Can a computer write poetry?

我有一個問題，

This is a provocative question.

電腦可以寫詩嗎？

You think about it for a minute,

這是個有爭議的問題。

and you suddenly have a bunch of other questions like:

你稍微想一下，

What is a computer?

腦海裡突然就會浮現出 很多其他的問題：

What is poetry?

例如，甚麼是電腦？

What is creativity?

甚麼是詩？

But these are questions

甚麼是創造力？

that people spend their entire lifetime trying to answer,

但這些問題，

not in a single TED Talk.

很多人窮盡一生才能試著給出答案，

So we're going to have to try a different approach.

單單一場ＴＥＤ演說並不能回答。

So up here, we have two poems.

所以，我們必須用不一樣的方法，

One of them is written by a human,

上面這裡有兩首詩，

and the other one's written by a computer.

其中一首是人類寫的，

I'm going to ask you to tell me which one's which.

另一首是電腦寫的。

Have a go:

我會讓各位來分辨哪首是誰寫的，

Poem 1: Little Fly / Thy summer's play, / My thoughtless hand / Has brush'd away.

我們開始吧：

Am I not / A fly like thee? / Or art not thou / A man like me?

１號詩：小蒼蠅，夏天的嘻戲， 我輕率的手，已揮走。

Poem 2: We can feel / Activist through your life's / morning /

難道我，不是像你一樣的蒼蠅， 抑或妳，是像我一樣的人？

Pauses to see, pope I hate the / Non all the night to start a / great otherwise (...)

２號詩：我們可以感受到， 激進派在妳每日生活的清晨出沒

Alright, time's up.

暫且停下感受，那我憎惡的教皇 並非每晚都能開始，一個偉大的其他可能...

Hands up if you think Poem 1 was written by a human.

好的，時間到。

OK, most of you.

認為１號詩是人寫的請舉手，

Hands up if you think Poem 2 was written by a human.

好的，你們大部分都是。

Very brave of you,

認為２號詩是人寫的請舉手，

because the first one was written by the human poet William Blake.

你們很勇敢，

The second one was written by an algorithm

因為第一首詩是由詩人William Blake所寫，

that took all the language from my Facebook feed on one day

第二首詩是由一個演算法所寫出來的，

and then regenerated it algorithmically,

這裡所有的文法是從我 臉書裡一天灌進去的，

according to methods that I'll describe a little bit later on.

然後，用演算法重新製作出來的，

So let's try another test.

關於方法我稍後會提到一些。

Again, you haven't got ages to read this,

我們來做另一個測驗，

so just trust your gut.

我再次說明， 你不用花太多時間去讀它，

Poem 1: A lion roars and a dog barks. It is interesting / and fascinating

所以，相信你的直覺。

that a bird will fly and not / roar or bark. Enthralling stories about animals

１號詩：獅吼，狗吠， 鳥飛，卻不吼也不吠，這真迷人且有趣吶

are in my dreams and I will sing them all if I / am not exhausted or weary.

我夢裡有著關於動物的迷人故事

Poem 2: Oh! kangaroos, sequins, chocolate sodas! / You are really beautiful!

如果我不筋疲力盡或疲憊不堪 我會為他們歌頌。

Pearls, / harmonicas, jujubes, aspirins! All / the stuff they've always talked about (...)

２號詩：喔！袋鼠、亮片、 巧克力蘇打！你們真漂亮！

Alright, time's up.

珍珠、口琴、棗子、阿斯匹林！ 全是他們一直提到的東西（...)

So if you think the first poem was written by a human,

好的，時間到。

put your hand up.

如果你認為第一首詩是人寫的，

OK.

請舉手。

And if you think the second poem was written by a human,

好的。

put your hand up.

如果你認為第二首詩是人寫的，

We have, more or less, a 50/50 split here.

請舉手。

It was much harder.

我們這裡大約是50／50比例，

The answer is,

這題比較難一點。

the first poem was generated by an algorithm called Racter,

答案是，

that was created back in the 1970s,

第一首詩是一個名叫Racter的 電腦演算法

and the second poem was written by a guy called Frank O'Hara,

在1970年所創造的，

who happens to be one of my favorite human poets.

