one of the fundamental problems of cognitive science

馬克．吐溫用一句妙語概括了， 我認為是認知科學的一個最根本問題。

with a single witticism.

他說:「科學的有趣之處在於，

He said, "There's something fascinating about science.

一個人可從微不足道的事得出了偉大的猜想。」

One gets such wholesale returns of conjecture

(笑聲)

out of such a trifling investment in fact."

馬克．吐溫當然只是開玩笑，但他是對的。

(Laughter)

科學有其有趣之處。

Twain meant it as a joke, of course, but he's right:

從幾塊骨頭，我們推測了恐龍的存在;

There's something fascinating about science.

從譜線得出了星雲的成份;

From a few bones, we infer the existence of dinosuars.

從果蠅得出了遺傳的機制;

From spectral lines, the composition of nebulae.

以及從血液流入大腦的重建影像，

From fruit flies,

在我的研究則是從幼兒的行為中，

the mechanisms of heredity,

我們嘗試解釋人類認知的基本機制。

and from reconstructed images of blood flowing through the brain,

我在麻省理工大腦及認知科學系實驗室中，

or in my case, from the behavior of very young children,

花了過去十年研究一個謎團，

we try to say something about the fundamental mechanisms

就是兒童如何從零開始， 快速地學到那麼多的東西。

of human cognition.

科學令人著迷之處，

In particular, in my lab in the Department of Brain and Cognitive Sciences at MIT,

亦正是孩子令人著迷的地方。

I have spent the past decade trying to understand the mystery

回應馬克．吐溫的話，

of how children learn so much from so little so quickly.

那就是孩子從零碎和離亂的訊息中， 能夠得出豐富而抽象的推論的能力。

Because, it turns out that the fascinating thing about science

我將舉出兩個例子:

is also a fascinating thing about children,

一個是關於廣義化的問題，

which, to put a gentler spin on Mark Twain,

另一個則是關於因果推理的。

is precisely their ability to draw rich, abstract inferences

雖然我將會談及我實驗室的研究，

rapidly and accurately from sparse, noisy data.

但這個研究的靈感是來自一個領域，

I'm going to give you just two examples today.

一個我要感謝世界各地的導師、 同事和工作夥伴付出的領域。

One is about a problem of generalization,

讓我先談談廣義化的問題。

and the other is about a problem of causal reasoning.

歸納數據樣本在科學上是不可或缺的，

And although I'm going to talk about work in my lab,

如我們調查一部分的選民，

this work is inspired by and indebted to a field.

然後預測國家大選的結果。

I'm grateful to mentors, colleagues, and collaborators around the world.

我們觀察一小撮病人在臨床試驗中的反應，

Let me start with the problem of generalization.

然後把藥物帶入市場，

Generalizing from small samples of data is the bread and butter of science.

但只有在整個人口中隨機抽樣才可行。

We poll a tiny fraction of the electorate

當我們刻意挑選樣本，

and we predict the outcome of national elections.

如我們只調查城市中的選民，

We see how a handful of patients responds to treatment in a clinical trial,

又或在治療心臟病的臨床試驗中，

and we bring drugs to a national market.

我們只研究男性，

But this only works if our sample is randomly drawn from the population.

這樣的結果便不能代表整個人口。

If our sample is cherry-picked in some way --

因此科學家著緊抽樣的方法是否隨機。

say, we poll only urban voters,

但這又跟嬰兒有甚麼關係?

or say, in our clinical trials for treatments for heart disease,

嬰兒在任何時候都要歸納數據樣本，

we include only men --

當他們看到幾隻橡皮鴨， 並知道它們浮在水面。

the results may not generalize to the broader population.

又或見到幾個皮球， 並知道它們能彈跳。

So scientists care whether evidence is randomly sampled or not,

從中他們建立對橡膠鴨和皮球的概念，

but what does that have to do with babies?

