ABOUT THE SPEAKER
Jeremy Howard - Data scientist
Jeremy Howard imagines how advanced machine learning can improve our lives.

Why you should listen

Jeremy Howard is the CEO of Enlitic, an advanced machine learning company in San Francisco. Previously, he was the president and chief scientist at Kaggle, a community and competition platform of over 200,000 data scientists. Howard is a faculty member at Singularity University, where he teaches data science. He is also a Young Global Leader with the World Economic Forum, and spoke at the World Economic Forum Annual Meeting 2014 on "Jobs for the Machines."

Howard advised Khosla Ventures as their Data Strategist, identifying opportunities for investing in data-driven startups and mentoring portfolio companies to build data-driven businesses. He was the founding CEO of two successful Australian startups, FastMail and Optimal Decisions Group.

More profile about the speaker
Jeremy Howard | Speaker | TED.com
TEDxBrussels

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

杰里米 霍华德: 会学习的电脑带来的美好和恐怖

Filmed:
2,532,971 views

如果我们教电脑学习会发生什么?技术专家杰里米 霍华德分享了一些快速发展的深度学习领域里令人惊奇的新发展,让计算机能够学习汉语,识别物体和图片,或者帮助医疗诊断。(一个深度学习工具在看了一个小时YouTube视频后,教会自己“猫”这个概念。)深度学习会改变你身边的电脑得行为...远比你想象的要快。
- Data scientist
Jeremy Howard imagines how advanced machine learning can improve our lives. Full bio

Double-click the English transcript below to play the video.

