Placeholder Image

字幕列表 影片播放

  • Today, I'm going to talk about AI and us.

    譯者: Judy Wan-Ling CHEN 審譯者: Wilde Luo

  • AI researchers have always said

    今天我要跟大家聊聊 「人工智慧與你我」,

  • that we humans do not need to worry,

    人工智慧研究專家常說,

  • because only menial jobs will be taken over by machines.

    人類不需要擔心,

  • Is that really true?

    因為機器只會取代那些 乏味枯燥的粗活。

  • They have also said that AI will create new jobs,

    是真的嗎?

  • so those who lose their jobs will find a new one.

    他們也說過人工智慧 可以創造新的工作,

  • Of course.

    因此那些失業的人還是可以再就業。

  • But the real question is:

    當然。

  • How many of those who may lose their jobs to AI

    只是問題的癥結是:

  • will be able to land a new one,

    多少人因為人工智慧失業後

  • especially when AI is smart enough to learn better than most of us?

    能夠真的找到新的工作,

  • Let me ask you a question:

    尤其當人工智慧已經成熟到 比大多數人都能更有效率地學習?

  • How many of you think

    讓我問你們一個問題:

  • that AI will pass the entrance examination of a top university by 2020?

    你們之中認為在 2020 年之前

  • Oh, so many. OK.

    人工智慧將可以通過 頂尖大學的入學測驗的人,請舉手。

  • So some of you may say, "Of course, yes!"

    哇,好多。好的。

  • Now singularity is the issue.

    相當多人會認為:「那當然!」

  • And some others may say, "Maybe,

    現在「人工智慧的奇點」 是個熱門話題。

  • because AI already won against a top Go player."

    也有很多人會說:「有可能,

  • And others may say, "No, never. Uh-uh."

    人工智慧都已經打敗過 世界頂尖的圍棋高手了。」

  • That means we do not know the answer yet, right?

    也有一些人持不同看法: 「絕對不可能。嗯,就是這樣。」

  • So that was the reason why I started Todai Robot Project,

    這說明了我們其實還沒有 真正的答案,不是嗎?

  • making an AI which passes the entrance examination

    這就是我啟動 「東大機器人計畫」的始末,

  • of the University of Tokyo,

    打造一個人工智慧機器人,

  • the top university in Japan.

    通過東京大學的入學考,

  • This is our Todai Robot.

    東大是全日本最頂尖的學府。

  • And, of course, the brain of the robot is working in the remote server.

    為您介紹東大機器人。

  • It is now writing a 600-word essay

    當然,它正接受遠端伺服器遙控著。

  • on maritime trade in the 17th century.

    正在寫一篇 600 字的論文,

  • How does that sound?

    闡述 17 世紀的海上貿易。

  • Why did I take the entrance exam as its benchmark?

    聽起來如何?

  • Because I thought we had to study the performance of AI

    為什麼當初要將入學考試 作為一個標竿?

  • in comparison to humans,

    因為我想,我們有必要研究人工智慧

  • especially on the skills and expertise

    相較人類的表現,

  • which are believed to be acquired only by humans

    尤其那些有規模 跟特殊性的專長領域,

  • and only through education.

    向來我們都相信 這些技能和知識惟有人類

  • To enter Todai, the University of Tokyo,

    透過教育才能獲得。

  • you have to pass two different types of exams.

    要考上日本的 第一學府,東京大學,

  • The first one is a national standardized test

    必須通過兩項測驗。

  • in multiple-choice style.

    第一個是日本高考, 全國標準化的測驗,

  • You have to take seven subjects

    全選擇題的題型。

  • and achieve a high score --

    你必須在七大科目中

  • I would say like an 85 percent or more accuracy rate --

    都獲得高分——

  • to be allowed to take the second stage written test

    大概要達到 85% 以上的正確率——

  • prepared by Todai.

    才可以獲准進入第二階段筆試,

  • So let me first explain how modern AI works,

    由東大命題。

  • taking the "Jeopardy!" challenge as an example.

    先讓我說明一下 當代的人工智慧運作的方式,

  • Here is a typical "Jeopardy!" question:

    舉一個 「Jeopardy!」 《危險邊緣》 的例子。

  • "Mozart's last symphony shares its name with this planet."

    典型的一道題目如下:

  • Interestingly, a "Jeopardy!" question always asks,

    「莫札特最後創作的交響曲 跟這個星球同名。」

  • always ends with "this" something:

    有趣的是,

  • "this" planet, "this" country,

    《危險邊緣》的例子總是有

  • "this" rock musician, and so on.

    「這個」字眼在裡頭:

  • In other words, "Jeopardy!" doesn't ask many different types of questions,

    「這個」星球、「這個」國家,

  • but a single type,

    「這個」搖滾樂手 或「這個」什麼什麼。

  • which we call "factoid questions."