並將這概念延伸至日後會見到的 所有橡膠鴨和皮球。

Well, babies have to generalize from small samples of data all the time.

嬰兒對橡膠鴨和皮球的這種概括，

They see a few rubber ducks and learn that they float,

他們會運用在每一件事上:

or a few balls and learn that they bounce.

鞋子、船、封蠟、捲心菜和皇帝。

And they develop expectations about ducks and balls

因此嬰兒留意這些細節能否代表整體。

that they're going to extend to rubber ducks and balls

我們一起看看吧。

for the rest of their lives.

我將讓你看兩段短片，

And the kinds of generalizations babies have to make about ducks and balls

這兩段短片分別代表實驗的兩個情況。

they have to make about almost everything:

由於你將看到兩段短片，

shoes and ships and sealing wax and cabbages and kings.

你只會看到兩個嬰兒，

So do babies care whether the tiny bit of evidence they see

而這兩個嬰兒在很多地方都是不同的。

is plausibly representative of a larger population?

但這兩個嬰兒將代表更大的群組，

Let's find out.

你將看到的不同之處則代表 嬰兒行為中的平均差異。

I'm going to show you two movies,

在每一段短片中你會見到嬰兒 在做些他們正常會做的事。

one from each of two conditions of an experiment,

嬰兒本身已是十分神奇的，

and because you're going to see just two movies,

但對我來說他們的神奇之處，

you're going to see just two babies,

也是我想你們留意的地方，

and any two babies differ from each other in innumerable ways.

就是這兩種情況之間的分別。

But these babies, of course, here stand in for groups of babies,

因為這兩段短片唯一不同的地方，

and the differences you're going to see

正是嬰兒將要觀察的資料。

represent average group differences in babies' behavior across conditions.

我們把一些藍色和黃色的球給嬰兒看。

In each movie, you're going to see a baby doing maybe

權孝媛當時是我的學生， 現在則是史丹佛大學的同事。

just exactly what you might expect a baby to do,

她將拿出三個藍色的球，

and we can hardly make babies more magical than they already are.

而當她拿出這些球時， 她會把球擠一下，

But to my mind the magical thing,

讓這些球發出吱吱聲。

and what I want you to pay attention to,

這對於嬰兒來說就像TED一樣，

is the contrast between these two conditions,

是件很美好的事。

because the only thing that differs between these two movies

（笑聲）

is the statistical evidence the babies are going to observe.

從一個裝滿藍色球的箱中， 抽出三個藍色球是件很容易的事。

We're going to show babies a box of blue and yellow balls,

你閉上眼睛也能做到，

and my then-graduate student, now colleague at Stanford, Hyowon Gweon,

這就像隨機抽樣。

is going to pull three blue balls in a row out of this box,

因此當你可以在箱中隨機地抽出 能吱吱叫的物件時，

and when she pulls those balls out, she's going to squeeze them,

也許箱中所有物件都能吱吱叫，

and the balls are going to squeak.

所以嬰兒可能會假設黃色球也能吱吱叫。

And if you're a baby, that's like a TED Talk.

但這些黃色球都有一根棒，

It doesn't get better than that.

所以嬰兒可用它們做些不同的事，

(Laughter)

他們可以拍打或搖動這些球。

But the important point is it's really easy to pull three blue balls in a row

就讓我們看看這嬰兒會做甚麼。

out of a box of mostly blue balls.

(影片) 權孝媛: 看看這個。 (球發出吱吱聲)

You could do that with your eyes closed.

看到這個嗎? (球發出吱吱聲)

It's plausibly a random sample from this population.

很酷吧!

And if you can reach into a box at random and pull out things that squeak,

看看這個。

then maybe everything in the box squeaks.

(球發出吱吱聲)

So maybe babies should expect those yellow balls to squeak as well.

哇!

Now, those yellow balls have funny sticks on the end,

羅拉·舒爾茨: 早就說了。 (笑聲)

so babies could do other things with them if they wanted to.