00:12
It used to be that if you wanted
to get a computer电脑 to do something new,
0
880
4013
在过去,如果你想让计算机做一件事
00:16
you would have to program程序 it.
1
4893
1554
你需要设计电脑程序
00:18
Now, programming程序设计, for those of you here
that haven't没有 doneDONE it yourself你自己,
2
6447
3411
你们可能从没做过这件事
00:21
requires要求 laying铺设 out in excruciating痛苦 detail详情
3
9858
3502
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
00:25
every一切 single step that you want
the computer电脑 to do
4
13360
3367
00:28
in order订购 to achieve实现 your goal目标.
5
16727
2362
00:31
Now, if you want to do something
that you don't know how to do yourself你自己,
6
19089
3496
假如你自己都不清楚完成这某件事的话
00:34
then this is going
to be a great challenge挑战.
7
22585
2063
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
00:36
So this was the challenge挑战 faced面对
by this man, Arthur亚瑟 Samuel塞缪尔.
8
24648
3483
这也是这个人,亚瑟 塞缪尔,所面临的挑战
00:40
In 1956, he wanted to get this computer电脑
9
28131
4077
在1956年,他想让这台电脑和他下国际象棋
00:44
to be able能够 to beat击败 him at checkers跳棋.
10
32208
2340
00:46
How can you write a program程序,
11
34548
2040
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?
00:48
lay铺设 out in excruciating痛苦 detail详情,
how to be better than you at checkers跳棋?
12
36588
3806
00:52
So he came来了 up with an idea理念:
13
40394
1722
他想出一个办法
00:54
he had the computer电脑 play
against反对 itself本身 thousands数千 of times
14
42116
3724
它让电脑和自己对战几千次
00:57
and learn学习 how to play checkers跳棋.
15
45840
2524
学习如何下象棋
01:00
And indeed确实 it worked工作,
and in fact事实, by 1962,
16
48364
3180
事实证明他做到了。1962年
01:03
this computer电脑 had beaten殴打
the Connecticut康涅狄格 state champion冠军.
17
51544
4017
这台电脑打败了美国康涅狄克州象棋冠军
01:07
So Arthur亚瑟 Samuel塞缪尔 was
the father父亲 of machine learning学习,
18
55561
2973
亚瑟 塞缪尔是机器学习之父
01:10
and I have a great debt债务 to him,
19
58534
1717
我非常敬畏他
01:12
because I am a machine
learning学习 practitioner从业者.
20
60251
2763
因为我是机器学习的实践者
01:15
I was the president主席 of KaggleKaggle,
21
63014
1465
我曾是Kaggle的主席
01:16
a community社区 of over 200,000
machine learning学习 practictionerspractictioners.
22
64479
3388
Kaggle是一个拥有200,000机器学习实践者地社区
01:19
KaggleKaggle puts看跌期权 up competitions比赛
23
67867
2058
Kaggle会组织竞赛
01:21
to try and get them to solve解决
previously先前 unsolved未解 problems问题,
24
69925
3708
让人们尝试解决过去未解决的问题
01:25
and it's been successful成功
hundreds数以百计 of times.
25
73633
3837
已成功解决问题几百次
01:29
So from this vantage华帝 point,
I was able能够 to find out
26
77470
2470
在这个有利环境中,我发现了
01:31
a lot about what machine learning学习
can do in the past过去, can do today今天,
27
79940
3950
机器学习在过去,现在,和将来可以做些什么
01:35
and what it could do in the future未来.
28
83890
2362
01:38
Perhaps也许 the first big success成功 of
machine learning学习 commercially商业 was Google谷歌.
29
86252
4423
第一个机器学习的商业成功案例应该是谷歌
01:42
Google谷歌 showed显示 that it is
possible可能 to find information信息
30
90675
3109
谷歌用计算机算法寻找信息
01:45
by using运用 a computer电脑 algorithm算法,
31
93784
1752
01:47
and this algorithm算法 is based基于
on machine learning学习.
32
95536
2901
而且这个算法以计算机学习为基础
01:50
Since以来 that time, there have been many许多
commercial广告 successes成功 of machine learning学习.
33
98437
3886
从那以后,机器学习得到了很多的商业成功
01:54
Companies公司 like Amazon亚马逊 and NetflixNetflix公司
34
102323
1837
像亚马逊、网飞这类公司
01:56
use machine learning学习 to suggest建议
products制品 that you might威力 like to buy购买,
35
104160
3716
通过机器学习向你推荐你可能想买的东西
01:59
movies电影 that you might威力 like to watch.
36
107876
2020
你可能想看的电影
02:01
Sometimes有时, it's almost几乎 creepy爬行.
37
109896
1807
有时候你会被吓一跳
02:03
Companies公司 like LinkedInLinkedIn and FacebookFacebook的
38
111703
1954
像领英、脸谱这类的公司
02:05
sometimes有时 will tell you about
who your friends朋友 might威力 be
39
113657
2594
有时会告诉你谁会是你的朋友
02:08
and you have no idea理念 how it did it,
40
116251
1977
你根本不知道他们是如何做到的
02:10
and this is because it's using运用
the power功率 of machine learning学习.
41
118228
2967
其实他们正是运用了机器学习的力量
02:13
These are algorithms算法 that have
learned学到了 how to do this from data数据
42
121195
2957
这种运算方法使用数据
02:16
rather than being存在 programmed程序 by hand.
43
124152
3247
而非手动编写程序
02:19
This is also how IBMIBM was successful成功
44
127399
2478
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
02:21
in getting得到 Watson沃森 to beat击败
the two world世界 champions冠军 at "Jeopardy危险,"
45
129877
3862
02:25
answering回答 incredibly令人难以置信 subtle微妙
and complex复杂 questions问题 like this one.
46
133739
3225
成功回答了这样一个极其模糊且复杂的问题
02:28
["The ancient 'Lion'狮子 of Nimrud'尼姆鲁德” went missing失踪
from this city's城市的 national国民 museum博物馆 in 2003
(along沿 with a lot of other stuff东东)"]
47
136964
2835
[“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)”]
02:31
This is also why we are now able能够
to see the first self-driving自驾车 cars汽车.
48
139799
3235
这也是为什么我们现在有了第一台自驾车
02:35
If you want to be able能够 to tell
the difference区别 between之间, say,
49
143034
2822
如果你想区分一棵树和一个行人
02:37
a tree and a pedestrian行人,
well, that's pretty漂亮 important重要.
50
145856
2632
显然这很重要
02:40
We don't know how to write
those programs程式 by hand,
51
148488
2587
但是我们不知道如何写这样一个程序
02:43
but with machine learning学习,
this is now possible可能.
52
151075
2997
有了机器学习,这就成为了可能
02:46
And in fact事实, this car汽车 has driven驱动
over a million百万 miles英里
53
154072
2608
这台自驾车已经行驶了十万英里
02:48
without any accidents事故 on regular定期 roads道路.
54
156680
3506
在正常路面上零事故
02:52
So we now know that computers电脑 can learn学习,
55
160196
3914
我们知道电脑能够学习
02:56
and computers电脑 can learn学习 to do things
56
164110
1900
学习做一件有时我们自己都不知道怎么做的事情
02:58
that we actually其实 sometimes有时
