TED2016
Anthony Goldbloom: The jobs we'll lose to machines -- and the ones we won't
安東尼．葛博倫: 即將要被機器取代以及無法取代的工作
Filmed:
Readability: 4.8
2,568,213 views
機器學習不再僅止於簡單任務，好比評估信用風險、分類郵件。如今，它能夠承擔更複雜的工作，像是幫作文打分數、診斷疾病。這些進步帶來了一個令人不安的問題：未來機器人會搶走你的工作嗎？
Anthony Goldbloom - Machine learning expert
Anthony Goldbloom crowdsources solutions to difficult problems using machine learning. Full bio
Anthony Goldbloom crowdsources solutions to difficult problems using machine learning. Full bio
Double-click the English transcript below to play the video.
00:12
So this is my niece.
0
968
1262
這是我的姪女，
00:14
Her name is Yahli.
1
2644
1535
她的名字是雅莉，
00:16
She is nine months old.
2
4203
1511
她現在九個月大，
00:18
Her mum is a doctor,
and her dad is a lawyer.
and her dad is a lawyer.
3
6201
2528
媽媽是位醫生、爸爸是位律師；
00:21
By the time Yahli goes to college,
4
9269
2006
不過等到她上大學的時候
00:23
the jobs her parents do
are going to look dramatically different.
are going to look dramatically different.
5
11299
3253
她父母親的工作將會迥然不同了。
00:27
In 2013, researchers at Oxford University
did a study on the future of work.
did a study on the future of work.
6
15347
5073
2013年，牛津大學的研究人員
做了一個對未來工作的研究，
做了一個對未來工作的研究，
00:32
They concluded that almost one
in every two jobs have a high risk
in every two jobs have a high risk
7
20766
4139
他們得出結論：差不多將近一半的工作
00:36
of being automated by machines.
8
24929
1824
都有被機器自動化取代的危險。
00:40
Machine learning is the technology
9
28388
1905
而「機器學習」
要對這種顛覆負主要的責任。
00:42
that's responsible for most
of this disruption.
of this disruption.
10
30317
2278
00:44
It's the most powerful branch
of artificial intelligence.
of artificial intelligence.
11
32619
2790
它是人工智慧最呼風喚雨的分支，
00:47
It allows machines to learn from data
12
35433
1882
它讓機器得以從數據中學習，
00:49
and mimic some of the things
that humans can do.
that humans can do.
13
37339
2592
並模仿一些人類可以做到的事情。
00:51
My company, Kaggle, operates
on the cutting edge of machine learning.
on the cutting edge of machine learning.
14
39955
3415
我的公司「Kaggle」算是能操控
機器學習的尖端科技公司。
機器學習的尖端科技公司。
00:55
We bring together
hundreds of thousands of experts
hundreds of thousands of experts
15
43394
2386
我們召集了成千上萬的專家
00:57
to solve important problems
for industry and academia.
for industry and academia.
16
45804
3118
為產、學界解決重要的難題。
01:01
This gives us a unique perspective
on what machines can do,
on what machines can do,
17
49279
3222
所以我們可以從獨特的角度
來觀察機器可以做什麽、
來觀察機器可以做什麽、
01:04
what they can't do
18
52525
1235
不可以做什麽。
01:05
and what jobs they might
automate or threaten.
automate or threaten.
19
53784
2939
哪些工作可以被自動化或者受到威脅。
01:09
Machine learning started making its way
into industry in the early '90s.
into industry in the early '90s.
20
57316
3550
機器學習是在90年代初
進入產業界的，
進入產業界的，
01:12
It started with relatively simple tasks.
21
60890
2124
一開始，它只是執行一些
簡單的任務。
簡單的任務。
01:15
It started with things like assessing
credit risk from loan applications,
credit risk from loan applications,
22
63406
4115
像評估貸款申請的信用風險、
01:19
sorting the mail by reading
handwritten characters from zip codes.
handwritten characters from zip codes.
23
67545
4053
查看郵遞區號的手寫字碼
來分類郵件。
來分類郵件。
01:24
Over the past few years, we have made
dramatic breakthroughs.
dramatic breakthroughs.
24
72036
3169
過去幾年來我們已經做出
多項重大的突破，
多項重大的突破，
01:27
Machine learning is now capable
of far, far more complex tasks.
of far, far more complex tasks.
25
75586
3916
機器學習現在已經可以完成
非常覆雜的任務。
非常覆雜的任務。
01:31
In 2012, Kaggle challenged its community
26
79860
3231
在 2012 年 Kaggle
給自家社群出了一道難題，
給自家社群出了一道難題，
01:35
to build an algorithm
that could grade high-school essays.
that could grade high-school essays.