    也就是說,《危險邊緣》 沒有太多類型的題目,

  • By the way, do you know the answer?

    幾乎只有這一類,

  • If you do not know the answer and if you want to know the answer,

    我們稱作「趣味小問題」。

  • what would you do?

    有沒有人剛好知道答案呀?

  • You Google, right? Of course.

    假如我們不知道答案, 但是又想要知道,

  • Why not?

    怎麼辦?

  • But you have to pick appropriate keywords

    當然是上網搜尋呀——

  • like "Mozart," "last" and "symphony" to search.

    為甚麼不呢?

  • The machine basically does the same.

    但要挑對關鍵字,

  • Then this Wikipedia page will be ranked top.

    譬如輸入像「莫札特」、 「最後」跟「交響曲」去搜尋。

  • Then the machine reads the page.

    基本上機器人也是這麼做。

  • No, uh-uh.

    維基百科的頁面就會出現在最上頭。

  • Unfortunately, none of the modern AIs,

    所以機器人就開始 「讀」這個頁面嗎?

  • including Watson, Siri and Todai Robot,

    並不是喔──

  • is able to read.

    不幸的是,所有當代的智慧機器人,

  • But they are very good at searching and optimizing.

    無論是 IBM 的華生、蘋果的 Siri 或是東大機器人,

  • It will recognize

    它們都沒有「閱讀」的能力。

  • that the keywords "Mozart," "last" and "symphony"

    但是它們在搜尋跟 得到最佳化結果上很在行。

  • are appearing heavily around here.

    它會找到關鍵字像是

  • So if it can find a word which is a planet

    「莫札特」、「最後」 跟「交響曲」,

  • and which is co-occurring with these keywords,

    重複地出現在這一帶。

  • that must be the answer.

    接著繼續尋找屬於星球的詞彙,

  • This is how Watson finds the answer "Jupiter," in this case.

    是跟前述這些關鍵字同時出現的,

  • Our Todai Robot works similarly, but a bit smarter

    那鐵定就是答案了。

  • in answering history yes-no questions,

    華生就是這樣找到「木星」的。

  • like, "'Charlemagne repelled the Magyars.' Is this sentence true or false?"

    東大機器人的運作方式很接近, 但在回答歷史科目的

  • Our robot starts producing a factoid question,

    判斷題上表現稍好,

  • like: "Charlemagne repelled [this person type]" by itself.

    例如: 查理曼大帝擊敗馬札爾人,

  • Then, "Avars" but not "Magyars" is ranked top.

    對還是錯?

  • This sentence is likely to be false.

    機器人自動轉換為一道 趣味小問題,

  • Our robot does not read, does not understand,

    變成:「 查理曼大帝 擊敗了這一種人」。

  • but it is statistically correct in many cases.

    結果最上頭出現了「阿瓦爾人」 而非「馬札爾人」。

  • For the second stage written test,

    所以這個陳述句很可能是錯誤的。

  • it is required to write a 600-word essay like this one:

    機器人不會閱讀,也不了解,

  • [Discuss the rise and fall of the maritime trade

    但從統計學角度評估, 卻具有高準確度。

  • in East and Southeast Asia in the 17th century ...]

    至於第二階段的筆試,

  • and as I have shown earlier,

    受測者必須寫一篇 600 字的論文,

  • our robot took the sentences from the textbooks and Wikipedia,

    如這一道題:

  • combined them together,

    (闡述 17 世紀時東亞與東南亞 海上貿易的興衰……)

  • and optimized it to produce an essay

    如同我稍早展示過的,

  • without understanding a thing.

    我們的機器人將教科書 與維基百科的句子

  • (Laughter)

    併在一起,

  • But surprisingly, it wrote a better essay

    優化後形成一篇文章,

  • than most of the students.

    完全不懂字裡行間的意涵。

  • (Laughter)

    (笑聲)

  • How about mathematics?

    但是令人驚訝的是, 機器人這樣寫出來的文章,

  • A fully automatic math-solving machine

    居然比大多數的學生好。

  • has been a dream

    (笑聲)

  • since the birth of the word "artificial intelligence,"

    那麼數學呢?

  • but it has stayed at the level of arithmetic for a long, long time.

    能全自動處理數學問題的機器人,

  • Last year, we finally succeeded in developing a system

    是大家都夢寐以求的,

  • which solved pre-university-level problems from end to end,

    打從「人工智慧」的概念 問世以來就是如此。

  • like this one.

    但是,它曾經 長期停滯在算術的階段。

  • This is the original problem written in Japanese,

    去年我們總算成功發展一套系統,

  • and we had to teach it 2,000 mathematical axioms

    可以從頭到尾地 解決中等教育程度的數學題目,

  • and 8,000 Japanese words

    像這一題。

  • to make it accept the problems written in natural language.