don't know how to do ourselves我们自己,
57
166010
2838
03:00
or maybe can do them better than us.
58
168848
2885
有时甚至比我们做得更好
03:03
One of the most amazing惊人 examples例子
I've seen看到 of machine learning学习
59
171733
4195
我见过机器学习最惊人的例子
是我在Kaggle做的一个项目
03:07
happened发生 on a project项目 that I ran at KaggleKaggle
60
175928
2392
03:10
where a team球队 run by a guy
called Geoffrey杰弗里 Hinton韩丁
61
178320
3591
一个叫杰弗里 辛顿的人毕业于多伦多大学,
带领一个团队
03:13
from the University大学 of Toronto多伦多
62
181911
1552
03:15
won韩元 a competition竞争 for
automatic自动 drug药物 discovery发现.
63
183463
2677
赢得了一个自动查毒的竞赛
03:18
Now, what was extraordinary非凡 here
is not just that they beat击败
64
186140
2847
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
03:20
all of the algorithms算法 developed发达 by Merck默克
or the international国际 academic学术的 community社区,
65
188987
4013
03:25
but nobody没有人 on the team球队 had any background背景
in chemistry化学 or biology生物学 or life sciences科学,
66
193000
5061
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
03:30
and they did it in two weeks.
67
198061
2169
却在两个星期内赢得了比赛
03:32
How did they do this?
68
200230
1381
他们是如何做到的?
03:34
They used an extraordinary非凡 algorithm算法
called deep learning学习.
69
202421
2921
他们应用了一种超凡的算法叫做深度学习
03:37
So important重要 was this that in fact事实
the success成功 was covered覆盖
70
205342
2949
几个星期后纽约时报在其首页
报道了此次的重要成功
03:40
in The New York纽约 Times in a front面前 page
article文章 a few少数 weeks later后来.
71
208291
3121
03:43
This is Geoffrey杰弗里 Hinton韩丁
here on the left-hand左手 side.
72
211412
2735
在左手边就是杰弗里 辛顿
03:46
Deep learning学习 is an algorithm算法
inspired启发 by how the human人的 brain works作品,
73
214147
4341
深度学习是受到人类大脑的启发
03:50
and as a result结果 it's an algorithm算法
74
218488
1812
也因此这种算法的能力不受任何理论限制
03:52
which哪一个 has no theoretical理论 limitations限制
on what it can do.
75
220300
3841
03:56
The more data数据 you give it and the more
computation计算 time you give it,
76
224141
2823
你给它越多的数据和运算时间
03:58
the better it gets得到.
77
226964
1312
它会工作的越好
04:00
The New York纽约 Times also
showed显示 in this article文章
78
228276
2339
纽约时报在其文章中
还说明了深度学习的另一非凡之处
04:02
another另一个 extraordinary非凡
result结果 of deep learning学习
79
230615
2242
04:04
which哪一个 I'm going to show显示 you now.
80
232857
2712
现在我要展示给你们看
04:07
It shows节目 that computers电脑
can listen and understand理解.
81
235569
4941
它表明电脑能够听懂信息
04:12
(Video视频) Richard理查德 Rashid拉希德: Now, the last step
82
240510
2711
(视频)理查德 拉希德:现在,
我要做的最后一步是
04:15
that I want to be able能够
to take in this process处理
83
243221
3025
04:18
is to actually其实 speak说话 to you in Chinese中文.
84
246246
4715
用汉语和大家说话
04:22
Now the key thing there is,
85
250961
2635
在这之前,我们已经通过很多说汉语的人
收集了大量信息
04:25
we've我们已经 been able能够 to take a large amount
of information信息 from many许多 Chinese中文 speakers音箱
86
253596
5002
04:30
and produce生产 a text-to-speech文字转语音 system系统
87
258598
2530
然后形成一个语音合成系统
04:33
that takes Chinese中文 text文本
and converts转换 it into Chinese中文 language语言,
88
261128
4673
把汉字转换成汉语言
04:37
and then we've我们已经 taken采取
an hour小时 or so of my own拥有 voice语音
89
265801
4128
之后我们收录了一个小时我的声音
04:41
and we've我们已经 used that to modulate调制
90
269929
1891
使声音合成系统的声音听起来像我
04:43
the standard标准 text-to-speech文字转语音 system系统
so that it would sound声音 like me.
91
271820
4544
04:48
Again, the result's结果的 not perfect完善.
92
276364
2540
再次,结果并不完美
04:50
There are in fact事实 quite相当 a few少数 errors错误.
93
278904
2648
他们会有不少错误
04:53
(In Chinese中文)
94
281552
2484
(中文)
04:56
(Applause掌声)
95
284036
3367
(掌声)
05:01
There's much work to be doneDONE in this area.
96
289446
3576
在这个领域还有很多工作要做
05:05
(In Chinese中文)
97
293022
3645
(中文)
05:08
(Applause掌声)
98
296667
3433
(掌声)
05:13
Jeremy杰里米 Howard霍华德: Well, that was at
a machine learning学习 conference会议 in China中国.
99
301345
3399
杰里米 霍华德:这是在一个中国的机器学习会议上
05:16
It's not often经常, actually其实,
at academic学术的 conferences会议
100
304744
2370
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
05:19
that you do hear spontaneous自发 applause掌声,
101
307114
1897
05:21
although虽然 of course课程 sometimes有时
at TEDx的TEDx conferences会议, feel free自由.
102
309011
3676
当然除了TEDx演讲可以随意鼓掌
05:24
Everything you saw there
was happening事件 with deep learning学习.
103
312687
2795
你所看到的一切都伴随着深入学习
05:27
(Applause掌声) Thank you.
104
315482
1525
(掌声)谢谢
05:29
The transcription转录 in English英语
was deep learning学习.
105
317007
2282
对英文的转录是深入学习
05:31
The translation翻译 to Chinese中文 and the text文本
in the top最佳 right, deep learning学习,
106
319289
3412
翻译成汉语以及屏幕右上方的文字是深入学习
05:34
and the construction施工 of the voice语音
was deep learning学习 as well.
107
322701
3307
声音的合成也是深入学习
05:38
So deep learning学习 is
this extraordinary非凡 thing.
108
326008
3234
深入学习就是这样神奇的事情
05:41
It's a single algorithm算法 that
can seem似乎 to do almost几乎 anything,
109
329242
3099
这个单一的算法似乎可以做任何事情
05:44
and I discovered发现 that a year earlier,
it had also learned学到了 to see.
110
332341
3111
而且一年前我发现他甚至有视觉
05:47
In this obscure朦胧 competition竞争 from Germany德国
111
335452
2176
这个名不见经传的德国竞赛
05:49
called the German德语 Traffic交通 Sign标志
Recognition承认 Benchmark基准,
112
337628
2597
叫做德国交通标志识别基准
05:52
deep learning学习 had learned学到了
to recognize认识 traffic交通 signs迹象 like this one.
113
340225
3393
深度学习已学得识别这些交通标识
05:55
Not only could it
recognize认识 the traffic交通 signs迹象
114
343618
2094
它不仅能够做的比其它算法好
05:57
better than any other algorithm算法,
115
345712
1758
05:59
the leaderboard排行榜 actually其实 showed显示
it was better than people,
116
347470
2719
排行榜显示它比人更厉害
06:02
about twice两次 as good as people.
117