(影片) 孝媛: 看到這個嗎? (球發出吱吱聲)

They could pound them or whack them.

克拉拉, 這個是給你的, 你拿去玩吧。

But let's see what the baby does.

(笑聲)

(Video) Hyowon Gweon: See this? (Ball squeaks)

羅拉: 我不用解釋, 對吧?

Did you see that? (Ball squeaks)

嬰兒把藍色球的特性套用到黃色球上。

Cool.

嬰兒從模仿我們中學習，這是很神奇的，

See this one?

但我們早就知道嬰兒能這樣做。

(Ball squeaks)

有趣的地方是當把一樣的東西給嬰兒看時， 甚麼事會發生。

Wow.

我們能肯定這是完全一樣的， 因我們有個秘密的空間，

Laura Schulz: Told you. (Laughs)

從中我們抽出這些球。

(Video) HG: See this one? (Ball squeaks)

但這次我們改變了抽樣的母體。

Hey Clara, this one's for you. You can go ahead and play.

這次我們在一個裝滿黃色球的箱中， 抽出三個藍色球給嬰兒看。

(Laughter)

想想甚麼事會發生?

LS: I don't even have to talk, right?

你大概不能隨機地在裝滿黃色球的箱中， 連續抽出三個藍色球，

All right, it's nice that babies will generalize properties

因此這很可能不是隨機抽樣。

of blue balls to yellow balls,

這反映了孝媛可能是刻意抽出藍色球，

and it's impressive that babies can learn from imitating us,

可能這些藍色球是特別的，

but we've known those things about babies for a very long time.

可能只有藍色球能吱吱叫。

The really interesting question

一起看看這嬰兒會做甚麼。

is what happens when we show babies exactly the same thing,

(影片) 孝媛: 看看這個。 (球發出吱吱聲)

and we can ensure it's exactly the same because we have a secret compartment

看到這個玩具嗎? (球發出吱吱聲)

and we actually pull the balls from there,

哇, 這很酷, 看到嗎? (球發出吱吱聲)

but this time, all we change is the apparent population

這個是給你的, 你拿去玩吧。

from which that evidence was drawn.

(不耐煩的) (笑聲)

This time, we're going to show babies three blue balls

羅拉: 你剛剛看到兩個15月大的嬰兒，

pulled out of a box of mostly yellow balls,

按他們觀察到樣本出現的機率， 而做出完全不同的事。

and guess what?

一起看看實驗的結果，

You [probably won't] randomly draw three blue balls in a row

垂直軸代表在每一個情況中， 有多少百分比的嬰兒擠壓球。

out of a box of mostly yellow balls.

你可看見嬰兒在樣本和整體一致時，

That is not plausibly randomly sampled evidence.

比刻意挑選的樣本，

That evidence suggests that maybe Hyowon was deliberately sampling the blue balls.

較會歸納他們看到的特徵。

Maybe there's something special about the blue balls.

因此這帶出一個有趣的預測。

Maybe only the blue balls squeak.

假設你在一個裝滿黃色球的箱中 只拿出一個藍色球，

Let's see what the baby does.

當然你很難隨機地連續抽出三個藍色球，

(Video) HG: See this? (Ball squeaks)

但你可以只用一個藍色球作樣本，

See this toy? (Ball squeaks)

這不一定是個不可行的樣本。

Oh, that was cool. See? (Ball squeaks)

當你隨機抽出一個會吱吱叫的東西時，

Now this one's for you to play. You can go ahead and play.

可能箱中所有的東西都會吱吱叫，

(Fussing) (Laughter)

因此雖然嬰兒會看到較少吱吱叫的例子，

LS: So you just saw two 15-month-old babies

而且在只抽出一個球的情況下， 他們會有較少的動作去模仿，

do entirely different things

但我們預計會有更多嬰兒擠壓球。

based only on the probability of the sample they observed.

