The wonderful and terrifying implications of computers that can learn | Jeremy Howard

翻译人员: Yuchen Shen 校对人员: Li Li

It used to be that if you wanted to get a computer to do something new,

在过去，如果你想让计算机做一件事

you would have to program it.

你需要设计电脑程序

Now, programming, for those of you here that haven't done it yourself,

你们可能从没做过这件事

requires laying out in excruciating detail

编程需要排列出你想让电脑做的 每一个细枝末节的小步骤来达到你的目的

every single step that you want the computer to do

in order to achieve your goal.

Now, if you want to do something that you don't know how to do yourself,

假如你自己都不清楚完成这某件事的话

then this is going to be a great challenge.

要编写处电脑程序来完成那件事就会显得 比登天还要困难

So this was the challenge faced by this man, Arthur Samuel.

这也是这个人，亚瑟 塞缪尔，所面临的挑战

In 1956, he wanted to get this computer

在1956年，他想让这台电脑和他下国际象棋

to be able to beat him at checkers.

How can you write a program,

你怎样才能罗列出所有的细枝末节， 并且让电脑下象棋比你厉害？

lay out in excruciating detail, how to be better than you at checkers?

So he came up with an idea:

他想出一个办法

he had the computer play against itself thousands of times

它让电脑和自己对战几千次

and learn how to play checkers.

学习如何下象棋

And indeed it worked, and in fact, by 1962,

事实证明他做到了。1962年

this computer had beaten the Connecticut state champion.

这台电脑打败了美国康涅狄克州象棋冠军

So Arthur Samuel was the father of machine learning,

亚瑟 塞缪尔是机器学习之父

and I have a great debt to him,

我非常敬畏他

because I am a machine learning practitioner.

因为我是机器学习的实践者

I was the president of Kaggle,

我曾是Kaggle的主席

a community of over 200,000 machine learning practictioners.

Kaggle是一个拥有200,000机器学习实践者地社区

Kaggle puts up competitions

Kaggle会组织竞赛

to try and get them to solve previously unsolved problems,

让人们尝试解决过去未解决的问题

and it's been successful hundreds of times.

已成功解决问题几百次

So from this vantage point, I was able to find out

在这个有利环境中，我发现了

a lot about what machine learning can do in the past, can do today,

机器学习在过去，现在，和将来可以做些什么

and what it could do in the future.

Perhaps the first big success of machine learning commercially was Google.

第一个机器学习的商业成功案例应该是谷歌

Google showed that it is possible to find information

谷歌用计算机算法寻找信息

by using a computer algorithm,

and this algorithm is based on machine learning.

而且这个算法以计算机学习为基础

Since that time, there have been many commercial successes of machine learning.

从那以后，机器学习得到了很多的商业成功

Companies like Amazon and Netflix

像亚马逊、网飞这类公司

use machine learning to suggest products that you might like to buy,

通过机器学习向你推荐你可能想买的东西

movies that you might like to watch.

你可能想看的电影

Sometimes, it's almost creepy.

有时候你会被吓一跳

Companies like LinkedIn and Facebook

像领英、脸谱这类的公司

sometimes will tell you about who your friends might be

有时会告诉你谁会是你的朋友

and you have no idea how it did it,

你根本不知道他们是如何做到的

and this is because it's using the power of machine learning.

其实他们正是运用了机器学习的力量

These are algorithms that have learned how to do this from data

这种运算方法使用数据

rather than being programmed by hand.

而非手动编写程序

This is also how IBM was successful

这也是IBM的Watson超级计算机 在《危险边缘》里打败两届世界冠军的秘诀

in getting Watson to beat the two world champions at "Jeopardy,"

answering incredibly subtle and complex questions like this one.

成功回答了这样一个极其模糊且复杂的问题

["The ancient 'Lion of Nimrud' went missing from this city's national museum in 2003 (along with a lot of other stuff)"]

［“古代‘尼姆鲁德狮像’于2003年在这个城市的国家博物馆消失（连同其它很多物品）”］

This is also why we are now able to see the first self-driving cars.

这也是为什么我们现在有了第一台自驾车

If you want to be able to tell the difference between, say,

如果你想区分一棵树和一个行人

a tree and a pedestrian, well, that's pretty important.