350189
1852
是人的准确率的两倍
06:04
So by 2011, we had the first example
118
352041
1996
到2011年,我们有了第一台视力高于人类的电脑
06:06
of computers电脑 that can see
better than people.
119
354037
3405
06:09
Since以来 that time, a lot has happened发生.
120
357442
2049
从此更多的电脑也可以做到
06:11
In 2012, Google谷歌 announced公布 that
they had a deep learning学习 algorithm算法
121
359491
3514
在2012年,谷歌宣布让一个深度学习的算法看YouTube视频
06:15
watch YouTubeYouTube的 videos视频
122
363005
1415
06:16
and crunched嘎吱嘎吱 the data数据
on 16,000 computers电脑 for a month,
123
364420
3437
收集16,000台电脑上的数据,为期一个月
06:19
and the computer电脑 independently独立地 learned学到了
about concepts概念 such这样 as people and cats
124
367857
4361
之后电脑便能仅通过看视频独立识别人和猫
06:24
just by watching观看 the videos视频.
125
372218
1809
06:26
This is much like the way
that humans人类 learn学习.
126
374027
2352
这近似于人类学习的过程
06:28
Humans人类 don't learn学习
by being存在 told what they see,
127
376379
2740
人类不需要被告诉他们看到了什么
06:31
but by learning学习 for themselves他们自己
what these things are.
128
379119
3331
而是在自己认知事物的过程中学习
06:34
Also in 2012, Geoffrey杰弗里 Hinton韩丁,
who we saw earlier,
129
382450
3369
同样在2012年,杰弗里 辛顿,我们之前看到的人
06:37
won韩元 the very popular流行 ImageNetImageNet competition竞争,
130
385819
2858
赢了很火的ImageNet比赛
06:40
looking to try to figure数字 out
from one and a half million百万 images图片
131
388677
4141
分辨出150万张图片的内容
06:44
what they're pictures图片 of.
132
392818
1438
06:46
As of 2014, we're now down
to a six percent百分 error错误 rate
133
394256
3533
到2014年,我们已经将图像识别的误差
降低到百分之六
06:49
in image图片 recognition承认.
134
397789
1453
06:51
This is better than people, again.
135
399242
2026
低于人类误差率
06:53
So machines really are doing
an extraordinarily异常 good job工作 of this,
136
401268
3769
这项非凡的工作现在已经用于工业
06:57
and it is now being存在 used in industry行业.
137
405037
2269
06:59
For example, Google谷歌 announced公布 last year
138
407306
3042
比如说,去年谷歌声明
07:02
that they had mapped映射 every一切 single
location位置 in France法国 in two hours小时,
139
410348
4585
他们在两小时内把法国的每一个地点汇成地图
07:06
and the way they did it was
that they fed美联储 street view视图 images图片
140
414933
3447
他们是将街景填入深度学习算法以辨认街道号
07:10
into a deep learning学习 algorithm算法
to recognize认识 and read street numbers数字.
141
418380
4319
07:14
Imagine想像 how long
it would have taken采取 before:
142
422699
2220
可以想象从前这件事要花费多少时间和精力
07:16
dozens许多 of people, many许多 years年份.
143
424919
3355
07:20
This is also happening事件 in China中国.
144
428274
1911
同样的事情也发生在中国
07:22
Baidu百度 is kind of
the Chinese中文 Google谷歌, I guess猜测,
145
430185
4036
百度大概类似于中国的谷歌
07:26
and what you see here in the top最佳 left
146
434221
2283
我们看到左上角
07:28
is an example of a picture图片 that I uploaded上传
to Baidu's百度的 deep learning学习 system系统,
147
436504
3974
是一张我上传到百度的深度学习系统的图片
07:32
and underneath you can see that the system系统
has understood了解 what that picture图片 is
148
440478
3769
下面你可以看到系统理解了这张照片
07:36
and found发现 similar类似 images图片.
149
444247
2236
并且找到了类似的图片
07:38
The similar类似 images图片 actually其实
have similar类似 backgrounds背景,
150
446483
2736
同样的背景
07:41
similar类似 directions方向 of the faces面孔,
151
449219
1658
同样的角度
07:42
even some with their tongue out.
152
450877
1788
有的甚至也有伸出来的舌头
07:44
This is not clearly明确地 looking
at the text文本 of a web卷筒纸 page.
153
452665
3030
网页上没有准确的文字
07:47
All I uploaded上传 was an image图片.
154
455695
1412
我只是上传了图片
07:49
So we now have computers电脑 which哪一个
really understand理解 what they see
155
457107
4021
所以说电脑能够真正理解它所看到的事物
07:53
and can therefore因此 search搜索 databases数据库
156
461128
1624
进而在数据库的几百万张图片中进行实时搜索
07:54
of hundreds数以百计 of millions百万
of images图片 in real真实 time.
157
462752
3554
07:58
So what does it mean
now that computers电脑 can see?
158
466306
3230
就现在而言,电脑的视力意味着什么呢?
08:01
Well, it's not just
that computers电脑 can see.
159
469536
2017
事实上不仅仅是电脑能够看见
08:03
In fact事实, deep learning学习
has doneDONE more than that.
160
471553
2069
深度学习其实可以做得更多
08:05
Complex复杂, nuanced细致入微 sentences句子 like this one
161
473622
2948
像这样一个细小复杂的语句
08:08
are now understandable可理解
with deep learning学习 algorithms算法.
162
476570
2824
对深度学习来说是相对易于理解的
08:11
As you can see here,
163
479394
1303
你可以看到
08:12
this Stanford-based基于斯坦福大学 system系统
showing展示 the red dot at the top最佳
164
480697
2768
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
08:15
has figured想通 out that this sentence句子
is expressing表达 negative sentiment情绪.
165
483465
3919
08:19
Deep learning学习 now in fact事实
is near human人的 performance性能
166
487384
3406
深度学习在理解语句内容方面已经接近人类水平
08:22
at understanding理解 what sentences句子 are about
and what it is saying about those things.
167
490802
5121
08:27
Also, deep learning学习 has
been used to read Chinese中文,
168
495923
2728
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
08:30
again at about native本地人
Chinese中文 speaker扬声器 level水平.
169
498651
3156
08:33
This algorithm算法 developed发达
out of Switzerland瑞士
170
501807
2168
这个算法开发于瑞士
08:35
by people, none没有 of whom speak说话
or understand理解 any Chinese中文.
171
503975
3356
没有一个人懂汉语
08:39
As I say, using运用 deep learning学习
172
507331
2051
要我说,深度学习是比较于人类
做这件事最好的系统
08:41
is about the best最好 system系统
in the world世界 for this,
173
509382
2219
08:43
even compared相比 to native本地人
human人的 understanding理解.
174
511601
5117
08:48
This is a system系统 that we
put together一起 at my company公司
175
516718
2964
这个系统是在我们公司建立的
08:51
which哪一个 shows节目 putting
all this stuff东东 together一起.
176
519682
2046
它要把这些东西集合起来
08:53
These are pictures图片 which哪一个
have no text文本 attached,
177
521728
2461
这些图片没有文字描述
08:56
and as I'm typing打字 in here sentences句子,
