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

597,885 views ใƒป 2014-12-16

TED


์•„๋ž˜ ์˜๋ฌธ์ž๋ง‰์„ ๋”๋ธ”ํด๋ฆญํ•˜์‹œ๋ฉด ์˜์ƒ์ด ์žฌ์ƒ๋ฉ๋‹ˆ๋‹ค.

๋ฒˆ์—ญ: Gemma Lee ๊ฒ€ํ† : Mingyu Choi
00:12
It used to be that if you wanted to get a computer to do something new,
0
12880
4013
์˜ˆ์ „์—๋Š” ์ปดํ“จํ„ฐ๊ฐ€ ์ƒˆ๋กœ์šด ์ผ์„ ํ•˜๊ฒŒ ๋งŒ๋“ค๋ ค๋ฉด
00:16
you would have to program it.
1
16893
1554
ํ”„๋กœ๊ทธ๋žจ์„ ์งœ์•ผ ํ–ˆ์Šต๋‹ˆ๋‹ค.
00:18
Now, programming, for those of you here that haven't done it yourself,
2
18447
3411
ํ”„๋กœ๊ทธ๋ž˜๋ฐ์„ ํ•ด๋ณธ ์ ์ด ์—†๋Š” ๋ถ„๋“ค์€
00:21
requires laying out in excruciating detail
3
21858
3502
๋ชฉํ‘œ๋ฅผ ๋‹ฌ์„ฑํ•˜๊ธฐ ์œ„ํ•ด์„œ
์ปดํ“จํ„ฐ๊ฐ€ ํ•ด์•ผ ํ•  ์ผ์„ ๋งค ๋‹จ๊ณ„๋งˆ๋‹ค
00:25
every single step that you want the computer to do
4
25360
3367
00:28
in order to achieve your goal.
5
28727
2362
๊ณ ํ†ต์Šค๋Ÿฌ์šธ์ •๋„๋กœ ์„ธ์„ธํ•˜๊ฒŒ ์„ค์ •ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
00:31
Now, if you want to do something that you don't know how to do yourself,
6
31089
3496
์ž, ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๋ชจ๋ฅด๋Š” ์ผ์„ ์—ฌ๋Ÿฌ๋ถ„์ด ํ•˜๊ณ  ์‹ถ๋‹ค๋ฉด
00:34
then this is going to be a great challenge.
7
34585
2063
๊ทธ๊ฑด ์•„์ฃผ ์ปค๋‹ค๋ž€ ๋„์ „์ด ๋˜๊ฒ ์ฃ .
00:36
So this was the challenge faced by this man, Arthur Samuel.
8
36648
3483
์ด๊ฒƒ์ด ์•„์„œ ์‚ฌ๋ฌด์—˜์ด ์ง๋ฉดํ•œ ๋„์ „์ด์—ˆ์Šต๋‹ˆ๋‹ค.
00:40
In 1956, he wanted to get this computer
9
40131
4077
1956๋…„ ๊ทธ๋Š” ์ปดํ“จํ„ฐ๊ฐ€
00:44
to be able to beat him at checkers.
10
44208
2340
์„œ์–‘์žฅ๊ธฐ์—์„œ ๊ทธ๋ฅผ ์ด๊ธฐ๊ธฐ๋ฅผ ๋ฐ”๋žฌ์Šต๋‹ˆ๋‹ค.
00:46
How can you write a program,
11
46548
2040
ํ”„๋กœ๊ทธ๋žจ์„ ์–ด๋–ป๊ฒŒ ์งค ์ˆ˜ ์žˆ์„๊นŒ์š”?
00:48
lay out in excruciating detail, how to be better than you at checkers?
12
48588
3806
์„œ์–‘์žฅ๊ธฐ์—์„œ ์—ฌ๋Ÿฌ๋ถ„๋ณด๋‹ค ์ž˜ํ•˜๋„๋ก ๊ทน์‹ฌํ•œ ์„ธ๋ถ€์‚ฌํ•ญ์„ ์“ธ ์ˆ˜ ์žˆ์„๊นŒ์š”?
