Placeholder Image

字幕列表 影片播放

  • I'm a neuroscientist.

    我是一個神經學家。

  • And in neuroscience,

    在神經學中,

  • we have to deal with many difficult questions about the brain.

    我們必須處理許多關於大腦的艱深問題。

  • But I want to start with the easiest question

    但是我想從最簡單的問題開始談起,

  • and the question you really should have all asked yourselves at some point in your life,

    而每個人在人生中都該問過自己這個問題,

  • because it's a fundamental question

    因為想了解大腦的運作,

  • if we want to understand brain function.

    是最根本的問題。

  • And that is, why do we and other animals

    這問題就是,為什麼我們和其他動物

  • have brains?

    會有大腦呢?

  • Not all species on our planet have brains,

    並非地球上所有的生物都有大腦,

  • so if we want to know what the brain is for,

    所以如果我們想知道大腦的作用,

  • let's think about why we evolved one.

    就得想想我們為何會進化出一個大腦。

  • Now you may reason that we have one

    你現在可能認為大腦的存在

  • to perceive the world or to think,

    是為了感覺這個世界或是思考,

  • and that's completely wrong.

    這是完全錯誤的。

  • If you think about this question for any length of time,

    如果你花了很多時間去思考這個問題,

  • it's blindingly obvious why we have a brain.

    這將會使你誤判為什麼我們會有大腦。

  • We have a brain for one reason and one reason only,

    我們擁有大腦的唯一原因,

  • and that's to produce adaptable and complex movements.

    是為了產生適合且複雜的動作。

  • There is no other reason to have a brain.

    這是大腦存在的唯一理由。

  • Think about it.

    想想看。

  • Movement is the only way you have

    動作是你感覺這個世界

  • of affecting the world around you.

    唯一的反應方式。

  • Now that's not quite true. There's one other way, and that's through sweating.

    這並非完全正確。還有一個方式,就是透過冒汗。

  • But apart from that,

    但是除了那個以外,

  • everything else goes through contractions of muscles.

    每件事都必須藉由肌肉的收縮。

  • So think about communication --

    所以, 看看溝通 --

  • speech, gestures, writing, sign language --

    說話、姿勢、寫字、手語 --

  • they're all mediated through contractions of your muscles.

    這些都藉由收縮你的肌肉來達成。

  • So it's really important to remember

    記住事情是很重要的,

  • that sensory, memory and cognitive processes are all important,

    感覺、記憶和理解程序都很重要,

  • but they're only important

    但這些是為了能夠做到

  • to either drive or suppress future movements.

    開始或結束後續的動作, 所以才會重要。

  • There can be no evolutionary advantage

    小時候的回憶累積、

  • to laying down memories of childhood

    或是對於玫瑰顏色的認知,

  • or perceiving the color of a rose

    對於進化並沒有什麼影響,

  • if it doesn't affect the way you're going to move later in life.

    如果它對你未來生活的行為沒有幫助。

  • Now for those who don't believe this argument,

    對於那些不相信這種說法的人們,

  • we have trees and grass on our planet without the brain,

    雖然地球上的樹和草都沒有大腦,

  • but the clinching evidence is this animal here --

    但這動物就是確切的證據 --

  • the humble sea squirt.

    這渺小的海鞘。

  • Rudimentary animal, has a nervous system,

    這種未進化的動物,擁有神經系統,

  • swims around in the ocean in its juvenile life.

    幼年時會在海洋中游盪著。

  • And at some point of its life,

    等到長大之後,

  • it implants on a rock.

    它便會攀附在岩石上。

  • And the first thing it does in implanting on that rock, which it never leaves,

    當它攀附上這永遠居住的岩石之後,

  • is to digest its own brain and nervous system

    它所作的第一件事,

  • for food.

    就是將它的大腦和神經系統當食物吃掉。

  • So once you don't need to move,

    所以一旦你不再需要移動,

  • you don't need the luxury of that brain.