178
524189
2352
随着我在这输入文字
08:58
in real真实 time it's understanding理解
these pictures图片
179
526541
2969
同时它会了解这些图片
09:01
and figuring盘算 out what they're about
180
529510
1679
理解它们是关于什么的
09:03
and finding发现 pictures图片 that are similar类似
to the text文本 that I'm writing写作.
181
531189
3163
然后找出和这些相似的图片
09:06
So you can see, it's actually其实
understanding理解 my sentences句子
182
534352
2756
所以你看,他真正在理解我的文字
09:09
and actually其实 understanding理解 these pictures图片.
183
537108
2224
理解这些图片
09:11
I know that you've seen看到
something like this on Google谷歌,
184
539332
2559
我知道你在谷歌上看到过类似的
09:13
where you can type类型 in things
and it will show显示 you pictures图片,
185
541891
2775
你可以输入文字,它会提供给你图片
09:16
but actually其实 what it's doing is it's
searching搜索 the webpage网页 for the text文本.
186
544666
3424
但实际上它是在网页上搜索文字
09:20
This is very different不同 from actually其实
understanding理解 the images图片.
187
548090
3001
这和理解图片是有很大不同的
09:23
This is something that computers电脑
have only been able能够 to do
188
551091
2752
理解图片是电脑在过去几个月里才刚刚会做的事情
09:25
for the first time in the last few少数 months个月.
189
553843
3248
09:29
So we can see now that computers电脑
can not only see but they can also read,
190
557091
4091
电脑不仅有视力,而且能够阅读
09:33
and, of course课程, we've我们已经 shown显示 that they
can understand理解 what they hear.
191
561182
3765
而且当然,电脑也能理解所听到的
09:36
Perhaps也许 not surprising奇怪 now that
I'm going to tell you they can write.
192
564947
3442
也许并不意外,我现在要告诉你们,电脑也可以写
09:40
Here is some text文本 that I generated产生
using运用 a deep learning学习 algorithm算法 yesterday昨天.
193
568389
4783
这是我昨天用深度学习算法写的文字
09:45
And here is some text文本 that an algorithm算法
out of Stanford斯坦福 generated产生.
194
573172
3924
这些是斯坦福的算法做的
09:49
Each of these sentences句子 was generated产生
195
577096
1764
每一句话都是深度学习算法对图片进行的描述
09:50
by a deep learning学习 algorithm算法
to describe描述 each of those pictures图片.
196
578860
4249
09:55
This algorithm算法 before has never seen看到
a man in a black黑色 shirt衬衫 playing播放 a guitar吉他.
197
583109
4472
算法没见过一个穿黑衣服的男人弹吉他
09:59
It's seen看到 a man before,
it's seen看到 black黑色 before,
198
587581
2220
它见过男人,见过黑色
10:01
it's seen看到 a guitar吉他 before,
199
589801
1599
见过吉他
10:03
but it has independently独立地 generated产生
this novel小说 description描述 of this picture图片.
200
591400
4294
它便自己对这个图片作出了这样的描述
10:07
We're still not quite相当 at human人的
performance性能 here, but we're close.
201
595694
3502
我们还做不到完全和人类同等水平,
但我们已经很接近了
10:11
In tests测试, humans人类 prefer比较喜欢
the computer-generated计算机生成的 caption标题
202
599196
4068
统计表明,四分之一的人更喜欢电脑做的图片说明
10:15
one out of four times.
203
603264
1527
10:16
Now this system系统 is now only two weeks old,
204
604791
2064
目前这个系统刚被开发两周之久
10:18
so probably大概 within the next下一个 year,
205
606855
1846
所以按这个速度,估计明年
10:20
the computer电脑 algorithm算法 will be
well past过去 human人的 performance性能
206
608701
2801
电脑算法会超过人类水平
10:23
at the rate things are going.
207
611502
1862
10:25
So computers电脑 can also write.
208
613364
3049
电脑会写
10:28
So we put all this together一起 and it leads引线
to very exciting扣人心弦 opportunities机会.
209
616413
3475
我们把这些都放在一起,会发现一个令人兴奋的机遇
10:31
For example, in medicine医学,
210
619888
1492
比如说,在医药业
10:33
a team球队 in Boston波士顿 announced公布
that they had discovered发现
211
621380
2525
一个波士顿团队宣布
10:35
dozens许多 of new clinically临床 relevant相应 features特征
212
623905
2949
他们发现了肿瘤的几十种临床表现
10:38
of tumors肿瘤 which哪一个 help doctors医生
make a prognosis预测 of a cancer癌症.
213
626854
4266
帮助医生预测癌症
10:44
Very similarly同样, in Stanford斯坦福,
214
632220
2296
同样的,在斯坦福
10:46
a group there announced公布 that,
looking at tissues组织 under magnification放大,
215
634516
3663
一个团队宣布通过用放大镜观察组织
10:50
they've他们已经 developed发达
a machine learning-based学习为主 system系统
216
638179
2381
开发了一个基于机器学习的系统
10:52
which哪一个 in fact事实 is better
than human人的 pathologists病理学家
217
640560
2582
可以比病理学家更有效地预测癌症患者的幸存率
10:55
at predicting预测 survival生存 rates利率
for cancer癌症 sufferers患者.
218
643142
4377
10:59
In both of these cases, not only
were the predictions预测 more accurate准确,
219
647519
3245
在这两个例子中,不仅预测更加准确
11:02
but they generated产生 new insightful见地 science科学.
220
650764
2502
而且他们创造了新的科学视角
11:05
In the radiology放射科 case案件,
221
653276
1505
在放射学中
11:06
they were new clinical临床 indicators指标
that humans人类 can understand理解.
222
654781
3095
新视角是人类可以明白的新临床表现
11:09
In this pathology病理 case案件,
223
657876
1792
在病理学中
11:11
the computer电脑 system系统 actually其实 discovered发现
that the cells细胞 around the cancer癌症
224
659668
4500
电脑发现癌细胞周围的细胞
11:16
are as important重要 as
the cancer癌症 cells细胞 themselves他们自己
225
664168
3340
在诊断中同癌细胞一样重要
11:19
in making制造 a diagnosis诊断.
226
667508
1752
11:21
This is the opposite对面 of what pathologists病理学家
had been taught for decades几十年.
227
669260
5361
这和病理学家几十年来的教学是相反的
11:26
In each of those two cases,
they were systems系统 developed发达
228
674621
3292
这两个案例中的系统都是由
11:29
by a combination组合 of medical experts专家
and machine learning学习 experts专家,
229
677913
3621
医学专家和机器学习专家共同开发的
11:33
but as of last year,
we're now beyond that too.
230
681534
2741
去年我们就已经超过了这个水平
11:36
This is an example of
identifying识别 cancerous癌的 areas
231
684275
3549
这个是用显微镜识别组织癌变区的例子
11:39
of human人的 tissue组织 under a microscope显微镜.
232
687824
2530
11:42
The system系统 being存在 shown显示 here
can identify鉴定 those areas more accurately准确,
233
690354
4613
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
11:46
or about as accurately准确,
as human人的 pathologists病理学家,
234
694967
2775
甚至比病理专家更准确
11:49
but was built内置 entirely完全 with deep learning学习