00:52
So he came up with an idea:
13
52394
1722
๊ทธ๋Š” ์ƒˆ๋กœ์šด ์ƒ๊ฐ์„ ํ–ˆ์Šต๋‹ˆ๋‹ค.
00:54
he had the computer play against itself thousands of times
14
54116
3724
์ปดํ“จํ„ฐ๊ฐ€ ์Šค์Šค๋กœ์™€ ์ˆ˜์ฒœ ๋ฒˆ์˜ ์„œ์–‘์žฅ๊ธฐ๋ฅผ ๋‘๊ฒŒ ํ•ด์„œ
00:57
and learn how to play checkers.
15
57840
2524
์„œ์–‘์žฅ๊ธฐ ๋‘๋Š” ๋ฒ•์„ ๋ฐฐ์šฐ๊ฒŒ ํ–ˆ์Šต๋‹ˆ๋‹ค.
01:00
And indeed it worked, and in fact, by 1962,
16
60364
3180
๊ทธ ๋ฐฉ๋ฒ•์€ ์ •๋ง ํšจ๊ณผ๊ฐ€ ์žˆ์—ˆ๊ณ  ์‚ฌ์‹ค 1962๋…„์—
01:03
this computer had beaten the Connecticut state champion.
17
63544
4017
์ด ์ปดํ“จํ„ฐ๋Š” ์ฝ”๋„คํ‹ฐ์ปท ์ฃผ์˜ ์šฐ์Šน์ž๋ฅผ ๋ฌด์ฐ”๋ €์Šต๋‹ˆ๋‹ค.
01:07
So Arthur Samuel was the father of machine learning,
18
67561
2973
๊ทธ๋ž˜์„œ ์•„์„œ ์‚ฌ๋ฌด์—˜์€ ๊ธฐ๊ณ„ ํ•™์Šต์˜ ์•„๋ฒ„์ง€์˜€๊ณ 
01:10
and I have a great debt to him,
19
70534
1717
์ €๋Š” ๊ทธ๋ถ„๊ป˜ ํฐ ๋นš์„ ์ง€๊ณ  ์žˆ์ฃ .
01:12
because I am a machine learning practitioner.
20
72251
2763
์™œ๋ƒํ•˜๋ฉด ์ €๋Š” ๊ธฐ๊ณ„ ํ•™์Šต ๊ธฐ์ˆ ์ž์ด๋‹ˆ๊นŒ์š”.
01:15
I was the president of Kaggle,
21
75014
1465
์ €๋Š” ์บ๊ธ€์˜ ํšŒ์žฅ์ธ๋ฐ
01:16
a community of over 200,000 machine learning practictioners.
22
76479
3388
์บ๊ธ€์€ 20๋งŒ ๋ช…์ด ๋„˜๋Š” ๊ธฐ๊ณ„ ํ•™์Šต ๊ธฐ์ˆ ์ž๋“ค์˜ ๋™ํ˜ธํšŒ์ž…๋‹ˆ๋‹ค.
01:19
Kaggle puts up competitions
23
79867
2058
์บ๊ธ€์€ ์ด์ „๊นŒ์ง€ ํ’€์ง€ ๋ชปํ–ˆ๋˜ ๋ฌธ์ œ๋ฅผ
01:21
to try and get them to solve previously unsolved problems,
24
81925
3708
ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•œ ๋Œ€ํšŒ๋ฅผ ์ฃผ์ตœํ•˜๋Š”๋ฐ
01:25
and it's been successful hundreds of times.
25
85633
3837
์ˆ˜๋ฐฑ๋ฒˆ ์„ฑ๊ณตํ–ˆ์Šต๋‹ˆ๋‹ค.
01:29
So from this vantage point, I was able to find out
26
89470
2470
๊ทธ๋ž˜์„œ ์ด๋Ÿฐ ์œ ๋ฆฌํ•œ ์‹œ์ ์—์„œ ์ €๋Š” ๊ธฐ๊ณ„ ํ•™์Šต์ด
01:31
a lot about what machine learning can do in the past, can do today,
27
91940
3950
๊ณผ๊ฑฐ์™€ ํ˜„์žฌ์— ํ•  ์ˆ˜ ์žˆ๋Š” ์ผ๊ณผ ๋ฏธ๋ž˜์— ํ•  ์ˆ˜ ์žˆ๋Š” ์ผ์„
01:35
and what it could do in the future.