    你就不需要大腦這種奢侈品了。

  • And this animal is often taken

    這種動物常被拿來

  • as an analogy to what happens at universities

    當作一種比喻,當大學教授

  • when professors get tenure,

    獲得終身職位之後會發生的事情,

  • but that's a different subject.

    不過那是另外一個話題了。

  • (Applause)

    (掌聲)

  • So I am a movement chauvinist.

    我是一個活動主義者。

  • I believe movement is the most important function of the brain --

    我認為大腦最重要的功能就是控制動作,

  • don't let anyone tell you that it's not true.

    別讓任何人告訴你這不是真的。

  • Now if movement is so important,

    如果動作那麼重要,

  • how well are we doing

    我們對於了解大腦如何控制動作

  • understanding how the brain controls movement?

    這方面的研究進展如何呢?

  • And the answer is we're doing extremely poorly; it's a very hard problem.

    答案是,少得可憐;這是很艱深的難題。

  • But we can look at how well we're doing

    但我們可以換個方向來思考,

  • by thinking about how well we're doing building machines

    看看我們對於建造能做出和人類一樣動作的機器

  • which can do what humans can do.

    這種研究進展如何。

  • Think about the game of chess.

    想想看西洋棋這種遊戲。

  • How well are we doing determining what piece to move where?

    我們決定該將哪個棋子移到哪個位置這種研究做得怎麼樣?

  • If you pit Garry Kasparov here, when he's not in jail,

    如果你在 Gary Kasparov 還沒去坐牢之前,

  • against IBM's Deep Blue,

    讓他跟 IBM 的深藍電腦進行比賽,

  • well the answer is IBM's Deep Blue will occasionally win.

    IBM 的深藍電腦有時候可以獲勝。

  • And I think if IBM's Deep Blue played anyone in this room, it would win every time.

    我想如果讓 IBM 的深藍電腦跟在座任何一位下棋,它每次都會獲勝。

  • That problem is solved.

    這個問題就被解決了。

  • What about the problem

    但如果這個問題是

  • of picking up a chess piece,

    拿起一個棋子,

  • dexterously manipulating it and putting it back down on the board?

    靈巧地拿起它,再放回棋盤上呢?

  • If you put a five year-old child's dexterity against the best robots of today,

    如果你讓一個五歲的小孩跟現今最棒的機器人進行比賽,

  • the answer is simple:

    答案很簡單:

  • the child wins easily.

    那個小孩可以輕易獲勝。

  • There's no competition at all.

    機器人完全不是對手。

  • Now why is that top problem so easy

    那麼,為什麼之前的問題很容易做到?

  • and the bottom problem so hard?

    而接著的問題卻很困難呢?

  • One reason is a very smart five year-old

    原因之一是,一個很聰明的五歲小孩

  • could tell you the algorithm for that top problem --

    可以告訴你上面問題的演算法則 --

  • look at all possible moves to the end of the game

    找出直到遊戲結束的所有可能移動步法,

  • and choose the one that makes you win.

    然後選擇可以讓你獲勝的步法。

  • So it's a very simple algorithm.

    所以這是很簡單的演算法則。

  • Now of course there are other moves,

    當然有其他的步法,

  • but with vast computers we approximate

    但是利用龐大的電腦系統,我們估算

  • and come close to the optimal solution.

    並且找出最佳解答。

  • When it comes to being dexterous,

    當討論到靈巧時,

  • it's not even clear what the algorithm is you have to solve to be dexterous.

    甚至沒有明確的演算法則告訴你什麼叫做靈巧。

  • And we'll see you have to both perceive and act on the world,

    於是你必須感覺同時去做出反應,

  • which has a lot of problems.

    這就會遇到很多問題。

  • But let me show you cutting-edge robotics.

    讓我介紹一些先進的機器人。

  • Now a lot of robotics is very impressive,

    現在有許多優秀的機器人,

  • but manipulation robotics is really just in the dark ages.

