司和阿里巴巴都已在定制新聞和廣告等業(yè)務(wù)中采用人工智能，并投資于新領(lǐng)域。

Similar technology, reliant on drilling down in texts such as Wikipedia to learn and parrot information, has been used for answering the presumably more prosaic questions that flood in from buyers (“where’s my package?”) on Alibaba’s Singles Day shopping festival.

閱讀技術(shù)依賴于精讀文本，比如維基百科(Wikipedia)的文章，以學(xué)習(xí)和模仿信息。類似的技術(shù)已被用于買家提出的常見問題(比如“我的包裹在哪里?”)。在阿里巴巴的光棍日購物節(jié)期間，這類問題會大量涌入。