A brain in a supercomputer | Henry Markram

譯者: Boyang Zhu 審譯者: Shelley Krishna Tsang

Our mission is to build

我們的任務是建造

a detailed, realistic

一個詳細而真實的

computer model of the human brain.

人類大腦的計算機模型。

And we've done, in the past four years,

在過去幾年里，我們在一小塊嚙齒類動物

a proof of concept

的腦上做了一個

on a small part of the rodent brain,

用來驗證這個概念的測試，

and with this proof of concept we are now scaling the project up

現在根據這個試驗我們要把項目的規模擴展到

to reach the human brain.

人類大腦的規模。

Why are we doing this?

為甚麼我們要做這項工作？

There are three important reasons.

有三個重要的原因。

The first is, it's essential for us to understand the human brain

首先，是理解人類大腦對我們來將是非常重要的

if we do want to get along in society,

如果我們想在社會中繼續前進

and I think that it is a key step in evolution.

我認為這是進化過程中非常關鍵的一步

The second reason is,

第二個原因是

we cannot keep doing animal experimentation forever,

我們不能總是繼續拿動物做試驗

and we have to embody all our data and all our knowledge

還有我們必須要把我們所有的數據和知識收錄進

into a working model.

一個有效的模型當中。

It's like a Noah's Ark. It's like an archive.

就好像諾亞的方舟。好像是一个文庫。

And the third reason is that there are two billion people on the planet

第三個原因是地球上有20億人

that are affected by mental disorder,

的生活被精神障礙所影響。

and the drugs that are used today

而目前所廣泛使用的藥物

are largely empirical.

都是經驗性的

I think that we can come up with very concrete solutions on

我認為我們能夠對治療精神障礙

how to treat disorders.

提出非常堅實的方案。

Now, even at this stage,

即使是在現階段

we can use the brain model

我們可以使用大腦模型

to explore some fundamental questions

來探究一些關於大腦

about how the brain works.

如何運作的根本的問題

And here, at TED, for the first time,

在這裡，TED大會上，第一次

I'd like to share with you how we're addressing

我想與大家分享我們如何來解決這個理論

one theory -- there are many theories --

有許多的理論

one theory of how the brain works.

其中的一個關於大腦如何工作的理論是

So, this theory is that the brain

所以，這個理論是大腦如何

creates, builds, a version of the universe,

創造、建立一個宇宙版本

and projects this version of the universe,

並將這個宇宙版本像泡泡一樣

like a bubble, all around us.

映射在我們的周圍。

Now, this is of course a topic of philosophical debate for centuries.

當然這是一個經過許多個世紀爭論的話題

But, for the first time, we can actually address this,

但是，這是歷史上第一次我們可以實際地利用

with brain simulation,

大腦模擬來解決它，

and ask very systematic and rigorous questions,

並提出非常系統性，非常嚴謹的問題

whether this theory could possibly be true.

這個理論是是不是正確的

The reason why the moon is huge on the horizon

我們感覺地平線上的月亮非常大的原因

is simply because our perceptual bubble

正是因為我們的感知泡泡

does not stretch out 380,000 kilometers.

並沒有延伸到三十八萬公里之外

It runs out of space.

這真是太遠了

And so what we do is we compare the buildings

我們所做的是在我們的感知泡泡中

within our perceptual bubble,

將其與附近的建築做比較，

and we make a decision.

接著我們做出了一個判斷

We make a decision it's that big,

我們判斷它是那麼大的

even though it's not that big.

即使在我們眼裡它並不大

And what that illustrates

這個例子說明的是

is that decisions are the key things

判斷是讓我們的感知泡泡成立

that support our perceptual bubble. It keeps it alive.

並保證它活躍的關鍵因素。

Without decisions you cannot see, you cannot think,

失去了判斷，你既看不見東西，也不能思考

you cannot feel.

甚麼都感覺不到

And you may think that anesthetics work

你可能以為麻醉劑的工作方式是

by sending you into some deep sleep,

讓你進入酣睡的狀態

or by blocking your receptors so that you don't feel pain,

或者阻撓神經受體的運作來讓你感覺不到疼痛

but in fact most anesthetics don't work that way.

但事實上大多數麻醉劑並不是這樣生效的

What they do is they introduce a noise

它們所做的是在大腦中產生一種噪音來

into the brain so that the neurons cannot understand each other.