    但是操控機器人仍舊處於黑暗的時代。

  • So this is the end of a Ph.D. project

    這是在某一個很棒的機器人學院中,

  • from one of the best robotics institutes.

    一個博士研究項目的成果。

  • And the student has trained this robot

    這位學生訓練這個機器人

  • to pour this water into a glass.

    將水倒進杯子裡面。

  • It's a hard problem because the water sloshes about, but it can do it.

    這是很困難的題目,因為水會濺出來,但是它可以辦到。

  • But it doesn't do it with anything like the agility of a human.

    但是它無法像人類做得那麼靈巧。

  • Now if you want this robot to do a different task,

    如果你希望這個機器人進行另一項任務,

  • that's another three-year Ph.D. program.

    那將是另一個三年期的博士研究計畫。

  • There is no generalization at all

    在機器人工程學裡,

  • from one task to another in robotics.

    一項任務和另一項任務是沒有共通性的。

  • Now we can compare this

    我們可以將這個

  • to cutting-edge human performance.

    和人類優異的表現做比較。

  • So what I'm going to show you is Emily Fox

    我要給大家看的是 Emily Fox,

  • winning the world record for cup stacking.

    她是贏得堆疊杯子世界冠軍的人。

  • Now the Americans in the audience will know all about cup stacking.

    觀眾席中如果有美國人,應該知道這個堆疊杯子的比賽。

  • It's a high school sport

    這是一項高中常見的運動,

  • where you have 12 cups you have to stack and unstack

    你得把 12 個杯子依據指定的順序

  • against the clock in a prescribed order.

    快速的堆疊再分開。

  • And this is her getting the world record in real time.

    這是她創下世界紀錄的畫面,以正常速度播放。

  • (Laughter)

    (笑聲)

  • (Applause)

    (掌聲)

  • And she's pretty happy.

    她非常開心。

  • We have no idea what is going on inside her brain when she does that,

    我們不知道當她做這件事情時,腦子裡發生了什麼事情,

  • and that's what we'd like to know.

    那是我們很想知道。

  • So in my group, what we try to do

    所以我的團隊,我們想要做的是

  • is reverse engineer how humans control movement.

    針對人類如何控制動作這件事去進行逆向工程。

  • And it sounds like an easy problem.

    這聽起來是很簡單的問題。

  • You send a command down, it causes muscles to contract.

    你送出一個指令,這會讓肌肉收縮。

  • Your arm or body moves,

    你的手臂或身體移動,

  • and you get sensory feedback from vision, from skin, from muscles and so on.

    然後你得到來自於視覺、皮膚、肌肉等處的感覺回饋。

  • The trouble is

    問題是,

  • these signals are not the beautiful signals you want them to be.

    這些訊息不如你預期的那樣完美。

  • So one thing that makes controlling movement difficult

    讓控制動作變得困難的其中一個因素是,

  • is, for example, sensory feedback is extremely noisy.

    舉例來說,感覺回饋是充滿雜訊的。

  • Now by noise, I do not mean sound.

    關於雜訊,我指的不是聲音。

  • We use it in the engineering and neuroscience sense

    雜訊一般用在工程學與神經科學的檢測中,

  • meaning a random noise corrupting a signal.

    是指干擾主要訊號的不規律且雜亂的訊號。

  • So the old days before digital radio when you were tuning in your radio

    所以在數位收音機出現之前,當你轉動舊式收音機,

  • and you heard "crrcckkk" on the station you wanted to hear,

    你會在你想聽得電台中聽見「嘎啦嘎啦」的聲音,

  • that was the noise.

    那就是雜訊。

  • But more generally, this noise is something that corrupts the signal.

    講白話一點,雜訊就是干擾訊號的東西。

  • So for example, if you put your hand under a table

    例如,當你將手放在桌子底下,

  • and try to localize it with your other hand,

    試著用另一隻手去找到它的位置,

  • you can be off by several centimeters

    你可能會誤差好幾公分,

  • due to the noise in sensory feedback.

