日本研發(fā)人工智能軟件 考試成績(jī)超學(xué)生

The latest test score results showed that Torobo-kun has a probability of at least 80 percent of passing the exams of 80 percent of 581 private universities across Japan.

Japanese researchers have revealed artificial intelligence software so smart it can beat most real students on a high school test.

Known as To-Robo, the AI software scored higher on the English section of Japan's standardized college entrance test than the average Japanese high school senior, its developers said.

It has also managed to double its score in just 12 months - raising hopes it will eventually pass the entrance exam for Tokyo University, Japan's most prestigious college.

'The average score for the English section of the standardized entrance exam was 93.1 (out of 200), but the AI scored 95,' a spokesman for NTT Science and Core Technology Laboratory Group said.

Last year the software scored 52.

The NTT lab is developing the software alongside the National Institute of Informatics, and is in charge of developing the software's English capabilities.

The project began in 2011 with a 10-year time frame for reaching its goal.

Questions from the test were turned into data that could be recognized by the software.

To-Robo then processed the information, distinguishing the logic of exchanges and correctly identifying the right answer out of multiple choices.

'One of our purposes is to reveal the limitations of AI,' said Noriko Arai, an NII professor who heads the project, told the Asahi Shimbun.

'Showing how humans and machines will be able to supplement each other is a key for Japan's economic growth in the future.'

The latest test score results showed that Torobo-kun has a probability of at least 80 percent of passing the exams of 80 percent of 581 private universities across Japan.

One reason is the great improvement in its scores for English and Japanese, Yoyogi Seminar officials said.

However, they admit the software has has more work to do.

The AI program cannot grasp unrealistic situations, such as an object with no mass or an environment with no friction, in the physics test.

It also struggled with concepts it had never heard of - for instance, it did not know what democracy is like because it has yet to learn the idea of majority rule, social justice concepts and other related information.

In the English test, it was unable to answer questions using figures and illustrations.

It was better at recognizing photographs, the researchers said.

日本研究人員透露，人工智能軟件已經(jīng)聰明到可以在高中考試中打敗大部分真正的學(xué)生了。

To-Robo的開(kāi)發(fā)人員稱，該智能軟件在日本的標(biāo)準(zhǔn)化大學(xué)入學(xué)測(cè)試的英語(yǔ)部分，獲得了比日本高中生平均成績(jī)還要高的分?jǐn)?shù)。

僅在過(guò)去的12個(gè)月內(nèi)，To-Robo就成功將成績(jī)翻倍，提升了它最終通過(guò)日本最負(fù)盛名的大學(xué)——東京大學(xué)的入學(xué)考試的希望。

一位NTT科學(xué)核心技術(shù)實(shí)驗(yàn)室的發(fā)言人稱，“標(biāo)準(zhǔn)化入學(xué)測(cè)試英語(yǔ)部分的平均分為93.1(以200人為例)，但智慧型軟件得分95。”

去年該軟件的得分為52。

NTT實(shí)驗(yàn)室正與國(guó)立情報(bào)學(xué)研究所一道開(kāi)發(fā)此軟件，并負(fù)責(zé)開(kāi)發(fā)該軟件的英語(yǔ)能力。

此開(kāi)發(fā)項(xiàng)目始于2011年，計(jì)劃在10年內(nèi)完成目標(biāo)。

測(cè)試中的問(wèn)題被轉(zhuǎn)換為可被軟件識(shí)別的數(shù)據(jù)。

然后，To-Robo處理信息，以區(qū)分其中的邏輯并正確地從多項(xiàng)選擇中識(shí)別正確答案。

“我們的目標(biāo)之一是揭示出智能軟件的局限性。”領(lǐng)導(dǎo)此項(xiàng)目的國(guó)立情報(bào)學(xué)研究所教授新井紀(jì)子( Noriko Arai)向《朝日新聞》透露。

“展示人類(lèi)和機(jī)器將怎樣互為補(bǔ)充，是未來(lái)日本經(jīng)濟(jì)發(fā)展的一個(gè)關(guān)鍵。”

最新的測(cè)試得分結(jié)果顯示，Torobo-kun 至少有80%的可能性，來(lái)通過(guò)全日本581所私立大學(xué)中80%的考試。

代代木研討會(huì)官員表示，原因之一是該智能軟件在英語(yǔ)和日語(yǔ)得分上的巨大進(jìn)步。

然而，他們承認(rèn)軟件還有更多的工作要做。

在物理測(cè)試中，智能軟件項(xiàng)目不能夠掌握不真實(shí)的情景，例如沒(méi)有質(zhì)量的物體，或沒(méi)有摩擦力的環(huán)境。

它也糾結(jié)于從未聽(tīng)聞的概念。例如，它不知道什么是民主，因?yàn)樗€未學(xué)習(xí)少數(shù)服從多數(shù)原則，社會(huì)公平概念和其他相關(guān)信息。

在英語(yǔ)測(cè)試中，它不能夠回答使用到數(shù)字和插圖的問(wèn)題。

研究人員稱，在識(shí)別照片方面，智能軟件表現(xiàn)更佳。