第二首詩是一位叫 Frank O'Hara的傢伙寫的，

(Laughter)

他意外地成為我最喜歡 的“ 人類詩人”其中之一，

So what we've just done now is a Turing test for poetry.

（笑聲)

The Turing test was first proposed by this guy, Alan Turing, in 1950,

所以，我們為這首詩 做了「圖靈測試」。

in order to answer the question,

「圖靈測試」在1950年， 由Alan Turing做第一次發表，

can computers think?

是為了回答一個問題：

Alan Turing believed that if a computer was able

「電腦會思考嗎？」

to have a to have a text-based conversation with a human,

Alan Turing相信，如果電腦能夠

with such proficiency such that the human couldn't tell

和人類進行一場流暢的以文字交流，

whether they are talking to a computer or a human,

結果讓人無法分辨

then the computer can be said to have intelligence.

對方是人還是一台電腦，

So in 2013, my friend Benjamin Laird and I,

那麼這台電腦可以被稱呼為 擁有人工智慧。

we created a Turing test for poetry online.

所以在2013年，我的朋友 Benjamin Laird和我，

It's called bot or not,

我們創造了一個 詩的線上圖靈測試程式，

and you can go and play it for yourselves.

叫做「bot or not」（是不是機器人），

But basically, it's the game we just played.

你可以上線自己玩玩看。

You're presented with a poem,

但基本上，它就是我們剛剛玩的遊戲，

you don't know whether it was written by a human or a computer

你會看到一首詩，

and you have to guess.

你不知道它是人寫的還是電腦寫的，

So thousands and thousands of people have taken this test online,

然後你必須猜一猜。

so we have results.

好幾千人已經在線上做測驗，

And what are the results?

所以，我們有一個結論，

Well, Turing said that if a computer could fool a human

那結論是甚麼呢？

30 percent of the time that it was a human,

Turing說如果電腦可以騙過30%的人，

then it passes the Turing test for intelligence.

那它就可以被當作人，

We have poems on the bot or not database

它就通過了圖靈測試。

that have fooled 65 percent of human readers into thinking

我們在 bot or not 資料庫裡的詩集

it was written by a human.

已經騙過65% 的人，

So, I think we have an answer to our question.

認為裡面的詩是人寫的。

According to the logic of the Turing test,

所以，我認為我們的問題有答案了，

can a computer write poetry?

根據圖靈測試的邏輯，

Well, yes, absolutely it can.

電腦可以寫詩嗎？

But if you're feeling a little bit uncomfortable

是的，它絕對可以。

with this answer, that's OK.

但，如果你覺得對這答案 有點讓你不太舒服，

If you're having a bunch of gut reactions to it,

也沒關係，

that's also OK because this isn't the end of the story.

如果你花了很多時間與它互動，

Let's play our third and final test.

這也沒關係，因為這還沒完。

Again, you're going to have to read

我們來玩第三個 最後一個測驗，

and tell me which you think is human.

我再說明一下，你們要讀完後，

Poem 1: Red flags the reason for pretty flags. / And ribbons.

告訴我哪一個是人寫的。

Ribbons of flags / And wearing material / Reasons for wearing material. (...)

１號詩：紅旗之所以漂亮 除了紅色，還有緞帶

Poem 2: A wounded deer leaps highest, / I've heard the daffodil

旗上的緞帶及耐磨的材質 耐磨材料之所以（．．．）

I've heard the flag to-day / I've heard the hunter tell; /

２號詩：受傷的鹿跳最高， 我聽見水仙在訴說，

'Tis but the ecstasy of death, / And then the brake is almost done (...)

我今天聽旗子說、 我聽到獵人說；

OK, time is up.

這是對死亡的狂喜， 而傷害幾乎已經造成（．．．）

So hands up if you think Poem 1 was written by a human.

好的，時間到。

Hands up if you think Poem 2 was written by a human.

認為１號詩是人寫的請舉手，

Whoa, that's a lot more people.

認為２號詩是人寫的請舉手，

So you'd be surprised to find that Poem 1

哇！多很多人！

was written by the very human poet Gertrude Stein.

你會很驚訝地發現，

And Poem 2 was generated by an algorithm called RKCP.

１號詩由一位純正的人類詩人Gertrude Stein所寫的，

Now before we go on, let me describe very quickly and simply,

而２號詩是一個叫 RKCP演算法所創造的，

how RKCP works.