讓神經元細胞互相之間無法理解

They are confused,

它們被搞糊塗了

and you cannot make a decision.

這樣你就不能做出判斷

So, while you're trying to make up your mind

所以，當你還在努力着集中注意力

what the doctor, the surgeon, is doing

要搞清楚醫生在你身上動手動腳的時候

while he's hacking away at your body, he's long gone.

對你的身體做了些甚麼，他早已經走人了。

He's at home having tea.

他已經在家喝茶了

(Laughter)

笑聲

So, when you walk up to a door and you open it,

當你走到一扇門前並打開它的時候

what you compulsively have to do to perceive

為了理解周圍環境，

is to make decisions,

你不得不做出判斷，

thousands of decisions about the size of the room,

無數的關於房間與牆壁的大小，高度

the walls, the height, the objects in this room.

以及房間里放的是甚麼東西的判斷。

99 percent of what you see

99％你所看見的東西

is not what comes in through the eyes.

並不是通過眼睛觀察到的

It is what you infer about that room.

而是你對房間所做出的推斷

So I can say, with some certainty,

所以我一定程度上同意

"I think, therefore I am."

‘我思故我在’

But I cannot say, "You think, therefore you are,"

但是我卻不能說“你思故你在”

because "you" are within my perceptual bubble.

因為“你”這個概念是存在與我的感知泡泡之中

Now, we can speculate and philosophize this,

目前我們能推測並進行在哲理層面上研究這個理論

but we don't actually have to for the next hundred years.

不過不用再這樣繼續幾百年了

We can ask a very concrete question.

我們可以問這樣一個具體的問題

"Can the brain build such a perception?"

大腦本身可以映射出這些感覺嗎？

Is it capable of doing it?

它是否有這種能力做到這一點？

Does it have the substance to do it?

它有沒有足夠的物質來產生感覺？

And that's what I'm going to describe to you today.

這就是我今天想要向你們描述的主題

So, it took the universe 11 billion years to build the brain.

在這個宇宙中經過了110億年的進化出了大腦

It had to improve it a little bit.

它需要不斷地改進

It had to add to the frontal part, so that you would have instincts,

需要加上一個額部以讓你能夠擁有本能

because they had to cope on land.

因為生物需要應付地面上的環境

But the real big step was the neocortex.

真正的巨大進步是大腦新皮質

It's a new brain. You needed it.

這是一個新的大腦，你需要它

The mammals needed it

哺乳動物需要它

because they had to cope with parenthood,

因為它們需要撫養幼崽

social interactions,

互相交流

complex cognitive functions.

並使用複雜的識別功能

So, you can think of the neocortex

所以你可以把大腦新皮質看成是

actually as the ultimate solution today,

到目前為止我們所知的

of the universe as we know it.

宇宙中的終極產品。

It's the pinnacle, it's the final product

它是一個巔峰，是宇宙

that the universe has produced.

所製造的最後產品。

It was so successful in evolution

它在進化史中是如此的成功

that from mouse to man it expanded

從老鼠到人類，大腦中神經元

about a thousandfold in terms of the numbers of neurons,

的數量擴展了大約一千倍，

to produce this almost frightening

來構成這個幾乎是嚇人的

organ, structure.

器官，結構。

And it has not stopped its evolutionary path.

它也沒有停止進化的步伐

In fact, the neocortex in the human brain

實際上人類大腦中的新皮質層

is evolving at an enormous speed.

一直在以驚人的速度進化。

If you zoom into the surface of the neocortex,

如果你深入新皮質的表面

you discover that it's made up of little modules,

你會發現它是由微小的模塊組成

G5 processors, like in a computer.

就好像電腦里的G5處理器

But there are about a million of them.

但大腦中有大約100萬個模塊

They were so successful in evolution

它們進化的如此成功

that what we did was to duplicate them

因此我們就不斷地複製它們

over and over and add more and more of them to the brain

不斷地在大腦中加入更多的模塊

until we ran out of space in the skull.

直到用盡所有頭顱中的空間

And the brain started to fold in on itself,

大腦自身開始摺疊起來

and that's why the neocortex is so highly convoluted.

這就是為甚麼新皮質是非常的捲曲的

We're just packing in columns,

它們不斷的往縱深發展形成功能住

so that we'd have more neocortical columns

這樣我們就有更多的皮質功能住

to perform more complex functions.