28
95890
2362
๋งŽ์ด ์•Œ ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค.
01:38
Perhaps the first big success of machine learning commercially was Google.
29
98252
4423
์•„๋งˆ๋„ ๊ธฐ๊ณ„ ํ•™์Šต์ด ์ƒ์—…์—์„œ ์ตœ์ดˆ๋กœ ๊ฐ€์žฅ ํฌ๊ฒŒ ์„ฑ๊ณตํ•œ ๊ฒƒ์€ ๊ตฌ๊ธ€์ด์—ˆ์Šต๋‹ˆ๋‹ค.
01:42
Google showed that it is possible to find information
30
102675
3109
๊ตฌ๊ธ€์€ ์ปดํ“จํ„ฐ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์‚ฌ์šฉํ•ด์„œ
01:45
by using a computer algorithm,
31
105784
1752
์ •๋ณด๋ฅผ ์ฐพ์„ ์ˆ˜ ์žˆ์Œ์„ ๋ณด์—ฌ์คฌ๋Š”๋ฐ
01:47
and this algorithm is based on machine learning.
32
107536
2901
์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๊ธฐ๊ณ„ ํ•™์Šต์„ ๋ฐ”ํƒ•์œผ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
01:50
Since that time, there have been many commercial successes of machine learning.
33
110437
3886
๊ทธ๋•Œ๋ถ€ํ„ฐ ๊ธฐ๊ณ„ ํ•™์Šต์˜ ์ƒ์—…์  ์„ฑ๊ณต์ด ๋งŽ์ด ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค.
01:54
Companies like Amazon and Netflix
34
114323
1837
์•„๋งˆ์กด๊ณผ ๋„ทํ”Œ๋ฆญ์Šค ๊ฐ™์€ ํšŒ์‚ฌ๋“ค์€
01:56
use machine learning to suggest products that you might like to buy,
35
116160
3716
๊ธฐ๊ณ„ ํ•™์Šต์„ ์ด์šฉํ•ด์„œ ์—ฌ๋Ÿฌ๋ถ„์ด ์‚ฌ๊ณ  ์‹ถ์€ ์ƒํ’ˆ์ด๋‚˜
01:59
movies that you might like to watch.
36
119876
2020
๋ณด๊ณ  ์‹ถ์€ ์˜ํ™”๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค.
02:01
Sometimes, it's almost creepy.
37
121896
1807
๋•Œ๋กœ๋Š” ์˜ค์‹นํ•  ์ง€๊ฒฝ์ด์ฃ .
02:03
Companies like LinkedIn and Facebook
38
123703
1954
๋งํฌ๋“œ์ธ๊ณผ ํŽ˜์ด์Šค๋ถ ๊ฐ™์€ ํšŒ์‚ฌ๋“ค์€
02:05
sometimes will tell you about who your friends might be
39
125657
2594
๋ˆ„๊ฐ€ ์—ฌ๋Ÿฌ๋ถ„์˜ ์นœ๊ตฌ์ธ์ง€๋ฅผ ๋งํ•ด์ค„ ๊ฒ๋‹ˆ๋‹ค.
02:08
and you have no idea how it did it,
40
128251
1977
์–ด๋–ป๊ฒŒ ๊ทธ๋ ‡๊ฒŒ ํ•˜๋Š”์ง€ ์—ฌ๋Ÿฌ๋ถ„์€ ๋ชจ๋ฆ…๋‹ˆ๋‹ค.
02:10
and this is because it's using the power of machine learning.
41
130228
2967
๊ทธ ์ด์œ ๋Š” ๊ธฐ๊ณ„ ํ•™์Šต์˜ ํž˜์„ ์ด์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์ด์ฃ .
02:13
These are algorithms that have learned how to do this from data
42
133195
2957
์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์†์œผ๋กœ ์“ด ํ”„๋กœ๊ทธ๋žจ ๋ณด๋‹ค๋Š”
02:16
rather than being programmed by hand.