在我們要繼續以前， 讓我簡單快速描述一下

So RKCP is an algorithm designed by Ray Kurzweil,

RKCP是如何運作的。

who's a director of engineering at Google

RKCP是Ray Kurzweil 所設計的演算法，

and a firm believer in artificial intelligence.

他是一位谷歌的工程師主管，

So, you give RKCP a source text,

也是一位人工智慧的堅定支持者。

it analyzes the source text in order to find out how it uses language,

那麼，你給 RKCP一個來源文字，

and then it regenerates language

為了找出要如何使用這個語言， 它會分析來源文字，

that emulates that first text.

然後，它會重新創造一段話來模仿源文字。

So in the poem we just saw before,

所以，我們剛剛看到的詩，

Poem 2, the one that you all thought was human,

你們認為是人類寫的２號詩，

it was fed a bunch of poems

它被灌入了很多一位名叫 Emily Dickinson詩人的詩，

by a poet called Emily Dickinson

它取用了這位詩人的語言，

it looked at the way she used language,

學習她的模式，

learned the model,

然後它依據同樣的結構 重製一首詩出來。

and then it regenerated a model according to that same structure.

但我們對RKCP最需要了解的是，

But the important thing to know about RKCP

它不明白它自己用的文字意義，

is that it doesn't know the meaning of the words it's using.

語言只是它的原料，

The language is just raw material,

它可以是中文，瑞典文，

it could be Chinese, it could be in Swedish,

它可以是你臉書上一天的文字。

it could be the collected language from your Facebook feed for one day.

它就只是個原料而已。

It's just raw material.

除此之外，它還有辦法寫一首

And nevertheless, it's able to create a poem

比Gertrude Stein寫的還要更有人味的詩，

that seems more human than Gertrude Stein's poem,

但Gertrude Stein才是人啊...

and Gertrude Stein is a human.

所以，我們剛剛做的 差不多就是，反向圖靈測試。

So what we've done here is, more or less, a reverse Turing test.

所以Gertrude Stein這位人類，

So Gertrude Stein, who's a human, is able to write a poem

可以寫出讓大部分人 誤認為是電腦寫出來的詩。

that fools a majority of human judges into thinking

所以，根據圖靈測試的邏輯，

that it was written by a computer.

Gertrude Stein這人是個電腦．．．（笑聲）

Therefore, according to the logic of the reverse Turing test,

感覺很困惑嗎？

Gertrude Stein is a computer.

我認為這情有可原。

(Laughter)

目前為止，我們有人可以寫出 像是人寫出的詩、

Feeling confused?

我們有電腦可以寫出 像是電腦寫出的詩、

I think that's fair enough.

我們有電腦可以寫出 像是人寫出的詩，

So far we've had humans that write like humans,

但我們同時也有會讓我們搞混 寫詩像電腦的人。

we have computers that write like computers,

所以，我們從這裏面了解到甚麼呢？

we have computers that write like humans,

我們會認為William Blake

but we also have, perhaps most confusingly,

比Gertrude Stein更像是個人嗎？

humans that write like computers.

或者Gertrude Stein比 William Blake更像是個電腦？

So what do we take from all of this?

（笑聲）

Do we take that William Blake is somehow more of a human

這兩年來，

than Gertrude Stein?

我一直在問我自己，

Or that Gertrude Stein is more of a computer than William Blake?

但我沒有任何答案，

(Laughter)

但我真的有領悟到很多有關於

These are questions I've been asking myself

我們與科技的關係。

for around two years now,

所以，我的第一個領悟是，

and I don't have any answers.

為了一些原因，我們把 人與詩結合一起，

But what I do have are a bunch of insights

所以，當我們問，＂電腦會寫詩嗎？＂

about our relationship with technology.

我們也在問，

So my first insight is that, for some reason,

人的定義是什麼？

we associate poetry with being human.

我們要如何界定、分類呢？

So that when we ask, "Can a computer write poetry?"

我們要如何分辨誰或是東西 是歸於哪一類？＂

we're also asking,

我相信，本質上這是一道哲學的問題，

"What does it mean to be human

而且，這不是像圖靈測試是個

and how do we put boundaries around this category?

對或錯的測試，

How do we say who or what can be part of this category?"