显然这很重要

We don't know how to write those programs by hand,

但是我们不知道如何写这样一个程序

but with machine learning, this is now possible.

有了机器学习，这就成为了可能

And in fact, this car has driven over a million miles

这台自驾车已经行驶了十万英里

without any accidents on regular roads.

在正常路面上零事故

So we now know that computers can learn,

我们知道电脑能够学习

and computers can learn to do things

学习做一件有时我们自己都不知道怎么做的事情

that we actually sometimes don't know how to do ourselves,

or maybe can do them better than us.

有时甚至比我们做得更好

One of the most amazing examples I've seen of machine learning

我见过机器学习最惊人的例子 是我在Kaggle做的一个项目

happened on a project that I ran at Kaggle

where a team run by a guy called Geoffrey Hinton

一个叫杰弗里 辛顿的人毕业于多伦多大学， 带领一个团队

from the University of Toronto

won a competition for automatic drug discovery.

赢得了一个自动查毒的竞赛

Now, what was extraordinary here is not just that they beat

然而真正精彩的不是他们打败了所有默克公司 或者国际学术团体设计的运算

all of the algorithms developed by Merck or the international academic community,

but nobody on the team had any background in chemistry or biology or life sciences,

而是他们团队里没有一个人有化学、生物 或者生命科学的背景

and they did it in two weeks.

却在两个星期内赢得了比赛

How did they do this?

他们是如何做到的？

They used an extraordinary algorithm called deep learning.

他们应用了一种超凡的算法叫做深度学习

So important was this that in fact the success was covered

几个星期后纽约时报在其首页 报道了此次的重要成功

in The New York Times in a front page article a few weeks later.

This is Geoffrey Hinton here on the left-hand side.

在左手边就是杰弗里 辛顿

Deep learning is an algorithm inspired by how the human brain works,

深度学习是受到人类大脑的启发

and as a result it's an algorithm

也因此这种算法的能力不受任何理论限制

which has no theoretical limitations on what it can do.

The more data you give it and the more computation time you give it,

你给它越多的数据和运算时间

the better it gets.

它会工作的越好

The New York Times also showed in this article

纽约时报在其文章中 还说明了深度学习的另一非凡之处

another extraordinary result of deep learning

which I'm going to show you now.

现在我要展示给你们看

It shows that computers can listen and understand.

它表明电脑能够听懂信息

(Video) Richard Rashid: Now, the last step

（视频）理查德 拉希德：现在， 我要做的最后一步是

that I want to be able to take in this process

is to actually speak to you in Chinese.

用汉语和大家说话

Now the key thing there is,

在这之前，我们已经通过很多说汉语的人 收集了大量信息

we've been able to take a large amount of information from many Chinese speakers

and produce a text-to-speech system

然后形成一个语音合成系统

that takes Chinese text and converts it into Chinese language,

把汉字转换成汉语言

and then we've taken an hour or so of my own voice

之后我们收录了一个小时我的声音

and we've used that to modulate

使声音合成系统的声音听起来像我

the standard text-to-speech system so that it would sound like me.

Again, the result's not perfect.

再次，结果并不完美

There are in fact quite a few errors.

他们会有不少错误

(In Chinese)

（中文）

(Applause)

（掌声）

There's much work to be done in this area.

在这个领域还有很多工作要做

(In Chinese)

（中文）

(Applause)

（掌声）

Jeremy Howard: Well, that was at a machine learning conference in China.

杰里米 霍华德：这是在一个中国的机器学习会议上

It's not often, actually, at academic conferences

事实上，一般来说，你不会在学术会议上 听到如此热烈的掌声

that you do hear spontaneous applause,

although of course sometimes at TEDx conferences, feel free.

当然除了TEDx演讲可以随意鼓掌

Everything you saw there was happening with deep learning.

你所看到的一切都伴随着深入学习

(Applause) Thank you.

（掌声）谢谢

The transcription in English was deep learning.

对英文的转录是深入学习

The translation to Chinese and the text in the top right, deep learning,

翻译成汉语以及屏幕右上方的文字是深入学习

and the construction of the voice was deep learning as well.