43
136152
3247
๋ฐ์ดํ„ฐ์—์„œ ๋ฐฐ์› ์Šต๋‹ˆ๋‹ค.
02:19
This is also how IBM was successful
44
139399
2478
IBM์ด ์™“์Šจ์„ ์ด์šฉํ•ด "์ œํผ๋””"์—์„œ
02:21
in getting Watson to beat the two world champions at "Jeopardy,"
45
141877
3862
2๋ช…์˜ ์„ธ๊ณ„ ์ฑ”ํ”ผ์–ธ์„ ์„ฑ๊ณต์ ์œผ๋กœ ๋ฌด์ฐŒ๋ฅธ ์ด์œ ์ด๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค.
02:25
answering incredibly subtle and complex questions like this one.
46
145739
3225
์ด์ฒ˜๋Ÿผ ์•„์ฃผ ๋ฏธ๋ฌ˜ํ•˜๊ณ  ๋ณต์žกํ•œ ์งˆ๋ฌธ์— ๋Œ€๋‹ตํ–ˆ์ฃ .
02:28
["The ancient 'Lion of Nimrud' went missing from this city's national museum in 2003 (along with a lot of other stuff)"]
47
148964
2835
["๊ณ ๋Œ€ '๋‹ˆ๋ฌด๋ฅด๋“œ์˜ ์‚ฌ์ž'๊ฐ€ 2003๋…„ ์ด ๋„์‹œ์˜ ๋ฐ•๋ฌผ๊ด€์—์„œ ์‚ฌ๋ผ์กŒ์Šต๋‹ˆ๋‹ค."
02:31
This is also why we are now able to see the first self-driving cars.
48
151799
3235
์ด ๋•Œ๋ฌธ์— ์šฐ๋ฆฌ๋Š” ์ด์ œ ์ตœ์ดˆ์˜ ๋ฌด์ธ ์ž๋™์ฐจ๋ฅผ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
02:35
If you want to be able to tell the difference between, say,
49
155034
2822
๋‚˜๋ฌด์™€ ๋ณดํ–‰์ž์˜ ์ฐจ์ด์ , ๊ทธ๊ฒŒ ์•„์ฃผ ์ค‘์š”ํ•œ๋ฐ
02:37
a tree and a pedestrian, well, that's pretty important.
50
157856
2632
๊ทธ๊ฑธ ๊ตฌ๋ณ„ํ•˜๊ณ  ์‹ถ์„ ๋•Œ
02:40
We don't know how to write those programs by hand,
51
160488
2587
์†์œผ๋กœ ํ”„๋กœ๊ทธ๋žจ์„ ์–ด๋–ป๊ฒŒ ์จ์•ผํ• ์ง€ ๋ชจ๋ฅด์ง€๋งŒ
02:43
but with machine learning, this is now possible.
52
163075
2997
๊ธฐ๊ณ„ ํ•™์Šต์œผ๋กœ ์ด์ œ ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค.
02:46
And in fact, this car has driven over a million miles
53
166072
2608
์‚ฌ์‹ค ์ด ์ž๋™์ฐจ๋Š” ์ผ๋ฐ˜ ๋„๋กœ์—์„œ ์‚ฌ๊ณ  ์—†์ด
02:48
without any accidents on regular roads.
54
168680
3506
์ˆ˜๋ฐฑ๋งŒ km๋ฅผ ๋‹ฌ๋ ธ์Šต๋‹ˆ๋‹ค.
02:52
So we now know that computers can learn,
55
172196
3914
์ด์ œ ์šฐ๋ฆฌ๋Š” ์ปดํ“จํ„ฐ๊ฐ€ ๋ฐฐ์šธ ์ˆ˜ ์žˆ๊ณ 
02:56
and computers can learn to do things
56
176110
1900
์šฐ๋ฆฌ๊ฐ€ ์‹ค์ œ๋กœ ํ•˜๋Š” ๋ฐฉ๋ฒ•์„
02:58
that we actually sometimes don't know how to do ourselves,
57
178010
2838
๋ชจ๋ฅด๋Š” ์ผ๋„ ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋ฐฐ์šธ ์ˆ˜ ์žˆ์Œ์„ ์••๋‹ˆ๋‹ค.