    因為感知回饋中有雜訊。

  • Similarly, when you put motor output on movement output,

    同樣地,當你將動力源的力量變成動作的力量時,

  • it's extremely noisy.

    訊號將是非常雜亂的。

  • Forget about trying to hit the bull's eye in darts,

    先不談射飛鏢時能射中靶心,

  • just aim for the same spot over and over again.

    只要試著去重複瞄準同一個點看看。

  • You have a huge spread due to movement variability.

    因為動作的差異性,你會丟到許多不同的點上去。

  • And more than that, the outside world, or task,

    更別提在外在世界,或是執行任務時,

  • is both ambiguous and variable.

    充滿著不確定性和變異性。

  • The teapot could be full, it could be empty.

    茶壺可能是滿的,也可能是空的。

  • It changes over time.

    每次都不一樣。

  • So we work in a whole sensory movement task soup of noise.

    所以我們是在充滿雜訊的環境中進行動作。

  • Now this noise is so great

    因為這個雜訊非常巨大,

  • that society places a huge premium

    所以我們的社會給予那些

  • on those of us who can reduce the consequences of noise.

    能夠抵抗雜訊的人鉅額獎賞。

  • So if you're lucky enough to be able to knock a small white ball

    所以如果你能將一顆小白球

  • into a hole several hundred yards away using a long metal stick,

    用一根金屬長棍打進幾百碼外的洞裡,

  • our society will be willing to reward you

    人們願意給你

  • with hundreds of millions of dollars.

    好幾億的獎金。

  • Now what I want to convince you of

    而我想要讓你知道的是

  • is the brain also goes through a lot of effort

    大腦做了許多的努力

  • to reduce the negative consequences

    去減少這些雜訊以及變異性

  • of this sort of noise and variability.

    所造成的負面效應。

  • And to do that, I'm going to tell you about a framework

    為此,我將會介紹一個

  • which is very popular in statistics and machine learning of the last 50 years

    在過去五十年間,常被用在統計與機械學習方面的架構,

  • called Bayesian decision theory.

    叫做貝葉斯決策理論。

  • And it's more recently a unifying way

    近來它已經逐漸變成用來解釋

  • to think about how the brain deals with uncertainty.

    大腦如何處理不確定性的主要方法。

  • And the fundamental idea is you want to make inferences and then take actions.

    它的基本概念是,你先做出假設,然後去行動。

  • So let's think about the inference.

    我們先來看看假設。

  • You want to generate beliefs about the world.

    你要產生出對事物的信念。

  • So what are beliefs?

    什麼是信念呢?

  • Beliefs could be: where are my arms in space?

    信念可以是:我的手臂在空間中的哪個位置?

  • Am I looking at a cat or a fox?

    我看見的是一隻貓還是一隻狐狸?

  • But we're going to represent beliefs with probabilities.

    而我們必須用可能性來表示信念。

  • So we're going to represent a belief

    我們要將信念表達為

  • with a number between zero and one --

    介於 0 到 1 之間的數字 --

  • zero meaning I don't believe it at all, one means I'm absolutely certain.

    0 代表我完全不相信,1 則表示我絕對相信。

  • And numbers in between give you the gray levels of uncertainty.

    而介於期間的數字則是代表不確定性的灰色地帶。

  • And the key idea to Bayesian inference

    貝葉斯假設的關鍵在於

  • is you have two sources of information

    你有兩種不同的資訊來源

  • from which to make your inference.

    用來建立起你的假設。

  • You have data,

    你會有資訊,

  • and data in neuroscience is sensory input.

    在神經科學中,這資訊就是你的感覺。

  • So I have sensory input, which I can take in to make beliefs.

    我有感覺,所以我可以將它用來建立信念。

  • But there's another source of information, and that's effectively prior knowledge.

    但還有另一種資訊的來源,就是已經擁有的知識。

  • You accumulate knowledge throughout your life in memories.