声音的合成也是深入学习

So deep learning is this extraordinary thing.

深入学习就是这样神奇的事情

It's a single algorithm that can seem to do almost anything,

这个单一的算法似乎可以做任何事情

and I discovered that a year earlier, it had also learned to see.

而且一年前我发现他甚至有视觉

In this obscure competition from Germany

这个名不见经传的德国竞赛

called the German Traffic Sign Recognition Benchmark,

叫做德国交通标志识别基准

deep learning had learned to recognize traffic signs like this one.

深度学习已学得识别这些交通标识

Not only could it recognize the traffic signs

它不仅能够做的比其它算法好

better than any other algorithm,

the leaderboard actually showed it was better than people,

排行榜显示它比人更厉害

about twice as good as people.

是人的准确率的两倍

So by 2011, we had the first example

到2011年，我们有了第一台视力高于人类的电脑

of computers that can see better than people.

Since that time, a lot has happened.

从此更多的电脑也可以做到

In 2012, Google announced that they had a deep learning algorithm

在2012年，谷歌宣布让一个深度学习的算法看YouTube视频

watch YouTube videos

and crunched the data on 16,000 computers for a month,

收集16，000台电脑上的数据，为期一个月

and the computer independently learned about concepts such as people and cats

之后电脑便能仅通过看视频独立识别人和猫

just by watching the videos.

This is much like the way that humans learn.

这近似于人类学习的过程

Humans don't learn by being told what they see,

人类不需要被告诉他们看到了什么

but by learning for themselves what these things are.

而是在自己认知事物的过程中学习

Also in 2012, Geoffrey Hinton, who we saw earlier,

同样在2012年，杰弗里 辛顿，我们之前看到的人

won the very popular ImageNet competition,

赢了很火的ImageNet比赛

looking to try to figure out from one and a half million images

分辨出150万张图片的内容

what they're pictures of.

As of 2014, we're now down to a six percent error rate

到2014年，我们已经将图像识别的误差 降低到百分之六

in image recognition.

This is better than people, again.

低于人类误差率

So machines really are doing an extraordinarily good job of this,

这项非凡的工作现在已经用于工业

and it is now being used in industry.

For example, Google announced last year

比如说，去年谷歌声明

that they had mapped every single location in France in two hours,

他们在两小时内把法国的每一个地点汇成地图

and the way they did it was that they fed street view images

他们是将街景填入深度学习算法以辨认街道号

into a deep learning algorithm to recognize and read street numbers.

Imagine how long it would have taken before:

可以想象从前这件事要花费多少时间和精力

dozens of people, many years.

This is also happening in China.

同样的事情也发生在中国

Baidu is kind of the Chinese Google, I guess,

百度大概类似于中国的谷歌

and what you see here in the top left

我们看到左上角

is an example of a picture that I uploaded to Baidu's deep learning system,

是一张我上传到百度的深度学习系统的图片

and underneath you can see that the system has understood what that picture is

下面你可以看到系统理解了这张照片

and found similar images.

并且找到了类似的图片

The similar images actually have similar backgrounds,

同样的背景

similar directions of the faces,

同样的角度

even some with their tongue out.

有的甚至也有伸出来的舌头

This is not clearly looking at the text of a web page.

网页上没有准确的文字

All I uploaded was an image.

我只是上传了图片

So we now have computers which really understand what they see

所以说电脑能够真正理解它所看到的事物

and can therefore search databases

进而在数据库的几百万张图片中进行实时搜索

of hundreds of millions of images in real time.

So what does it mean now that computers can see?

就现在而言，电脑的视力意味着什么呢？

Well, it's not just that computers can see.

事实上不仅仅是电脑能够看见

In fact, deep learning has done more than that.

深度学习其实可以做得更多

Complex, nuanced sentences like this one

像这样一个细小复杂的语句

are now understandable with deep learning algorithms.

对深度学习来说是相对易于理解的

As you can see here,

你可以看到

this Stanford-based system showing the red dot at the top

斯坦福基础系统显示上面的红点指出 这个语句表达的是否定语气

has figured out that this sentence is expressing negative sentiment.

Deep learning now in fact is near human performance

深度学习在理解语句内容方面已经接近人类水平

at understanding what sentences are about and what it is saying about those things.