27
83115
3189
要大家設計出一個演算法
來評判高中作文。
來評判高中作文。
01:38
The winning algorithms
were able to match the grades
were able to match the grades
28
86328
2604
獲勝的演算法給出的分數居然
和真正老師給出的分數相符
和真正老師給出的分數相符
01:40
given by human teachers.
29
88956
1665
去年，我們出了一道更難的題目：
01:43
Last year, we issued
an even more difficult challenge.
an even more difficult challenge.
30
91092
2984
01:46
Can you take images of the eye
and diagnose an eye disease
and diagnose an eye disease
31
94100
2953
你可不可以藉由眼球的影像
診斷出一種叫「糖尿病視網膜病變」的眼疾？
01:49
called diabetic retinopathy?
32
97077
1694
01:51
Again, the winning algorithms
were able to match the diagnoses
were able to match the diagnoses
33
99164
4040
果然，獲勝的演算法給出的診斷
可以和人類眼科醫師的診斷相媲美。
01:55
given by human ophthalmologists.
34
103228
1825
01:57
Now, given the right data,
machines are going to outperform humans
machines are going to outperform humans
35
105561
3212
只要給定正確的數據 ，
機器在類似的任務中
機器在類似的任務中
02:00
at tasks like this.
36
108797
1165
將完全超越人類。
02:01
A teacher might read 10,000 essays
over a 40-year career.
over a 40-year career.
37
109986
3992
一位老師，在他的40年職業生涯中
也許只能審閱10000篇作文
也許只能審閱10000篇作文
02:06
An ophthalmologist might see 50,000 eyes.
38
114407
2360
一名眼科醫生，大概可以看50,000隻眼睛
02:08
A machine can read millions of essays
or see millions of eyes
or see millions of eyes
39
116791
3913
但一部機器可以在短短幾分鐘內
讀完上百萬篇文章
讀完上百萬篇文章
02:12
within minutes.
40
120728
1276
或是看完上百萬顆眼睛。
02:14
We have no chance of competing
against machines
against machines
41
122456
2858
在頻繁、大批量的任務上
我們無法與機器抗衡。
我們無法與機器抗衡。
02:17
on frequent, high-volume tasks.
42
125338
2321
02:20
But there are things we can do
that machines can't do.
that machines can't do.
43
128665
3724
不過還是有我們能做
而機器做不到的事情，
而機器做不到的事情，
02:24
Where machines have made
very little progress
very little progress
44
132791
2200
機器在解決複雜的新情況方面
02:27
is in tackling novel situations.
45
135015
1854
進展甚微。
02:28
They can't handle things
they haven't seen many times before.
they haven't seen many times before.
46
136893
3899
它們對還沒看到很多次的事情無法掌握。
02:33
The fundamental limitations
of machine learning
of machine learning
47
141321
2584
機器學習的先天限制就是：
02:35
is that it needs to learn
from large volumes of past data.
from large volumes of past data.
48
143929
3394
它需要從大量的過往資料中學習。
02:39
Now, humans don't.
49
147347
1754
人類就不一樣了，
02:41
We have the ability to connect
seemingly disparate threads
seemingly disparate threads
50
149125
3030
我們有一種能把看似毫不相關的事物
聯系起來的能力，
聯系起來的能力，
02:44
to solve problems we've never seen before.
51
152179
2238
從而解決我們先前還不曾見過的難題。
02:46
Percy Spencer was a physicist
working on radar during World War II,
working on radar during World War II,
52
154808
4411
波西‧史賓塞是二次世界大戰期間，
從事雷達研究的物理學家，
從事雷達研究的物理學家，
02:51
when he noticed the magnetron
was melting his chocolate bar.
was melting his chocolate bar.
53
159243
3013
當他注意到磁控管不斷融化
他的巧克力棒時，
他的巧克力棒時，
02:54
He was able to connect his understanding
of electromagnetic radiation
of electromagnetic radiation
54
162970
3295
他能夠把他對電磁波的認知
02:58
with his knowledge of cooking
55
166289
1484
與烹飪的知識做結合，
02:59
in order to invent -- any guesses? --
the microwave oven.
the microwave oven.
56
167797
3258
因此發明了--各位猜猜是什麽？
微波爐。
微波爐。
03:03
Now, this is a particularly remarkable
example of creativity.
example of creativity.
57
171444
3073
這是個特別傑出的創新例子
03:06
But this sort of cross-pollination
happens for each of us in small ways
happens for each of us in small ways
58
174541
3664
但是這種跨領域的碰撞，
每天在我們的周遭會上演好幾千回。
03:10
thousands of times per day.