03:00
or maybe can do them better than us.
58
180848
2885
์–ด์ฉŒ๋ฉด ์šฐ๋ฆฌ๋ณด๋‹ค ์ž˜ํ•  ์ˆ˜๋„ ์žˆ์–ด์š”.
03:03
One of the most amazing examples I've seen of machine learning
59
183733
4195
๊ธฐ๊ณ„ ํ•™์Šต์—์„œ ๊ฐ€์žฅ ๋†€๋ผ์šด ์˜ˆ๊ฐ€
03:07
happened on a project that I ran at Kaggle
60
187928
2392
์ œ๊ฐ€ ์บ๊ธ€์—์„œ ํ•˜๋Š” ํ”„๋กœ์ ํŠธ์—์„œ ์ผ์–ด๋‚ฌ์Šต๋‹ˆ๋‹ค.
03:10
where a team run by a guy called Geoffrey Hinton
61
190320
3591
ํ† ๋ก ํ†  ๋Œ€ํ•™ ์ถœ์‹ ์˜ ์ œํ”„๋ฆฌ ํžŒํŠผ์ด
03:13
from the University of Toronto
62
193911
1552
์ด๋„๋Š” ํŒ€์€
03:15
won a competition for automatic drug discovery.
63
195463
2677
์ž๋™ ์‹ ์•ฝ ๊ฐœ๋ฐœ์„ ์œ„ํ•œ ๋Œ€ํšŒ์—์„œ ์ด๊ฒผ์Šต๋‹ˆ๋‹ค.
03:18
Now, what was extraordinary here is not just that they beat
64
198140
2847
์ž, ์—ฌ๊ธฐ์„œ ๋†€๋ผ์šด ์‚ฌ์‹ค์€ ๊ทธ๋“ค์ด ๋จธํฌ ๋˜๋Š” ๊ตญ์ œ ํ•™ํšŒ๊ฐ€
03:20
all of the algorithms developed by Merck or the international academic community,
65
200987
4013
๊ฐœ๋ฐœํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ด๊ฒผ์„ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ
03:25
but nobody on the team had any background in chemistry or biology or life sciences,
66
205000
5061
์–ด๋–ค ํŒ€์›๋„ ํ™”ํ•™, ์ƒ๋ฌผํ•™, ์ƒ๋ช…๊ณผํ•™์— ๊ด€ํ•œ ์ง€์‹์ด ์—†์—ˆ๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค.
03:30
and they did it in two weeks.
67
210061
2169
๊ทธ๋“ค์€ 2์ฃผ์•ˆ์— ์™„์„ฑํ–ˆ์ฃ .
03:32
How did they do this?
68
212230
1381
์–ด๋–ป๊ฒŒ ํ–ˆ์„๊นŒ์š”?
03:34
They used an extraordinary algorithm called deep learning.
69
214421
2921
๊ทธ๋“ค์€ ์‹ฌํ™” ํ•™์Šต์ด๋ผ๋Š” ๋†€๋ผ์šด ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.
03:37
So important was this that in fact the success was covered
70
217342
2949
์ด๊ฒƒ์€ ์‚ฌ์‹ค ์•„์ฃผ ์ค‘์š”ํ•ด์„œ ๋ช‡ ์ฃผ๊ฐ€ ์ง€๋‚œ ๋’ค
03:40
in The New York Times in a front page article a few weeks later.
71
220291
3121
๋‰ด์š• ํƒ€์ž„์ฆˆ์—์„œ ์•ž๋ฉด ๊ธฐ์‚ฌ๋กœ ๋‹ค๋ค˜์Šต๋‹ˆ๋‹ค.
03:43
This is Geoffrey Hinton here on the left-hand side.
72
223412
2735
์™ผ์ชฝ์ด ์ œํ”„๋ฆฌ ํžŒํŠผ์ž…๋‹ˆ๋‹ค.