Also, deep learning has been used to read Chinese,

同样，深度学习在用于阅读汉语上已经相当于中国本土人水平

again at about native Chinese speaker level.

This algorithm developed out of Switzerland

这个算法开发于瑞士

by people, none of whom speak or understand any Chinese.

没有一个人懂汉语

As I say, using deep learning

要我说，深度学习是比较于人类 做这件事最好的系统

is about the best system in the world for this,

even compared to native human understanding.

This is a system that we put together at my company

这个系统是在我们公司建立的

which shows putting all this stuff together.

它要把这些东西集合起来

These are pictures which have no text attached,

这些图片没有文字描述

and as I'm typing in here sentences,

随着我在这输入文字

in real time it's understanding these pictures

同时它会了解这些图片

and figuring out what they're about

理解它们是关于什么的

and finding pictures that are similar to the text that I'm writing.

然后找出和这些相似的图片

So you can see, it's actually understanding my sentences

所以你看，他真正在理解我的文字

and actually understanding these pictures.

理解这些图片

I know that you've seen something like this on Google,

我知道你在谷歌上看到过类似的

where you can type in things and it will show you pictures,

你可以输入文字，它会提供给你图片

but actually what it's doing is it's searching the webpage for the text.

但实际上它是在网页上搜索文字

This is very different from actually understanding the images.

这和理解图片是有很大不同的

This is something that computers have only been able to do

理解图片是电脑在过去几个月里才刚刚会做的事情

for the first time in the last few months.

So we can see now that computers can not only see but they can also read,

电脑不仅有视力，而且能够阅读

and, of course, we've shown that they can understand what they hear.

而且当然，电脑也能理解所听到的

Perhaps not surprising now that I'm going to tell you they can write.

也许并不意外，我现在要告诉你们，电脑也可以写

Here is some text that I generated using a deep learning algorithm yesterday.

这是我昨天用深度学习算法写的文字

And here is some text that an algorithm out of Stanford generated.

这些是斯坦福的算法做的

Each of these sentences was generated

每一句话都是深度学习算法对图片进行的描述

by a deep learning algorithm to describe each of those pictures.

This algorithm before has never seen a man in a black shirt playing a guitar.

算法没见过一个穿黑衣服的男人弹吉他

It's seen a man before, it's seen black before,

它见过男人，见过黑色

it's seen a guitar before,

见过吉他

but it has independently generated this novel description of this picture.

它便自己对这个图片作出了这样的描述

We're still not quite at human performance here, but we're close.

我们还做不到完全和人类同等水平， 但我们已经很接近了

In tests, humans prefer the computer-generated caption

统计表明，四分之一的人更喜欢电脑做的图片说明

one out of four times.

Now this system is now only two weeks old,

目前这个系统刚被开发两周之久

so probably within the next year,

所以按这个速度，估计明年

the computer algorithm will be well past human performance

电脑算法会超过人类水平

at the rate things are going.

So computers can also write.

电脑会写

So we put all this together and it leads to very exciting opportunities.

我们把这些都放在一起，会发现一个令人兴奋的机遇

For example, in medicine,

比如说，在医药业

a team in Boston announced that they had discovered

一个波士顿团队宣布

dozens of new clinically relevant features

他们发现了肿瘤的几十种临床表现

of tumors which help doctors make a prognosis of a cancer.

帮助医生预测癌症

Very similarly, in Stanford,

同样的，在斯坦福

a group there announced that, looking at tissues under magnification,

一个团队宣布通过用放大镜观察组织

they've developed a machine learning-based system

开发了一个基于机器学习的系统

which in fact is better than human pathologists

可以比病理学家更有效地预测癌症患者的幸存率

at predicting survival rates for cancer sufferers.

In both of these cases, not only were the predictions more accurate,

在这两个例子中，不仅预测更加准确

but they generated new insightful science.

而且他们创造了新的科学视角

In the radiology case,

在放射学中

they were new clinical indicators that humans can understand.

新视角是人类可以明白的新临床表现

In this pathology case,

在病理学中

the computer system actually discovered that the cells around the cancer

电脑发现癌细胞周围的细胞

are as important as the cancer cells themselves

在诊断中同癌细胞一样重要

in making a diagnosis.