這正是我們發現的結果。

Let me show you the experimental results.

因此15月大的嬰兒在這方面就像科學家，

On the vertical axis, you'll see the percentage of babies

他們留意抽樣的方法是否隨機，

who squeezed the ball in each condition,

並以此建立對事物的概念:

and as you'll see, babies are much more likely to generalize the evidence

甚麼會吱吱叫而甚麼不會，

when it's plausibly representative of the population

甚麼需要探索而甚麼可忽略。

than when the evidence is clearly cherry-picked.

現在讓我給你們看看另一個例子，

And this leads to a fun prediction:

這次是關於因果推理的。

Suppose you pulled just one blue ball out of the mostly yellow box.

每人都要面對這個問題，

You [probably won't] pull three blue balls in a row at random out of a yellow box,

因為我們都是這世界的一部份。

but you could randomly sample just one blue ball.

這看似不是一個問題， 但和其他問題一樣，

That's not an improbable sample.

事情會出狀況。

And if you could reach into a box at random

以這個嬰兒為例，

and pull out something that squeaks, maybe everything in the box squeaks.

所有事都出了問題，

So even though babies are going to see much less evidence for squeaking,

他想開動這個玩具，但他做不到。

and have many fewer actions to imitate

我會讓你看一段幾秒的影片。

in this one ball condition than in the condition you just saw,

這有兩個可能的原因，

we predicted that babies themselves would squeeze more,

可能是他做錯了一些事，

and that's exactly what we found.

又或是那個玩具有些問題。

So 15-month-old babies, in this respect, like scientists,

因此在這個實驗中，

care whether evidence is randomly sampled or not,

我們會給嬰兒們少許資料。

and they use this to develop expectations about the world:

這些資料會傾向支持其中一個可能性，

what squeaks and what doesn't,

我們將研究這些嬰兒能否運用這些資料，

what to explore and what to ignore.

而作出不同的決定。

Let me show you another example now,

這個實驗是這樣的:

this time about a problem of causal reasoning.

孝媛嘗試開動那個玩具並成功了，

And it starts with a problem of confounded evidence

而我的兩次嘗試都失敗了，

that all of us have,

之後孝媛再嘗試，並再次成功了。

which is that we are part of the world.

這就像我和我的學生在使用新科技的情況。

And this might not seem like a problem to you, but like most problems,

重要的是這提供了少許的資料，

it's only a problem when things go wrong.

這反映玩具並沒有問題，而是人的問題。

Take this baby, for instance.

有些人可以開動這玩具，

Things are going wrong for him.

有些人則不能。

He would like to make this toy go, and he can't.

當這嬰兒拿到玩具時，他要作一個選擇。

I'll show you a few-second clip.

他的母親在旁，

And there's two possibilities, broadly:

所以他可以把玩具交給母親， 換另一人嘗試。

Maybe he's doing something wrong,

同時在毛巾上有另一個玩具，

or maybe there's something wrong with the toy.

所以他也可以把玩具拉向自己， 換另一個玩具。

So in this next experiment,

一起看看嬰兒會怎樣做。

we're going to give babies just a tiny bit of statistical data

(影片) 孝媛: 二、三、開始! (音樂)

supporting one hypothesis over the other,

羅拉: 一、二、三、開始!

and we're going to see if babies can use that to make different decisions

亞瑟，讓我再試一次， 一、二、三、開始!

about what to do.

孝媛: 亞瑟，讓我再試吧，好嗎?

Here's the setup.

一、二、三、開始! (音樂)

Hyowon is going to try to make the toy go and succeed.

看看這裡，記得這些玩具嗎?

I am then going to try twice and fail both times,

看到嗎? 對，我會把這個放在這裡，

and then Hyowon is going to try again and succeed,

把另一個給你。

and this roughly sums up my relationship to my graduate students

你拿去玩吧。

in technology across the board.