using运用 no medical expertise专门知识
235
697742
3392
但是建立系统的都是深度学习的专家
11:53
by people who have
no background背景 in the field领域.
236
701134
2526
没有一个医学专家
11:56
Similarly同样, here, this neuron神经元 segmentation分割.
237
704730
2555
类似的,这是神经细胞分裂
11:59
We can now segment分割 neurons神经元
about as accurately准确 as humans人类 can,
238
707285
3668
我们已经可以和人类一样准确地分裂细胞
12:02
but this system系统 was developed发达
with deep learning学习
239
710953
2717
但这是个深度学习系统
12:05
using运用 people with no previous以前
background背景 in medicine医学.
240
713670
3251
没有一个开发者拥有医学背景
12:08
So myself, as somebody with
no previous以前 background背景 in medicine医学,
241
716921
3227
对于我这个完全没有医学背景的人来说
12:12
I seem似乎 to be entirely完全 well qualified合格
to start开始 a new medical company公司,
242
720148
3727
看起来我也完全可以开一个医药公司
12:15
which哪一个 I did.
243
723875
2146
我确实这么做了
12:18
I was kind of terrified of doing it,
244
726021
1740
我开始有点不知所措
12:19
but the theory理论 seemed似乎 to suggest建议
that it ought应该 to be possible可能
245
727761
2889
但理论上说这件事是可行的
12:22
to do very useful有用 medicine医学
using运用 just these data数据 analytic解析 techniques技术.
246
730650
5492
用这些数据分析技术制作医药
12:28
And thankfully感激地, the feedback反馈
has been fantastic奇妙,
247
736142
2480
所幸的是,反响非常好
12:30
not just from the media媒体
but from the medical community社区,
248
738622
2356
不仅是媒体的,包括医药行业
12:32
who have been very supportive支持.
249
740978
2344
都很支持
12:35
The theory理论 is that we can take
the middle中间 part部分 of the medical process处理
250
743322
4149
理论表明我们可以将制药的中间过程
12:39
and turn that into data数据 analysis分析
as much as possible可能,
251
747471
2893
充分转换成数据分析
12:42
leaving离开 doctors医生 to do
what they're best最好 at.
252
750364
3065
让医生去做他们最擅长的
12:45
I want to give you an example.
253
753429
1602
我有一个例子
12:47
It now takes us about 15 minutes分钟
to generate生成 a new medical diagnostic诊断 test测试
254
755031
4944
制作一个医学诊断测试需要十五分钟
12:51
and I'll show显示 you that in real真实 time now,
255
759975
1954
我会给你们实际展示
12:53
but I've compressed压缩 it down to
three minutes分钟 by cutting切割 some pieces out.
256
761929
3487
但是我去掉了一部分,把它压缩到了三分钟
12:57
Rather than showing展示 you
creating创建 a medical diagnostic诊断 test测试,
257
765416
3061
不要医学诊断试验
13:00
I'm going to show显示 you
a diagnostic诊断 test测试 of car汽车 images图片,
258
768477
3369
我要给你们展示制作一个汽车图片的诊断测试
13:03
because that's something
we can all understand理解.
259
771846
2222
因为这个我们都能懂
13:06
So here we're starting开始 with
about 1.5 million百万 car汽车 images图片,
260
774068
3201
现在我们有150万张汽车图片
13:09
and I want to create创建 something
that can split分裂 them into the angle角度
261
777269
3206
我想要根据拍照的角度对他们进行分类
13:12
of the photo照片 that's being存在 taken采取.
262
780475
2223
13:14
So these images图片 are entirely完全 unlabeled未标记,
so I have to start开始 from scratch.
263
782698
3888
这些图片完全没有标签,所以我要先对他们进行简单描述
13:18
With our deep learning学习 algorithm算法,
264
786586
1865
有深度学习算法
13:20
it can automatically自动 identify鉴定
areas of structure结构体 in these images图片.
265
788451
3707
它可以自动识别图片的结构要素
13:24
So the nice不错 thing is that the human人的
and the computer电脑 can now work together一起.
266
792158
3620
令人高兴的是人和电脑可以合作
13:27
So the human人的, as you can see here,
267
795778
2178
你可以看到,这个人
13:29
is telling告诉 the computer电脑
about areas of interest利益
268
797956
2675
正在告诉电脑什么是感兴趣的要素
13:32
which哪一个 it wants the computer电脑 then
to try and use to improve提高 its algorithm算法.
269
800631
4650
为之后电脑用来完善算法
13:37
Now, these deep learning学习 systems系统 actually其实
are in 16,000-dimensional space空间,
270
805281
4296
现在,这些深度学习算法处在16,000维空间中
13:41
so you can see here the computer电脑
rotating旋转 this through通过 that space空间,
271
809577
3432
所以你看到电脑让他们在这个空间中旋转
13:45
trying to find new areas of structure结构体.
272
813009
1992
尝试找到新的结构要素
13:47
And when it does so successfully顺利,
273
815001
1781
当他成功时
13:48
the human人的 who is driving主动 it can then
point out the areas that are interesting有趣.
274
816782
4004
开车的人就可以指出感兴趣的要素
13:52
So here, the computer电脑 has
successfully顺利 found发现 areas,
275
820786
2422
现在电脑成功找出这些要素
13:55
for example, angles.
276
823208
2562
比如,角度
13:57
So as we go through通过 this process处理,
277
825770
1606
我们在这个过程中
13:59
we're gradually逐渐 telling告诉
the computer电脑 more and more
278
827376
2340
逐渐的告诉电脑更多
14:01
about the kinds of structures结构
we're looking for.
279
829716
2428
我们想寻找的结构
14:04
You can imagine想像 in a diagnostic诊断 test测试
280
832144
1772
你可以想象一个诊断测试
14:05
this would be a pathologist病理学家 identifying识别
areas of pathosis病态, for example,
281
833916
3350
这就像是病理学家识别病态区域
14:09
or a radiologist放射科医生 indicating说明
potentially可能 troublesome麻烦 nodules结节.
282
837266
5026
或者放射学专家找出潜在的问题囊肿
14:14
And sometimes有时 it can be
difficult for the algorithm算法.
283
842292
2559
有时候这对算法来说有些难度
14:16
In this case案件, it got kind of confused困惑.
284
844851
1964
我们的例子就比较麻烦
14:18
The fronts战线 and the backs
of the cars汽车 are all mixed up.
285
846815
2550
车的正面和背面全部混淆了
14:21
So here we have to be a bit more careful小心,
286
849365
2072
所以我们要仔细一些
14:23
manually手动 selecting选择 these fronts战线
as opposed反对 to the backs,
287
851437
3232
人工地选出正面和背面
14:26
then telling告诉 the computer电脑
that this is a type类型 of group
288
854669
5506
人后告诉电脑这是我们所感兴趣的一类
14:32
that we're interested有兴趣 in.
289
860175
1348
14:33
So we do that for a while,
we skip跳跃 over a little bit,
290
861523
2677
做这件事花了一些时间,所以我们跳过
14:36
and then we train培养 the
machine learning学习 algorithm算法
291
864200
2246
之后我们用这几百个东西训练机器学习算法
14:38
based基于 on these couple一对 of hundred things,
292
866446
1974
14:40
and we hope希望 that it's gotten得到 a lot better.