我也相信，

This is an essentially philosophical question, I believe,

Alan Turing在1950年發明這個理論時， 也了解這一點，

and it can't be answered with a yes or no test,

他當時引發了一個哲學上的爭議。

like the Turing test.

我的第二個領悟是， 當我們在為詩做圖靈測試時，

I also believe that Alan Turing understood this,

我們並不是真的在測試電腦的能力，

and that when he devised his test back in 1950,

因為用演算法作詩相當簡單，

he was doing it as a philosophical provocation.

而且它們大約在1950年代 早就已經存在了。

So my second insight is that, when we take the Turing test for poetry,

我們現在為詩做的圖靈測試，

we're not really testing the capacity of the computers

反而，比較像是在收集 甚麼是構成人性的條件。

because poetry-generating algorithms,

所以，我發現，

they're pretty simple and have existed, more or less, since the 1950s.

稍早我們今天看到的，

What we are doing with the Turing test for poetry, rather,

我們說William Blake

is collecting opinions about what constitutes humanness.

比Gertrude Stein更像個人，

So, what I've figured out,

當然，這不代表

we've seen this when earlier today,

William Blake比較有人性

we say that William Blake is more of a human

或者Gertrude Stein比較像電腦。

than Gertrude Stein.

這只能單純的說明， 對人類的界定是不穩定的。

Of course, this doesn't mean that William Blake

這讓我明白了一件事，

was actually more human

就是人性不是冷的、死板的事實，

or that Gertrude Stein was more of a computer.

反倒是一種由我們 的意見所構成的東西，

It simply means that the category of the human is unstable.

而這個東西會隨著時間而改變。

This has led me to understand

所以我最後的領悟是，

that the human is not a cold, hard fact.

電腦，或多或少只是

Rather, it is something that's constructed with our opinions

一面反映我們輸入進去的人類思想的鏡子。

and something that changes over time.

我們向它展示Emily Dickinson，

So my final insight is that the computer, more or less,

它僅是模仿Emily Dickinson給我們，

works like a mirror that reflects any idea of a human

我們向它展示William Blake，

that we show it.

它就回應William Blake給我們的，

We show it Emily Dickinson,

我們向它展示Gertrude Stein，

it gives Emily Dickinson back to us.

我們得到的回應僅是Gertrude Stein。

We show it William Blake,

還有其他更多的科技也是，

that's what it reflects back to us.

電腦只是我們教它甚麼 它就反應甚麼的一面鏡子。

We show it Gertrude Stein,

所以，我確定你們大部分人都曾聽過

what we get back is Gertrude Stein.

很多有關人工智慧的事情。

More than any other bit of technology,

而大部分的對話就類似：

the computer is a mirror that reflects any idea of the human we teach it.

「我們該建造它嗎？」

So I'm sure a lot of you have been hearing

「我們可以建立一個智慧型電腦嗎？」

a lot about artificial intelligence recently.

「我們可以建立一個創造型電腦嗎？」

And much of the conversation is,

我們一次又一次的被問到，

can we build it?

我們可以建立一個 類似人類的電腦嗎？

Can we build an intelligent computer?

但就我們剛剛看到的，

Can we build a creative computer?

人類不是一個科學事實，

What we seem to be asking over and over

人類是一個會不斷地變化、串聯想法、

is can we build a human-like computer?

隨時間改變的物種。

But what we've seen just now

所以，當我們開始要努力克服

is that the human is not a scientific fact,

未來人工智慧的這個想法時，

that it's an ever-shifting, concatenating idea

我們不應該只問我們自己，

and one that changes over time.

「我們可以建造它嗎？」

So that when we begin to grapple with the ideas

我們還得問我們自己，

of artificial intelligence in the future,

「我們希望可以得到甚麼樣的人性回應？」

we shouldn't only be asking ourselves,

這絕對是個哲學想法，

"Can we build it?"

而且不是單靠軟體就可以回答出來的，

But we should also be asking ourselves,

但我認為，這需要一個各類物種 共存的反應時刻，

"What idea of the human do we want to have reflected back to us?"

謝謝各位。

This is an essentially philosophical idea,

（掌聲）

and it's one that can't be answered with software alone,

but I think requires a moment of species-wide, existential reflection.

Thank you.

(Applause)