羅拉: 你或許會說嬰兒都愛他們的母親，

But the important point here is it provides a little bit of evidence

因此當玩具出現問題時， 嬰兒自然會把它交給母親。

that the problem isn't with the toy, it's with the person.

因此，問題在於當我們稍微改變資料時， 甚麼事會發生。

Some people can make this toy go,

這次，嬰兒將看到這玩具 按同一次序成功運作和失敗，

and some can't.

但我們改變了資料的分佈。

Now, when the baby gets the toy, he's going to have a choice.

這次孝媛和我各有一次成功和一次失敗，

His mom is right there,

這代表誰人嘗試都沒有分別， 那件玩具是壞的，

so he can go ahead and hand off the toy and change the person,

它不是每次都能運作的。

but there's also going to be another toy at the end of that cloth,

同樣地，嬰兒要作出一個選擇，

and he can pull the cloth towards him and change the toy.

她的母親在旁，所以她可換另一人嘗試，

So let's see what the baby does.

同時另一個玩具就在毛巾上。

(Video) HG: Two, three. Go! (Music)

看看她會怎樣做。

LS: One, two, three, go!

(影片) 孝媛: 二、三、開始! (音樂)

Arthur, I'm going to try again. One, two, three, go!

讓我再試一次， 一、二、三、開始!

YG: Arthur, let me try again, okay?

嗯...

One, two, three, go! (Music)

羅拉: 讓我試試吧，克拉拉。

Look at that. Remember these toys?

一、二、三、開始!

See these toys? Yeah, I'm going to put this one over here,

嗯...讓我再試試。

and I'm going to give this one to you.

一、二、三、開始! (音樂)

You can go ahead and play.

孝媛: 我把這個放在這裡，

LS: Okay, Laura, but of course, babies love their mommies.

這個則交給你，

Of course babies give toys to their mommies

你拿去玩吧。

when they can't make them work.

(掌聲)

So again, the really important question is what happens when we change

羅拉: 看看這個實驗的結果，

the statistical data ever so slightly.

在垂直軸上，你會看到在每種情況下， 嬰兒作出不同選擇的分佈。

This time, babies are going to see the toy work and fail in exactly the same order,

你會發現他們作的選擇是 基於他們觀察到的資料。

but we're changing the distribution of evidence.

因此當他們兩歲時，

This time, Hyowon is going to succeed once and fail once, and so am I.

嬰兒已經可以運用細微的資料，

And this suggests it doesn't matter who tries this toy, the toy is broken.

在兩個完全不同的選項中作出決定:

It doesn't work all the time.

尋求幫忙或自行探索。

Again, the baby's going to have a choice.

我剛才讓你們看了兩個實驗，

Her mom is right next to her, so she can change the person,

在這領域中有數千個得出相同結果的實驗。

and there's going to be another toy at the end of the cloth.

當中反映的重點是，

Let's watch what she does.

兒童擁有充分解讀零碎資訊的能力，

(Video) HG: Two, three, go! (Music)

這超出了所有文化的學習方式。

Let me try one more time. One, two, three, go!

孩子從少數的例子便能學到新技能，

Hmm.

他們從少數的例子便能領略到新的因果關係，

LS: Let me try, Clara.

他們甚至能學到新的生字，如美國手語。

One, two, three, go!

我會提出兩個重點作總結。

Hmm, let me try again.

如果你近年有留意大腦和認知科學領域，

One, two, three, go! (Music)

你會聽到三個重要的概念。

HG: I'm going to put this one over here,

第一，現在是大腦的時代。

and I'm going to give this one to you.

的確，神經科學近來有不少驚人的發現，

You can go ahead and play.

例如標記了大腦皮層負責不同功能的位置、

(Applause)

製造出透明的老鼠大腦、

LS: Let me show you the experimental results.

以及利用光線啟動神經元。

On the vertical axis, you'll see the distribution

第二個重要的概念是，