293
868420
2025
希望他会有很大进步
14:42
You can see, it's now started开始 to fade褪色
some of these pictures图片 out,
294
870445
3073
你能看到,它正在消退一些图片
14:45
showing展示 us that it already已经 is recognizing认识
how to understand理解 some of these itself本身.
295
873518
4708
说明他已经开始可以自己理解这些图片了
14:50
We can then use this concept概念
of similar类似 images图片,
296
878226
2902
我们可以用相似图片的概念
14:53
and using运用 similar类似 images图片, you can now see,
297
881128
2094
用相似的图片,你可以看到
14:55
the computer电脑 at this point is able能够 to
entirely完全 find just the fronts战线 of cars汽车.
298
883222
4019
电脑现在能够只找出正面的车
14:59
So at this point, the human人的
can tell the computer电脑,
299
887241
2948
在这个时候,人可以告诉电脑
15:02
okay, yes, you've doneDONE
a good job工作 of that.
300
890189
2293
对的,没错,你做的很好
15:05
Sometimes有时, of course课程, even at this point
301
893652
2185
当然,有时,即使在这个阶段
15:07
it's still difficult
to separate分离 out groups.
302
895837
3674
分组仍然是很困难的
15:11
In this case案件, even after we let the
computer电脑 try to rotate回转 this for a while,
303
899511
3884
像我们这里,让电脑在这里旋转了一段时间了
15:15
we still find that the left sides双方
and the right sides双方 pictures图片
304
903399
3345
我们还是看到左面的和右面的图片有混淆
15:18
are all mixed up together一起.
305
906744
1478
15:20
So we can again give
the computer电脑 some hints提示,
306
908222
2140
所以我们可以再一次给电脑一些提示
15:22
and we say, okay, try and find
a projection投影 that separates中隔离 out
307
910362
2976
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
15:25
the left sides双方 and the right sides双方
as much as possible可能
308
913338
2607
15:27
using运用 this deep learning学习 algorithm算法.
309
915945
2122
15:30
And giving it that hint暗示 --
ah, okay, it's been successful成功.
310
918067
2942
有了这个指示——好的,它已经完成了
15:33
It's managed管理 to find a way
of thinking思维 about these objects对象
311
921009
2882
它要想办法分开这一部分
15:35
that's separated分离 out these together一起.
312
923891
2380
15:38
So you get the idea理念 here.
313
926271
2438
你现在知道了
15:40
This is a case案件 not where the human人的
is being存在 replaced更换 by a computer电脑,
314
928709
8197
这不是电脑取代人类
15:48
but where they're working加工 together一起.
315
936906
2640
而是一起合作
15:51
What we're doing here is we're replacing更换
something that used to take a team球队
316
939546
3550
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
15:55
of five or six people about seven years年份
317
943096
2002
15:57
and replacing更换 it with something
that takes 15 minutes分钟
318
945098
2605
变成只需一个人花十五分钟就能完成
15:59
for one person acting演戏 alone单独.
319
947703
2505
16:02
So this process处理 takes about
four or five iterations迭代.
320
950208
3950
这个过程需要四到五次反复
16:06
You can see we now have 62 percent百分
321
954158
1859
你可以看到我们已经将150万张图片的62%正确分类
16:08
of our 1.5 million百万 images图片
classified分类 correctly正确地.
322
956017
2959
16:10
And at this point, we
can start开始 to quite相当 quickly很快
323
958976
2472
现在我们就可以快速地检查整个分组
16:13
grab whole整个 big sections部分,
324
961448
1297
16:14
check through通过 them to make sure
that there's no mistakes错误.
325
962745
2919
确保没有错误
16:17
Where there are mistakes错误, we can
let the computer电脑 know about them.
326
965664
3952
如果哪里有错误,我们可以告诉电脑
16:21
And using运用 this kind of process处理
for each of the different不同 groups,
327
969616
3045
每个分组我们都这样做
16:24
we are now up to
an 80 percent百分 success成功 rate
328
972661
2487
现在这150万张图片已经达到80%的成功率
16:27
in classifying分类 the 1.5 million百万 images图片.
329
975148
2415
16:29
And at this point, it's just a case案件
330
977563
2078
现在这个阶段
16:31
of finding发现 the small number
that aren't classified分类 correctly正确地,
331
979641
3579
只需要找出几个不正确的分类
16:35
and trying to understand理解 why.
332
983220
2888
并让电脑明白为什么
16:38
And using运用 that approach途径,
333
986108
1743
到了这个步骤
16:39
by 15 minutes分钟 we get
to 97 percent百分 classification分类 rates利率.
334
987851
4121
十五分钟后我们达到了97%的正确率
16:43
So this kind of technique技术
could allow允许 us to fix固定 a major重大的 problem问题,
335
991972
4600
这种技术能帮助我们解决一个问题
16:48
which哪一个 is that there's a lack缺乏
of medical expertise专门知识 in the world世界.
336
996578
3036
医疗专家不足的问题
16:51
The World世界 Economic经济 Forum论坛 says
that there's between之间 a 10x and a 20x
337
999614
3489
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
16:55
shortage短缺 of physicians医师
in the developing发展 world世界,
338
1003103
2624
16:57
and it would take about 300 years年份
339
1005727
2113
而弥补这一短缺需要300年的时间
16:59
to train培养 enough足够 people
to fix固定 that problem问题.
340
1007840
2894
17:02
So imagine想像 if we can help
enhance提高 their efficiency效率
341
1010734
2885
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?
17:05
using运用 these deep learning学习 approaches方法?
342
1013619
2839
17:08
So I'm very excited兴奋
about the opportunities机会.
343
1016458
2232
我对这个机会表示很激动
17:10
I'm also concerned关心 about the problems问题.
344
1018690
2589
我同样的担心一些问题
17:13
The problem问题 here is that
every一切 area in blue蓝色 on this map地图
345
1021279
3124
问题是在这张地图上的蓝色区域内
17:16
is somewhere某处 where services服务
are over 80 percent百分 of employment雇用.
346
1024403
3769
服务占就业的80%以上
17:20
What are services服务?
347
1028172
1787
什么是服务?
17:21
These are services服务.
348
1029959
1514
这些是服务
17:23
These are also the exact精确 things that
computers电脑 have just learned学到了 how to do.
349
1031473
4154
这些也是电脑才刚刚开始学习的事情
17:27
So 80 percent百分 of the world's世界 employment雇用
in the developed发达 world世界
350
1035627
3804
也就是说世界上发达国家的80%的就业
17:31
is stuff东东 that computers电脑
have just learned学到了 how to do.
351
1039431
2532
是电脑刚开始学习的
17:33
What does that mean?
352
1041963
1440
这是什么意思?
17:35