This is the opposite of what pathologists had been taught for decades.

这和病理学家几十年来的教学是相反的

In each of those two cases, they were systems developed

这两个案例中的系统都是由

by a combination of medical experts and machine learning experts,

医学专家和机器学习专家共同开发的

but as of last year, we're now beyond that too.

去年我们就已经超过了这个水平

This is an example of identifying cancerous areas

这个是用显微镜识别组织癌变区的例子

of human tissue under a microscope.

The system being shown here can identify those areas more accurately,

所显示的这个系统能够与病理学专家同样准确地识别癌变区

or about as accurately, as human pathologists,

甚至比病理专家更准确

but was built entirely with deep learning using no medical expertise

但是建立系统的都是深度学习的专家

by people who have no background in the field.

没有一个医学专家

Similarly, here, this neuron segmentation.

类似的，这是神经细胞分裂

We can now segment neurons about as accurately as humans can,

我们已经可以和人类一样准确地分裂细胞

but this system was developed with deep learning

但这是个深度学习系统

using people with no previous background in medicine.

没有一个开发者拥有医学背景

So myself, as somebody with no previous background in medicine,

对于我这个完全没有医学背景的人来说

I seem to be entirely well qualified to start a new medical company,

看起来我也完全可以开一个医药公司

which I did.

我确实这么做了

I was kind of terrified of doing it,

我开始有点不知所措

but the theory seemed to suggest that it ought to be possible

但理论上说这件事是可行的

to do very useful medicine using just these data analytic techniques.

用这些数据分析技术制作医药

And thankfully, the feedback has been fantastic,

所幸的是，反响非常好

not just from the media but from the medical community,

不仅是媒体的，包括医药行业

who have been very supportive.

都很支持

The theory is that we can take the middle part of the medical process

理论表明我们可以将制药的中间过程

and turn that into data analysis as much as possible,

充分转换成数据分析

leaving doctors to do what they're best at.

让医生去做他们最擅长的

I want to give you an example.

我有一个例子

It now takes us about 15 minutes to generate a new medical diagnostic test

制作一个医学诊断测试需要十五分钟

and I'll show you that in real time now,

我会给你们实际展示

but I've compressed it down to three minutes by cutting some pieces out.

但是我去掉了一部分，把它压缩到了三分钟

Rather than showing you creating a medical diagnostic test,

不要医学诊断试验

I'm going to show you a diagnostic test of car images,

我要给你们展示制作一个汽车图片的诊断测试

because that's something we can all understand.

因为这个我们都能懂

So here we're starting with about 1.5 million car images,

现在我们有150万张汽车图片

and I want to create something that can split them into the angle

我想要根据拍照的角度对他们进行分类

of the photo that's being taken.

So these images are entirely unlabeled, so I have to start from scratch.

这些图片完全没有标签，所以我要先对他们进行简单描述

With our deep learning algorithm,

有深度学习算法

it can automatically identify areas of structure in these images.

它可以自动识别图片的结构要素

So the nice thing is that the human and the computer can now work together.

令人高兴的是人和电脑可以合作

So the human, as you can see here,

你可以看到，这个人

is telling the computer about areas of interest

正在告诉电脑什么是感兴趣的要素

which it wants the computer then to try and use to improve its algorithm.

为之后电脑用来完善算法

Now, these deep learning systems actually are in 16,000-dimensional space,

现在，这些深度学习算法处在16，000维空间中

so you can see here the computer rotating this through that space,

所以你看到电脑让他们在这个空间中旋转

trying to find new areas of structure.

尝试找到新的结构要素

And when it does so successfully,

当他成功时

the human who is driving it can then point out the areas that are interesting.

开车的人就可以指出感兴趣的要素

So here, the computer has successfully found areas,

现在电脑成功找出这些要素

for example, angles.

比如，角度

So as we go through this process,

我们在这个过程中

we're gradually telling the computer more and more

逐渐的告诉电脑更多

about the kinds of structures we're looking for.

我们想寻找的结构

You can imagine in a diagnostic test

你可以想象一个诊断测试

this would be a pathologist identifying areas of pathosis, for example,

这就像是病理学家识别病态区域

or a radiologist indicating potentially troublesome nodules.