59
178229
1828
03:12
Machines cannot compete with us
60
180501
1661
在解決新的棘手問題方面
機器無法與我們媲美，
機器無法與我們媲美，
03:14
when it comes to tackling
novel situations,
novel situations,
61
182186
2251
03:16
and this puts a fundamental limit
on the human tasks
on the human tasks
62
184461
3117
而這使機器自動化取代人工
03:19
that machines will automate.
63
187602
1717
受到了限制。
03:22
So what does this mean
for the future of work?
for the future of work?
64
190041
2405
那麽這對未來的工作意味著什麽呢？
03:24
The future state of any single job lies
in the answer to a single question:
in the answer to a single question:
65
192804
4532
未來工作的狀態完全取決於一個問題：
03:29
To what extent is that job reducible
to frequent, high-volume tasks,
to frequent, high-volume tasks,
66
197360
4981
「該工作有多少程度可以縮減成
經常性、高產量的任務，
經常性、高產量的任務，
03:34
and to what extent does it involve
tackling novel situations?
tackling novel situations?
67
202365
3253
以及有多少程度是在解決新的棘手問題？」
03:37
On frequent, high-volume tasks,
machines are getting smarter and smarter.
machines are getting smarter and smarter.
68
205975
4035
對於那些頻繁，大批量的任務，
機器變得越來越聰明。
機器變得越來越聰明。
03:42
Today they grade essays.
They diagnose certain diseases.
They diagnose certain diseases.
69
210034
2714
今天它們能給作文打分數、
診斷特定疾病，
診斷特定疾病，
03:44
Over coming years,
they're going to conduct our audits,
they're going to conduct our audits,
70
212772
3157
過了幾年後，它們將可以進行審計、
03:47
and they're going to read boilerplate
from legal contracts.
from legal contracts.
71
215953
2967
從法律合約中解讀法律語言。
03:50
Accountants and lawyers are still needed.
72
218944
1997
盡管會計師和律師還是需要的
03:52
They're going to be needed
for complex tax structuring,
for complex tax structuring,
73
220965
2682
但僅能研究覆雜的稅務結構及
無例可循的法律問題，
無例可循的法律問題，
03:55
for pathbreaking litigation.
74
223671
1357
03:57
But machines will shrink their ranks
75
225052
1717
不過機器將會減少他們的就業機會，
增加就業難度。
增加就業難度。
03:58
and make these jobs harder to come by.
76
226793
1872
04:00
Now, as mentioned,
77
228689
1151
如同我說過的：
04:01
machines are not making progress
on novel situations.
on novel situations.
78
229864
2949
機器在處理複雜的新情境上
沒有進步！
沒有進步！
04:04
The copy behind a marketing campaign
needs to grab consumers' attention.
needs to grab consumers' attention.
79
232837
3457
行銷推案的文宣必須擄獲消費者的青睞，
04:08
It has to stand out from the crowd.
80
236318
1715
它必須脫俗出眾。
04:10
Business strategy means
finding gaps in the market,
finding gaps in the market,
81
238057
2444
商業策略必須在市場上找到一些
04:12
things that nobody else is doing.
82
240525
1756
其它人還沒開始做的領域。
04:14
It will be humans that are creating
the copy behind our marketing campaigns,
the copy behind our marketing campaigns,
83
242305
4118
人類才是營銷文案的創造者，
04:18
and it will be humans that are developing
our business strategy.
our business strategy.
84
246447
3517
人類才是商業戰略的拓展人
04:21
So Yahli, whatever you decide to do,
85
249988
2817
所以，雅莉，不管妳決定要做什麼，
04:24
let every day bring you a new challenge.
86
252829
2361
讓每一天帶給妳新的挑戰，
04:27
If it does, then you will stay
ahead of the machines.
ahead of the machines.
87
255587
2809
如果是這樣，那麼妳將永遠領先機器一步。
04:31
Thank you.
88
259126
1176
謝謝大家！
04:32
(Applause)
89
260326
3104
（掌聲）
ABOUT THE SPEAKER
Anthony Goldbloom - Machine learning expertAnthony Goldbloom crowdsources solutions to difficult problems using machine learning.
Why you should listen
Anthony Goldbloom is the co-founder and CEO of Kaggle. Kaggle hosts machine learning competitions, where data scientists download data and upload solutions to difficult problems. Kaggle has a community of over 600,000 data scientists and has worked with companies ranging Facebook to GE on problems ranging from predicting friendships to flight arrival times.
Before Kaggle, Anthony worked as an econometrician at the Reserve Bank of Australia, and before that the Australian Treasury. In 2011 and 2012, Forbes named Anthony one of the 30 under 30 in technology; in 2013 the MIT Tech Review named him one of top 35 innovators under the age of 35, and the University of Melbourne awarded him an Alumni of Distinction Award. He holds a first call honors degree in Econometrics from the University of Melbourne.
Anthony Goldbloom | Speaker | TED.com