Well, it'll它会 be fine.
They'll他们会 be replaced更换 by other jobs工作.
353
1043403
2583
其实也没什么大不了的,他们会被其他职业替代
17:37
For example, there will be
more jobs工作 for data数据 scientists科学家们.
354
1045986
2707
比如说会有更多的数据学家
17:40
Well, not really.
355
1048693
817
也不尽然
17:41
It doesn't take data数据 scientists科学家们
very long to build建立 these things.
356
1049510
3118
数据学家不需要太久的时间做这些事
17:44
For example, these four algorithms算法
were all built内置 by the same相同 guy.
357
1052628
3252
比如这四个算法都是同时一个人开发的
17:47
So if you think, oh,
it's all happened发生 before,
358
1055880
2438
如果你认为这些曾经都发生过
17:50
we've我们已经 seen看到 the results结果 in the past过去
of when new things come along沿
359
1058318
3808
我们看到过新的事物出现
17:54
and they get replaced更换 by new jobs工作,
360
1062126
2252
然后被新的职业所取代
17:56
what are these new jobs工作 going to be?
361
1064378
2116
那这些新的职业又会是什么?
17:58
It's very hard for us to estimate估计 this,
362
1066494
1871
很难做出估计
18:00
because human人的 performance性能
grows成长 at this gradual rate,
363
1068365
2739
因为人的能力以这个均匀的速度增长
18:03
but we now have a system系统, deep learning学习,
364
1071104
2562
但是现在我们有了深度学习系统
18:05
that we know actually其实 grows成长
in capability能力 exponentially成倍.
365
1073666
3227
它的能力以指数方式增长
18:08
And we're here.
366
1076893
1605
我们现在在这
18:10
So currently目前, we see the things around us
367
1078498
2061
目前,我们看周围的事物
18:12
and we say, "Oh, computers电脑
are still pretty漂亮 dumb." Right?
368
1080559
2676
会说:“电脑还是很笨。”对吧?
18:15
But in five years'年份' time,
computers电脑 will be off this chart图表.
369
1083235
3429
但是在五年内,电脑会超出这张图
18:18
So we need to be starting开始 to think
about this capability能力 right now.
370
1086664
3865
所以我们现在要开始考虑这样的能力了
18:22
We have seen看到 this once一旦 before, of course课程.
371
1090529
2050
当然,我们曾经见过这个
18:24
In the Industrial产业 Revolution革命,
372
1092579
1387
在工业革命时期
18:25
we saw a step change更改
in capability能力 thanks谢谢 to engines引擎.
373
1093966
2851
发动机让生产力迈进一大步
18:29
The thing is, though虽然,
that after a while, things flattened扁平 out.
374
1097667
3138
然而问题是,一段时间之后,形势转平了
18:32
There was social社会 disruption瓦解,
375
1100805
1702
是由于社会的破坏
18:34
but once一旦 engines引擎 were used
to generate生成 power功率 in all the situations情况,
376
1102507
3439
但当发动机被普遍应用时
18:37
things really settled安定 down.
377
1105946
2354
一切都稳定下来了
18:40
The Machine Learning学习 Revolution革命
378
1108300
1473
机器学习革命
18:41
is going to be very different不同
from the Industrial产业 Revolution革命,
379
1109773
2909
将和工业革命有很大不同
18:44
because the Machine Learning学习 Revolution革命,
it never settles结算 down.
380
1112682
2950
因为机器学习革命不会停止
18:47
The better computers电脑 get
at intellectual知识分子 activities活动,
381
1115632
2982
电脑越擅长智能活动
18:50
the more they can build建立 better computers电脑
to be better at intellectual知识分子 capabilities功能,
382
1118614
4248
它们越能制造出更加擅长智能活动的电脑
18:54
so this is going to be a kind of change更改
383
1122862
1908
这将会是世界从未经历过的改变
18:56
that the world世界 has actually其实
never experienced有经验的 before,
384
1124770
2478
18:59
so your previous以前 understanding理解
of what's possible可能 is different不同.
385
1127248
3306
所以你之前理解的可能性是不一样的
19:02
This is already已经 impacting影响 us.
386
1130974
1780
这正在影响我们的生活
19:04
In the last 25 years年份,
as capital首都 productivity生产率 has increased增加,
387
1132754
3630
在过去的25年里,随着资本生产力的增加
19:08
labor劳动 productivity生产率 has been flat平面,
in fact事实 even a little bit down.
388
1136400
4188
劳动生产力在变缓,甚至下降
19:13
So I want us to start开始
having this discussion讨论 now.
389
1141408
2741
所以我希望可以发起大家的讨论
19:16
I know that when I often经常 tell people
about this situation情况,
390
1144149
3027
我知道当我和人们讲述这样的处境时
19:19
people can be quite相当 dismissive不屑一顾.
391
1147176
1490
人们往往表现出不以为然
19:20
Well, computers电脑 can't really think,
392
1148666
1673
电脑不会思考
19:22
they don't emote作表情,
they don't understand理解 poetry诗歌,
393
1150339
3028
它们没有情感,也不懂诗
19:25
we don't really understand理解 how they work.
394
1153367
2521
它们甚至都不知道自己是如何运作的
19:27
So what?
395
1155888
1486
那又怎样?
19:29
Computers电脑 right now can do the things
396
1157374
1804
电脑现在可以做
19:31
that humans人类 spend most
of their time being存在 paid支付 to do,
397
1159178
2719
人类用大部分有偿的劳动时间做的事情
19:33
so now's现在是 the time to start开始 thinking思维
398
1161897
1731
所以现在该到我们思考
19:35
about how we're going to adjust调整 our
social社会 structures结构 and economic经济 structures结构
399
1163628
4387
我们将如何调整我们的社会结构和经济结构
19:40
to be aware知道的 of this new reality现实.
400
1168015
1840
来应对新形势
19:41
Thank you.
401
1169855
1533
谢谢
19:43
(Applause掌声)
402
1171388
802
(鼓掌)
Translated by Yuchen Shen
Reviewed by Lee Li

▲Back to top

ABOUT THE SPEAKER
Jeremy Howard - Data scientist
Jeremy Howard imagines how advanced machine learning can improve our lives.

Why you should listen

Jeremy Howard is the CEO of Enlitic, an advanced machine learning company in San Francisco. Previously, he was the president and chief scientist at Kaggle, a community and competition platform of over 200,000 data scientists. Howard is a faculty member at Singularity University, where he teaches data science. He is also a Young Global Leader with the World Economic Forum, and spoke at the World Economic Forum Annual Meeting 2014 on "Jobs for the Machines."

Howard advised Khosla Ventures as their Data Strategist, identifying opportunities for investing in data-driven startups and mentoring portfolio companies to build data-driven businesses. He was the founding CEO of two successful Australian startups, FastMail and Optimal Decisions Group.

More profile about the speaker
Jeremy Howard | Speaker | TED.com