    藉由生命中的回憶,知識會被累積下來。

  • And the point about Bayesian decision theory

    而貝葉斯決策理論的重點在於

  • is it gives you the mathematics

    它提供你一種

  • of the optimal way to combine

    數學的最佳化方式

  • your prior knowledge with your sensory evidence

    來合併你原有的知識和你的感覺

  • to generate new beliefs.

    以產生出新的信念。

  • And I've put the formula up there.

    它的公式在這裡。

  • I'm not going to explain what that formula is, but it's very beautiful.

    我不會解釋公式是什麼,但是它很漂亮。

  • And it has real beauty and real explanatory power.

    它擁有真實的美感,和真實的說服力。

  • And what it really says, and what you want to estimate,

    它真正表達的,以及你想要估計出的,

  • is the probability of different beliefs

    是由你的感覺所產生出

  • given your sensory input.

    不同信念的可能性。

  • So let me give you an intuitive example.

    我舉一個很直接的例子。

  • Imagine you're learning to play tennis

    想像你正在學習打網球,

  • and you want to decide where the ball is going to bounce

    當球飛過網子朝你過來時,

  • as it comes over the net towards you.

    你要決定球會掉在哪個位置。

  • There are two sources of information

    依據貝葉斯的理論,

  • Bayes' rule tells you.

    你有兩個資訊來源。

  • There's sensory evidence -- you can use visual information auditory information,

    一個是感覺證據 -- 你可以藉由視覺和聽覺的資訊,

  • and that might tell you it's going to land in that red spot.

    那可能會讓你判斷在紅點處。

  • But you know that your senses are not perfect,

    而你知道你的感覺並不完美,

  • and therefore there's some variability of where it's going to land

    所以它的落點會有誤差,

  • shown by that cloud of red,

    這就是紅色區域,

  • representing numbers between 0.5 and maybe 0.1.

    而可能性大概是在 0.5 到 0.1 之間。

  • That information is available in the current shot,

    這資訊來自於這一次的發球,

  • but there's another source of information

    還有另外的資訊

  • not available on the current shot,

    並非由這次發球而來,

  • but only available by repeated experience in the game of tennis,

    而是來自於反覆進行網球比賽的經驗,

  • and that's that the ball doesn't bounce

    經驗告訴你,在這場比賽中,

  • with equal probability over the court during the match.

    球落在球場上每個位置的可能性並不相等。

  • If you're playing against a very good opponent,

    如果你的對手技術很棒,

  • they may distribute it in that green area,

    他們會讓球落在綠色區域,

  • which is the prior distribution,

    就是所謂的先驗分布,

  • making it hard for you to return.

    這會讓你難以回擊。

  • Now both these sources of information carry important information.

    這些訊息來源都帶有重要的訊息。

  • And what Bayes' rule says

    依據貝葉斯理論所說,

  • is that I should multiply the numbers on the red by the numbers on the green

    我應該將紅色區域的機率和綠色區域的機率相乘,

  • to get the numbers of the yellow, which have the ellipses,

    就會得到橢圓形黃色區域的機率,

  • and that's my belief.

    而這就是我的信念。

  • So it's the optimal way of combining information.

    這是合併訊息的最佳方式。

  • Now I wouldn't tell you all this if it wasn't that a few years ago,

    幾年前我們的研究發現,

  • we showed this is exactly what people do

    人們在學習新的動作技巧時,

  • when they learn new movement skills.

    確實有同樣的現象。

  • And what it means

    也就是說,

  • is we really are Bayesian inference machines.

    我們就像是使用貝葉斯假設的機器。

  • As we go around, we learn about statistics of the world and lay that down,

    在生活中,我們學習並累積了關於世界的許多統計資料,

  • but we also learn

    但我們也學習了

  • about how noisy our own sensory apparatus is,

    我們自身感知器官產生的雜訊有多少,

  • and then combine those

    然後將這些合併在一起,

  • in a real Bayesian way.

    這些正是貝葉斯法則。

  • Now a key