來執行更複雜的機能

So you can think of the neocortex actually as

你也可以將大腦新皮質

a massive grand piano,

看成一架巨大的鋼琴。

a million-key grand piano.

一部有一百萬個琴鍵的大鋼琴

Each of these neocortical columns

其中的每一個皮質功能住

would produce a note.

會奏出一個音符

You stimulate it; it produces a symphony.

你對它施加刺激，它奏出一部交響曲

But it's not just a symphony of perception.

不過這不僅僅是感覺的交響曲

It's a symphony of your universe, your reality.

是你的宇宙的交響曲，你的現實世界

Now, of course it takes years to learn how

當然一個人需要花費很多年來學習如何彈奏

to master a grand piano with a million keys.

一架有着一百萬個琴鍵的鋼琴

That's why you have to send your kids to good schools,

這就是為甚麼你送孩子去好的學校

hopefully eventually to Oxford.

希望最後去到牛津大學

But it's not only education.

不過不只是教育

It's also genetics.

基因也會影響結果。

You may be born lucky,

你可能生來就很有天賦

where you know how to master your neocortical column,

或者你知道如何來操控你的新皮質功能柱

and you can play a fantastic symphony.

來演奏美妙的交響樂

In fact, there is a new theory of autism

關於自閉症有一種是

called the "intense world" theory,

稱作“激烈世界”理論

which suggests that the neocortical columns are super-columns.

它提出這些人的新皮質功能柱是超級功能柱

They are highly reactive, and they are super-plastic,

它們反應非常劇烈，而且非常有可塑性

and so the autists are probably capable of

所以自閉症患者或許可以

building and learning a symphony

構造並學習一個對我們來說

which is unthinkable for us.

無法想像的交響樂。

But you can also understand

同樣也可以理解

that if you have a disease

如果在這些功能柱中

within one of these columns,

產生任何病變，

the note is going to be off.

音調就會有偏差

The perception, the symphony that you create

這些感覺，這些你創造的交響樂

is going to be corrupted,

會被破壞，

and you will have symptoms of disease.

你會有得到有缺陷的交響曲。

So, the Holy Grail for neuroscience

所以神經科學的終極目的是

is really to understand the design of the neocoritical column --

真正地理解新皮質功能柱的設計

and it's not just for neuroscience;

這不光是對神經科學

it's perhaps to understand perception, to understand reality,

很有可能會讓人們理解感覺，理解現實

and perhaps to even also understand physical reality.

甚至理解促進對物理現實的理解

So, what we did was, for the past 15 years,

在過去的15年中我們所做的是

was to dissect out the neocortex, systematically.

系統地分解大腦新皮質

It's a bit like going and cataloging a piece of the rainforest.

這過程有點類似對一片熱帶雨林里的樹木進行分類

How many trees does it have?

一共有多少樹木？

What shapes are the trees?

它們都有些甚麼形狀？

How many of each type of tree do you have? Where are they positioned?

每一種的樹有多少？它們分布在何處？

But it's a bit more than cataloging because you actually have to

但又不只是分類，因為我們還需要

describe and discover all the rules of communication,

描述和發現它們互相交流的規則

the rules of connectivity,

連接的規則

because the neurons don't just like to connect with any neuron.

因為神經元不僅僅是與任何一個神經元細胞連接起來

They choose very carefully who they connect with.

它們有非常仔細地挑選與哪一個神經原連接

It's also more than cataloging

還有一點不同與分類的是

because you actually have to build three-dimensional

我們必須在三度空間中

digital models of them.

建立它們的數位化模型

And we did that for tens of thousands of neurons,

我們為所發現的所有不同種類的

built digital models of all the different types

神經元構建了

of neurons we came across.

成千上萬的數位模型。

And once you have that, you can actually

一旦我們有了這些模型，就可以

begin to build the neocortical column.

開始建造一個新皮質功能柱

And here we're coiling them up.

我們正將它們纏繞起來

But as you do this, what you see

當我們在這樣做的時候發現

is that the branches intersect

神經元的分支在

actually in millions of locations,

無數的地方互相交叉。

and at each of these intersections

而在每一個交叉點，

they can form a synapse.

他們都會形成一個突觸。

And a synapse is a chemical location

突觸是一個神經元之間利

where they communicate with each other.