或者放射学专家找出潜在的问题囊肿

And sometimes it can be difficult for the algorithm.

有时候这对算法来说有些难度

In this case, it got kind of confused.

我们的例子就比较麻烦

The fronts and the backs of the cars are all mixed up.

车的正面和背面全部混淆了

So here we have to be a bit more careful,

所以我们要仔细一些

manually selecting these fronts as opposed to the backs,

人工地选出正面和背面

then telling the computer that this is a type of group

人后告诉电脑这是我们所感兴趣的一类

that we're interested in.

So we do that for a while, we skip over a little bit,

做这件事花了一些时间，所以我们跳过

and then we train the machine learning algorithm

之后我们用这几百个东西训练机器学习算法

based on these couple of hundred things,

and we hope that it's gotten a lot better.

希望他会有很大进步

You can see, it's now started to fade some of these pictures out,

你能看到，它正在消退一些图片

showing us that it already is recognizing how to understand some of these itself.

说明他已经开始可以自己理解这些图片了

We can then use this concept of similar images,

我们可以用相似图片的概念

and using similar images, you can now see,

用相似的图片，你可以看到

the computer at this point is able to entirely find just the fronts of cars.

电脑现在能够只找出正面的车

So at this point, the human can tell the computer,

在这个时候，人可以告诉电脑

okay, yes, you've done a good job of that.

对的，没错，你做的很好

Sometimes, of course, even at this point

当然，有时，即使在这个阶段

it's still difficult to separate out groups.

分组仍然是很困难的

In this case, even after we let the computer try to rotate this for a while,

像我们这里，让电脑在这里旋转了一段时间了

we still find that the left sides and the right sides pictures

我们还是看到左面的和右面的图片有混淆

are all mixed up together.

So we can again give the computer some hints,

所以我们可以再一次给电脑一些提示

and we say, okay, try and find a projection that separates out

我们让它通过深度学习算法尽可能分离出左面和右面的图片

the left sides and the right sides as much as possible

using this deep learning algorithm.

And giving it that hint -- ah, okay, it's been successful.

有了这个指示——好的，它已经完成了

It's managed to find a way of thinking about these objects

它要想办法分开这一部分

that's separated out these together.

So you get the idea here.

你现在知道了

This is a case not where the human is being replaced by a computer,

这不是电脑取代人类

but where they're working together.

而是一起合作

What we're doing here is we're replacing something that used to take a team

我们在做的是将过去需要五六人的团队 用七年时间做的事情

of five or six people about seven years

and replacing it with something that takes 15 minutes

变成只需一个人花十五分钟就能完成

for one person acting alone.

So this process takes about four or five iterations.

这个过程需要四到五次反复

You can see we now have 62 percent

你可以看到我们已经将150万张图片的62%正确分类

of our 1.5 million images classified correctly.

And at this point, we can start to quite quickly

现在我们就可以快速地检查整个分组

grab whole big sections,

check through them to make sure that there's no mistakes.

确保没有错误

Where there are mistakes, we can let the computer know about them.

如果哪里有错误，我们可以告诉电脑

And using this kind of process for each of the different groups,

每个分组我们都这样做

we are now up to an 80 percent success rate

现在这150万张图片已经达到80%的成功率

in classifying the 1.5 million images.

And at this point, it's just a case

现在这个阶段

of finding the small number that aren't classified correctly,

只需要找出几个不正确的分类

and trying to understand why.

并让电脑明白为什么

And using that approach,

到了这个步骤

by 15 minutes we get to 97 percent classification rates.

十五分钟后我们达到了97%的正确率

So this kind of technique could allow us to fix a major problem,

这种技术能帮助我们解决一个问题

which is that there's a lack of medical expertise in the world.

医疗专家不足的问题

The World Economic Forum says that there's between a 10x and a 20x

世界经济论坛表明，在发展中国家， 内科医生有十倍到二十倍的短缺

shortage of physicians in the developing world,

and it would take about 300 years

而弥补这一短缺需要300年的时间

to train enough people to fix that problem.