    原文是日文。

  • And it is now translating the original problems

    我們必須先教會機器人 2,000 個數學公理,

  • into machine-readable formulas.

    與 8,000 個日文字,

  • Weird, but it is now ready to solve it, I think.

    才能讓機器人 看懂原文的數學題目。

  • Go and solve it.

    它現在正在翻譯原來的題目

  • Yes! It is now executing symbolic computation.

    成為機器人的語言。

  • Even more weird,

    很怪,不過應該可以開始計算了。

  • but probably this is the most fun part for the machine.

    開始解題。

  • (Laughter)

    沒錯!它正在進行符號運算。

  • Now it outputs a perfect answer,

    更怪了,

  • though its proof is impossible to read, even for mathematicians.

    但或許機器人會覺得這個才好玩。

  • Anyway, last year our robot was among the top one percent

    (笑聲)

  • in the second stage written exam in mathematics.

    好了,它產出了一個完美的解。

  • (Applause)

    儘管連數學家都證實了, 完全沒有人看得懂。

  • Thank you.

    無論如何,去年我們的機器人 在第二階段的數學表現中

  • So, did it enter Todai?

    被歸類在排名前 1% 高分的群組中。

  • No, not as I expected.

    (掌聲)

  • Why?

    謝謝。

  • Because it doesn't understand any meaning.

    所以它最終有沒有考上東大呢?

  • Let me show you a typical error it made in the English test.

    它並沒有如預期的金榜題名。

  • [Nate: We're almost at the bookstore. Just a few more minutes.

    為什麼?

  • Sunil: Wait. ______ . Nate: Thank you! That always happens ...]

    因為它根本什麼也不懂。

  • Two people are talking.

    讓我展示一個在英文科的典型錯誤。

  • For us, who can understand the situation --

    (奈特:我們快到書店了, 再過幾分鐘就到了。

  • [1. "We walked for a long time." 2. "We're almost there."

    桑妮:等一下。______。 奈特:謝謝!每次都這樣……)

  • 3. "Your shoes look expensive." 4. "Your shoelace is untied."]

    兩個人在對話。

  • it is obvious number four is the correct answer, right?

    我們都明白發生了什麽——

  • But Todai Robot chose number two,

    (選項:1. 我們走了很久的路 2. 我們幾乎快到了

  • even after learning 15 billion English sentences

    3. 你的鞋子看起來好昂貴 4. 你的鞋帶鬆了)

  • using deep learning technologies.

    很明顯地,標準答案 是選 4,同意嗎?

  • OK, so now you might understand what I said:

    可是東大機器人選 2,

  • modern AIs do not read,

    就算已經學習了 150 億個英文詞句,

  • do not understand.

    還透過深度學習技術。

  • They only disguise as if they do.

    好吧,現在你可能明白我所說的:

  • This is the distribution graph

    當代人工智慧沒有辦法閱讀,

  • of half a million students who took the same exam as Todai Robot.

    不能理解。

  • Now our Todai Robot is among the top 20 percent,

    它們只是佯裝成什麽都懂。

  • and it was capable to pass

    這個分佈圖,

  • more than 60 percent of the universities in Japan --

    代表跟東大機器人一起接受入學 考試的其他 50 萬名考生的成績。

  • but not Todai.

    機器人排名其中的前 20%,

  • But see how it is beyond the volume zone

    可以考進日本

  • of to-be white-collar workers.

    超過六成的大學──

  • You might think I was delighted.

    但就是考不上東大。

  • After all, my robot was surpassing students everywhere.

    可是看看被它超越的廣大區塊,

  • Instead, I was alarmed.

    所謂的白領階級。

  • How on earth could this unintelligent machine outperform students --

    你可能會猜想我應該很開心。

  • our children?

    畢竟我的機器人正在 全面性地超越學生。

  • Right?

    其實不然,我很驚恐。

  • I decided to investigate what was going on in the human world.

    這個一點都不聰明的機器人居然 表現得比學生們──

  • I took hundreds of sentences from high school textbooks

    也是我們的孩子們,更好?

  • and made easy multiple-choice quizzes,

    怎麼可以?

  • and asked thousands of high school students to answer.

    我決定深入調查 人類世界究竟發生了什麼事。

  • Here is an example:

    我收集了上百個 高中教科書裡頭的詞句,

  • [Buddhism spread to ... , Christianity to ... and Oceania,

    然後編成簡單的選擇題測驗,

  • and Islam to ...]

    讓上千位高中生接受測試。

  • Of course, the original problems are written in Japanese,

    這是其中一個範例:

  • their mother tongue.