用化學媒介互相交流的地方。

And these synapses together

這麼多突觸一起

form the network

形成了網路，

or the circuit of the brain.

或者說是大腦的迴路。

Now, the circuit, you could also think of as

這種迴路也可以

the fabric of the brain.

看成是大腦的纖維。

And when you think of the fabric of the brain,

當我們研究大腦的纖維

the structure, how is it built? What is the pattern of the carpet?

它的結構，不禁要問，它是如何構建的？按照甚麼樣的規律？

You realize that this poses

我們意識到這引出了一個

a fundamental challenge to any theory of the brain,

對任何關於大腦的理論的最根本的挑戰，

and especially to a theory that says

特別是有個理論

that there is some reality that emerges

認為現實是從大腦中

out of this carpet, out of this particular carpet

按照特定的規律

with a particular pattern.

湧現出來的。

The reason is because the most important design secret of the brain

因為大腦的設計中字重要的祕密

is diversity.

是差異化。

Every neuron is different.

每個神經元都是不同的

It's the same in the forest. Every pine tree is different.

這就好像叢林一樣，每棵松樹都是不同的

You may have many different types of trees,

或許有許多不同種類的樹

but every pine tree is different. And in the brain it's the same.

每一棵都是不同的，大腦也是這樣

So there is no neuron in my brain that is the same as another,

所以在我腦中的神經元絕對不會和別人的一樣

and there is no neuron in my brain that is the same as in yours.

也不會和你腦中的一樣

And your neurons are not going to be oriented and positioned

我們的神經元的方向和

in exactly the same way.

位置也不會是一樣的。

And you may have more or less neurons.

可能你的神經元會多一些或者少一些

So it's very unlikely

所以不大可能

that you got the same fabric, the same circuitry.

我們會有相同的纖維，相同的迴路

So, how could we possibly create a reality

所以我們怎麼可能創造出一個

that we can even understand each other?

我們在其中都能互相理解的現實？

Well, we don't have to speculate.

我們不用再繼續猜疑

We can look at all 10 million synapses now.

我們現在可以觀察這1000多萬的突觸

We can look at the fabric. And we can change neurons.

我們可以觀察這纖維，可以改變其中的神經元

We can use different neurons with different variations.

也可以使用各種各樣的神經元

We can position them in different places,

置放它們在不同的地方

orient them in different places.

讓它們朝向不同的方向

We can use less or more of them.

增加或者減少數量

And when we do that

當我們這樣去做

what we discovered is that the circuitry does change.

我們發現儘管大腦的迴路被改變

But the pattern of how the circuitry is designed does not.

但是迴路的模式是注定不變的

So, the fabric of the brain,

所以我們的大腦纖維，

even though your brain may be smaller, bigger,

可能有小有大

it may have different types of neurons,

可能有不同種類的神經元

different morphologies of neurons,

或者不同形狀的神經元

we actually do share

我們確實擁有着

the same fabric.

同樣的纖維。

And we think this is species-specific,

我們認為這是物種特有的。

which means that that could explain

這可能就解釋了為甚麼我們

why we can't communicate across species.

不能和其他物種交流溝通

So, let's switch it on. But to do it, what you have to do

讓我們開始行動。不過要進行這個計畫，我們需要做的是

is you have to make this come alive.

賦予它生命。

We make it come alive

我們用各種方程式來賦予它生命，

with equations, a lot of mathematics.

涉及到非常多的公式和算術。

And, in fact, the equations that make neurons into electrical generators

讓神經元產生電流的方程式

were discovered by two Cambridge Nobel Laureates.

是由兩位劍橋大學的諾貝爾獎得主發現的

So, we have the mathematics to make neurons come alive.

我們知道了賦予神經元生命的數學公式

We also have the mathematics to describe

我們也有用來描述神經元

how neurons collect information,

如何收集信息,

and how they create a little lightning bolt

如何使用電信號互相

to communicate with each other.

溝通交流的數學公式。

And when they get to the synapse,

當電流到達突觸的時候

what they do is they effectively,

它們會非常有效地

literally, shock the synapse.

衝擊突觸，

It's like electrical shock

就好像讓突觸釋放出

that releases the chemicals from these synapses.

化學物質的電擊一樣。

And we've got the mathematics to describe this process.

我們擁有描述這個過程的數學公式。

So we can describe the communication between the neurons.