So imagine if we can help enhance their efficiency

所以想象一下，是否我们能够用深度学习的方法 帮助他们提高效率？

using these deep learning approaches?

So I'm very excited about the opportunities.

我对这个机会表示很激动

I'm also concerned about the problems.

我同样的担心一些问题

The problem here is that every area in blue on this map

问题是在这张地图上的蓝色区域内

is somewhere where services are over 80 percent of employment.

服务占就业的80%以上

What are services?

什么是服务？

These are services.

这些是服务

These are also the exact things that computers have just learned how to do.

这些也是电脑才刚刚开始学习的事情

So 80 percent of the world's employment in the developed world

也就是说世界上发达国家的80%的就业

is stuff that computers have just learned how to do.

是电脑刚开始学习的

What does that mean?

这是什么意思？

Well, it'll be fine. They'll be replaced by other jobs.

其实也没什么大不了的，他们会被其他职业替代

For example, there will be more jobs for data scientists.

比如说会有更多的数据学家

Well, not really.

也不尽然

It doesn't take data scientists very long to build these things.

数据学家不需要太久的时间做这些事

For example, these four algorithms were all built by the same guy.

比如这四个算法都是同时一个人开发的

So if you think, oh, it's all happened before,

如果你认为这些曾经都发生过

we've seen the results in the past of when new things come along

我们看到过新的事物出现

and they get replaced by new jobs,

然后被新的职业所取代

what are these new jobs going to be?

那这些新的职业又会是什么？

It's very hard for us to estimate this,

很难做出估计

because human performance grows at this gradual rate,

因为人的能力以这个均匀的速度增长

but we now have a system, deep learning,

但是现在我们有了深度学习系统

that we know actually grows in capability exponentially.

它的能力以指数方式增长

And we're here.

我们现在在这

So currently, we see the things around us

目前，我们看周围的事物

and we say, "Oh, computers are still pretty dumb." Right?

会说：“电脑还是很笨。”对吧？

But in five years' time, computers will be off this chart.

但是在五年内，电脑会超出这张图

So we need to be starting to think about this capability right now.

所以我们现在要开始考虑这样的能力了

We have seen this once before, of course.

当然，我们曾经见过这个

In the Industrial Revolution,

在工业革命时期

we saw a step change in capability thanks to engines.

发动机让生产力迈进一大步

The thing is, though, that after a while, things flattened out.

然而问题是，一段时间之后，形势转平了

There was social disruption,

是由于社会的破坏

but once engines were used to generate power in all the situations,

但当发动机被普遍应用时

things really settled down.

一切都稳定下来了

The Machine Learning Revolution

机器学习革命

is going to be very different from the Industrial Revolution,

将和工业革命有很大不同

because the Machine Learning Revolution, it never settles down.

因为机器学习革命不会停止

The better computers get at intellectual activities,

电脑越擅长智能活动

the more they can build better computers to be better at intellectual capabilities,

它们越能制造出更加擅长智能活动的电脑

so this is going to be a kind of change

这将会是世界从未经历过的改变

that the world has actually never experienced before,

so your previous understanding of what's possible is different.

所以你之前理解的可能性是不一样的

This is already impacting us.

这正在影响我们的生活

In the last 25 years, as capital productivity has increased,

在过去的25年里，随着资本生产力的增加

labor productivity has been flat, in fact even a little bit down.

劳动生产力在变缓，甚至下降

So I want us to start having this discussion now.

所以我希望可以发起大家的讨论

I know that when I often tell people about this situation,

我知道当我和人们讲述这样的处境时

people can be quite dismissive.

人们往往表现出不以为然

Well, computers can't really think,

电脑不会思考

they don't emote, they don't understand poetry,

它们没有情感，也不懂诗

we don't really understand how they work.

它们甚至都不知道自己是如何运作的

So what?

那又怎样？

Computers right now can do the things

电脑现在可以做

that humans spend most of their time being paid to do,

人类用大部分有偿的劳动时间做的事情

so now's the time to start thinking

所以现在该到我们思考

about how we're going to adjust our social structures and economic structures

我们将如何调整我们的社会结构和经济结构

to be aware of this new reality.

来应对新形势

Thank you.

谢谢

(Applause)

（鼓掌）