    題目都是用他們的母語 ──日文寫的。

  • [ ______ has spread to Oceania.

    (題目:______傳播到了大洋洲。

  • 1. Hinduism 2. Christianity 3. Islam 4. Buddhism ]

    1. 印度教 2. 基督教 3. 伊斯蘭教 4. 佛教)

  • Obviously, Christianity is the answer, isn't it?

    顯而易見的答案是基督教,對吧?

  • It's written!

    都包含在題目所給信息裡了。

  • And Todai Robot chose the correct answer, too.

    東大機器人也選出了正確的答案。

  • But one-third of junior high school students

    但是有三分之一的國中生

  • failed to answer this question.

    無法回答這個問題。

  • Do you think it is only the case in Japan?

    你以為這個問題只存在於日本嗎?

  • I do not think so,

    我不這麼認為,

  • because Japan is always ranked among the top in OECD PISA tests,

    日本總是在國際學生能力評估計劃 測驗中名列前茅。

  • measuring 15-year-old students' performance in mathematics,

    那是一套衡量 15 歲青少年在數學、

  • science and reading

    科學與閱讀素質的測驗,

  • every three years.

    每三年考一次。

  • We have been believing

    我們一直相信,

  • that everybody can learn

    每個人都能學習,

  • and learn well,

    並且學得出色,

  • as long as we provide good learning materials

    只要我們提供高質量的學習資料,

  • free on the web

    這些免費資源,

  • so that they can access through the internet.

    讓他們透過網路取得使用。

  • But such wonderful materials may benefit only those who can read well,

    但這些優質的資料 只會讓那些能夠有效閱讀的人受益,

  • and the percentage of those who can read well

    而能夠有效閱讀的人所佔的比例,

  • may be much less than we expected.

    可能遠低於我們的預期。

  • How we humans will coexist with AI

    人類要如何與人工智慧共存

  • is something we have to think about carefully,

    是我們要謹慎思考的課題,

  • based on solid evidence.

    客觀考量各項可靠證據。

  • At the same time, we have to think in a hurry

    同時,我們也要加緊腳步思考,

  • because time is running out.

    因為所剩時間不多了。

  • Thank you.

    謝謝。

  • (Applause)

    (掌聲)

  • Chris Anderson: Noriko, thank you.

    克里斯 · 安德森:紀子,謝謝你。

  • Noriko Arai: Thank you.

    新井紀子:謝謝。

  • CA: In your talk, you so beautifully give us a sense of how AIs think,

    克里斯:您方才向我們 說明了人工智慧的運作方式。

  • what they can do amazingly

    它們可以完美勝任的,

  • and what they can't do.

    以及無法勝任的工作。

  • But -- do I read you right,

    但是,我的解讀是否是正確的,

  • that you think we really need quite an urgent revolution in education

    你認為我們急需教育改革,

  • to help kids do the things that humans can do better than AIs?

    以協助學子在特定的領域中 讓人工智慧難以望其項背?

  • NA: Yes, yes, yes.

    紀子:是的。

  • Because we humans can understand the meaning.

    因為我們人類可以理解意義。

  • That is something which is very, very lacking in AI.

    而這在人工智慧中是相當缺乏的。

  • But most of the students just pack the knowledge

    但是大多數的學生都只會囫圇吞棗,

  • without understanding the meaning of the knowledge,

    而非深入理解知識,

  • so that is not knowledge, that is just memorizing,

    這就只是單純記憶的動作,

  • and AI can do the same thing.

    人工智慧也辦得到。

  • So we have to think about a new type of education.

    所以我們應該要思考新型態的教育。

  • CA: A shift from knowledge, rote knowledge, to meaning.

    克里斯:從「死記硬背」 到「深入理解」的轉變。

  • NA: Mm-hmm.

    紀子:嗯嗯嗯。

  • CA: Well, there's a challenge for the educators. Thank you so much.

    克里斯:我想這是給 教育家的一大挑戰,

  • NA: Thank you very much. Thank you.

    再次感謝您。

  • (Applause)

    紀子:謝謝,非常謝謝你們。

Today, I'm going to talk about AI and us.

譯者: Judy Wan-Ling CHEN 審譯者: Wilde Luo

字幕與單字

影片操作 你可以在這邊進行「影片」的調整,以及「字幕」的顯示

A2 初級 中文 美國腔 TED 機器人 人工 題目 測驗 數學

【TED】新井理子。機器人能通過大學入學考試嗎?(機器人能通過大學入學考試嗎?新井典子) (【TED】Noriko Arai: Can a robot pass a university entrance exam? (Can a robot pass a university entrance exam? | Noriko Arai))

  • 59 9
    Zenn 發佈於 2021 年 01 月 14 日
影片單字