它們可以描述神經元之間互相通信。

There literally are only a handful

實際上激活大腦新皮質

of equations that you need to simulate

互相交流只需要

the activity of the neocortex.

少量的公式就可以了。

But what you do need is a very big computer.

你所需要的是一台巨大的電腦。

And in fact you need one laptop

每一個神經元就需要

to do all the calculations just for one neuron.

一台筆記型電腦來運算。

So you need 10,000 laptops.

所以我們需要10000台筆記型電腦。

So where do you go? You go to IBM,

去哪裡找這麼多電腦？我們找到IBM,

and you get a supercomputer, because they know how to take

在那裡我們有機會使用超級電腦，因為他們知道

10,000 laptops and put it into the size of a refrigerator.

怎麼把10000台筆記型電腦放進一個冰箱大小的機櫃當中。

So now we have this Blue Gene supercomputer.

有了這台深藍基因超級電腦。

We can load up all the neurons,

我們就可以載入所有的神經元，

each one on to its processor,

每一個神經元分配到一個處理器，

and fire it up, and see what happens.

然後啓動他們

Take the magic carpet for a ride.

來觀察會發生甚麼情況。

Here we activate it. And this gives the first glimpse

這裡是啓動後的情形。這是第一手的資料

of what is happening in your brain

揭露了你的大腦接受到

when there is a stimulation.

外界的刺激後會發生甚麼。

It's the first view.

這是第一批影象。

Now, when you look at that the first time, you think,

如果是第一次面對着它，你會覺得：

"My god. How is reality coming out of that?"

“我的天哪，怎麼從這裡面看的出現實？”

But, in fact, you can start,

但實際上，就算我們

even though we haven't trained this neocortical column

還從來沒有教過這些新皮質功能柱

to create a specific reality.

來創造一個專門的現實。

But we can ask, "Where is the rose?"

我們可以問它，“玫瑰在哪裡？”

We can ask, "Where is it inside,

我們很好奇這個現實會在大腦的哪裡湧現，

if we stimulate it with a picture?"

如果我們用一張照片來刺激它。

Where is it inside the neocortex?

它會處在新皮質的裡面的甚麼位置呢？

Ultimately it's got to be there if we stimulated it with it.

如果我們用圖片刺激它，“玫瑰”最終一定會在某個地方出現。

So, the way that we can look at that

我們觀察的方法是

is to ignore the neurons, ignore the synapses,

忽略神經元，忽略突觸，

and look just at the raw electrical activity.

只看最初始的電流活動。

Because that is what it's creating.

因為這是大腦應該產生的，

It's creating electrical patterns.

它產生電流活動。

So when we did this,

當我們這樣做的時候，

we indeed, for the first time,

我們的確第一次確實地看見了

saw these ghost-like structures:

這個虛幻地結構，

electrical objects appearing

電流形成的物體

within the neocortical column.

出現在新皮質功能柱中。

And it's these electrical objects

這些電流形成的物體

that are holding all the information about

承載着所有關於任何外來的刺激

whatever stimulated it.

所形成的信息。

And then when we zoomed into this,

我們深入進這個影像，

it's like a veritable universe.

它就像是一個真正的宇宙。

So the next step

下一步將是

is just to take these brain coordinates

按照大腦中的坐標再把這

and to project them into perceptual space.

產生的現實投射到感知空間。

And if you do that,

如果這樣做，

you will be able to step inside

我們就會步入

the reality that is created

由這個機器，

by this machine,

由這個部份大腦

by this piece of the brain.

所產生的現實當中。

So, in summary,

總的來講，

I think that the universe may have --

我認為宇宙進化出了

it's possible --

一個大腦

evolved a brain to see itself,

來觀察自己可能是

which may be a first step in becoming aware of itself.

產生自我意識的第一步。

There is a lot more to do to test these theories,

要驗證這些理論還有很多工作要做，

and to test any other theories.

還有測試其他的理論。

But I hope that you are at least partly convinced

我希望至少可以說服大家

that it is not impossible to build a brain.

創造一個大腦不是天方夜譚。

We can do it within 10 years,

我們在10年內就可以做到，

and if we do succeed,

如果成功了，

we will send to TED, in 10 years,

十年內，我們就會送一個全息圖像

a hologram to talk to you. Thank you.

到TED來跟大家交流。謝謝。

(Applause)

（掌聲）