Placeholder Image

字幕列表 影片播放

  • (playful music)

  • This is Geoff Hinton.

  • Because of a back condition, he hasn't been able

  • to sit down for more than 12 years.

  • I hate standing, I would much rather sit down,

  • but if I sit down I have a disc that comes out.

  • So. Okay.

  • Well, at least now standing desks are fashionable and--

  • Yeah, but I was ahead.

  • (laughs)

  • I was standing when they weren't fashionable.

  • Since he can't sit in a car or on a bus,

  • Hinton walks everywhere.

  • (playful music)

  • The walk says a lot about Hinton and his resolve.

  • For nearly 40 years, Hinton has been trying

  • to get computers to learn like people do.

  • A quest almost everyone thought was crazy,

  • or at least hopeless.

  • Right up until the moment it revolutionized the field.

  • Google thinks this is the future of the company.

  • Amazon thinks this is the future of the company.

  • Apple thinks it's future of the company.

  • My own department thinks this stuff's probably nonsense

  • and we shouldn't be doing any more of it.

  • (laughs)

  • So, I talked everybody into it except my own department.

  • (playful music)

  • You obviously grew up in the UK,

  • and you had this very prestigious family

  • full of famous mathematicians and economists

  • and, I was curious what that was like for you.

  • Yeah, there was a lot of pressure.

  • I think by the time I was about seven,

  • I realized I was gonna have to get a Ph.D.

  • (laughing)

  • Did you rebel against that?

  • Or you went along with it?

  • I dropped out every so often.

  • I became a carpenter for a while.

  • Geoff Hinton pretty early on became obsessed

  • with this idea of figuring out how the mind works.

  • He started off getting into physiology,

  • the anatomy of how the brain works,

  • then he got into psychology, and then finally,

  • he settled on more of a computer science approach

  • to modeling the brain, and got into artificial intelligence.

  • My feeling is, if you want to understand

  • a really complicated device like a brain,

  • you should build one.

  • I mean, you can look at cars,

  • and you could think you could understand cars.

  • When you try to build a car, you suddenly discover

  • then there's this stuff that has to go under the hood,

  • otherwise it doesn't work.

  • Yeah. (laughs)

  • As Geoff was starting to think about these ideas,

  • he got inspired by some AI researchers across the pond.

  • Specifically, this guy: Frank Rosenblatt.

  • Rosenblatt, in the the late 1950s,

  • developed what he called a perceptron,

  • and it was a neural network, a computing system

  • that would mimic the brain.

  • The basic idea is a collection

  • of small units, called neurons.

  • These are little computing units,

  • but they're actually modeled on the way

  • that the human brain does it's computation.

  • They take their incoming data like we do from our senses,

  • and they actually learn, so the neural net

  • can learn to make decisions over time.

  • Rosenblatts's hope was that you could feed

  • a neural network a bunch of data,

  • like pictures of men and women,

  • and it would eventually learn how to tell them apart.

  • Just like humans do.

  • There was just one problem: it didn't work very well.

  • Rosenblatt, his neural network was the single layer

  • of neurons, and it was limited in what it could do.

  • Extremely limited.

  • And a colleague of his wrote a book in the late 60s

  • that showed these limitations.

  • And, it kind of put the whole area of research

  • into a deep freeze for a good 10 years.

  • No one wanted to work in this area.

  • They were sure it would never work.

  • Well, almost no one.

  • It was just obvious to me that everything

  • was about ready to go.

  • The brain's a big neural network,

  • and so, it has to be that stuff like this can work,

  • because it works in our brains.

  • There's just never any doubt about that.

  • And what do you think that it was inside of you

  • that kept you wanting to pursue this

  • when everyone else was giving up?

  • Just, that you thought it was the right direction to go?

  • No, that everyone else was wrong.

  • Okay.

  • (laughs)

  • (upbeat music)

  • Hinton decides he's got an idea

  • of how these neural nets might work,

  • and he's going to pursue it no matter what.

  • For a little while, he's bouncing around

  • research institutions in the US.

  • He kind of gets fed up that most of them

  • were funded by the Defense Department,

  • and he starts looking for somewhere else he can go.

  • I didn't want to take Defense Department money.

  • I sort of didn't like the idea that this stuff

  • was going to be used for purposes

  • that I didn't think were good.

  • He suddenly hears that Canada might be interested

  • in funding artificial intelligence.

  • And that was very attractive,

  • that I could go off to this civilized town,

  • and just get on with it.

  • So I came to the University of Toronto.

  • And then in the mid-80s, we discovered

  • how to make more complicated neural nets

  • so they could solve those problems

  • that the simple ones couldn't solve.

  • He and his collaborators developed

  • a multi-layered neural network, a deep neural network.

  • And this started to work in a lot of ways.

  • Using a neural network, a guy named

  • Dean Pomerleau built a self-driving car in the late 80s.

  • And it drove on public roads.

  • Yann LeCun, in the 90s, built a system

  • that could recognize handwritten digits,

  • and this ended up being used commercially.

  • But again, they hit a ceiling.

  • (upbeat music)

  • It didn't work quite well enough,

  • because we didn't have enough data,

  • we didn't have enough compute power.

  • And people in AI and computer science,

  • decided that neural networks

  • were wishful thinking, basically.

  • So, it was a big disappointment.

  • Through the 90s, into the 2000s,

  • Geoff was one of only a handful of people on the planet

  • who were still pursuing this technology.

  • He would show up at academic conferences

  • and be banished to the back rooms,

  • he was treated as, really like a pariah.

  • Was there like a time when you thought

  • this just wasn't going to work?

  • And you had some self-doubt?

  • I mean there were many times when I thought,

  • "I'm not going to make this work."

  • (laughs)

  • But Geoff was consumed by this and couldn't stop.

  • He just kept pursuing the idea

  • that computers could learn.

  • Until about 2006, when the world catches up

  • to Hinton's ideas.

  • (upbeat music)

  • Computers are now a lot faster.

  • And now, it's behaving like I thought

  • it would behave in the mid-80s.

  • It's solving everything.

  • The arrival of super-fast chips,

  • and the massive amounts of data produced on the internet

  • gave Hinton's algorithms a magical boost.

  • Suddenly, computers could identify what was in an image.

  • Then, they could recognize speech

  • and translate from one language to another.

  • By 2012, words like neural nets and machine learning

  • were popping up on the front page

  • of the New York Times.

  • You have to go all these years,