03:46
Deep learning is an algorithm inspired by how the human brain works,
73
226147
4341
์‹ฌํ™” ํ•™์Šต์€ ์‚ฌ๋žŒ์˜ ๋‡Œ๊ฐ€ ์ž‘์šฉํ•˜๋Š” ๋ฐฉ์‹์— ์˜๊ฐ์„ ๋ฐ›์•„์„œ ๋งŒ๋“ 
03:50
and as a result it's an algorithm
74
230488
1812
์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ๊ทธ ๊ฒฐ๊ณผ
03:52
which has no theoretical limitations on what it can do.
75
232300
3841
ํ•  ์ˆ˜ ์žˆ๋Š” ์ผ์— ๋Œ€ํ•œ ์ด๋ก ์  ํ•œ๊ณ„๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค.
03:56
The more data you give it and the more computation time you give it,
76
236141
2823
๋” ๋งŽ์€ ๋ฐ์ดํ„ฐ์™€ ๋” ๋งŽ์€ ๊ณ„์‚ฐ ์‹œ๊ฐ„์„ ์ค„์ˆ˜๋ก
03:58
the better it gets.
77
238964
1312
๋” ์ข‹์€ ๊ฒฐ๊ณผ๋ฅผ ๋ƒ…๋‹ˆ๋‹ค.
04:00
The New York Times also showed in this article
78
240276
2339
๋‰ด์š• ํƒ€์ž„์ฆˆ๋Š” ์ด ๊ธฐ์‚ฌ์—์„œ
04:02
another extraordinary result of deep learning
79
242615
2242
์‹ฌํ™” ํ•™์Šต์˜ ๋˜๋‹ค๋ฅธ ๋†€๋ผ์šด ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์คฌ๋Š”๋ฐ
04:04
which I'm going to show you now.
80
244857
2712
์—ฌ๋Ÿฌ๋ถ„๊ป˜ ๋ณด์—ฌ๋“œ๋ฆฌ์ฃ .
04:07
It shows that computers can listen and understand.
81
247569
4941
์ปดํ“จํ„ฐ๊ฐ€ ๋“ฃ๊ณ  ์ดํ•ดํ•  ์ˆ˜ ์žˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
04:12
(Video) Richard Rashid: Now, the last step
82
252510
2711
(์˜์ƒ) ๋ฆฌ์ฑ ๋“œ ๋ผ์‹œ๋“œ: ์ œ๊ฐ€ ์ด ๊ณผ์ •์—์„œ
04:15
that I want to be able to take in this process
83
255221
3025
๋งˆ์ง€๋ง‰์œผ๋กœ ๋ณด์—ฌ๋“œ๋ฆด ๋‹จ๊ณ„๋Š”
04:18
is to actually speak to you in Chinese.
84
258246
4715
์‹ค์ œ ์ค‘๊ตญ์–ด๋กœ ๋งํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
04:22
Now the key thing there is,
85
262961
2635
์ค‘์š”ํ•œ ์ ์€
04:25
we've been able to take a large amount of information from many Chinese speakers
86
265596
5002
๋งŽ์€ ์ค‘๊ตญ์ธ๋“ค๋กœ๋ถ€ํ„ฐ ์—„์ฒญ๋‚œ ์–‘์˜ ์ •๋ณด๋ฅผ ๋ชจ์„ ์ˆ˜ ์žˆ์—ˆ๊ณ 
04:30
and produce a text-to-speech system
87
270598
2530
๊ธ€์ž๋ฅผ ์Œ์„ฑ์œผ๋กœ ๋ฐ”๊พธ๋Š” ์‹œ์Šคํ…œ์„ ๋งŒ๋“ค์–ด
04:33
that takes Chinese text and converts it into Chinese language,
88
273128
4673
์ค‘๊ตญ ๊ธ€์ž๋ฅผ ์ค‘๊ตญ ๋ง๋กœ ๋ณ€ํ™˜์‹œํ‚ค๊ณ 
04:37
and then we've taken an hour or so of my own voice
89
277801
4128
์ œ ๋ชฉ์†Œ๋ฆฌ๋ฅผ ํ•œ ์‹œ๊ฐ„ ์ •๋„ ๋…น์Œํ•ด์„œ
04:41
and we've used that to modulate
90
281929
1891
ํ‘œ์ค€ ๋ฌธ์ž - ์Œ์„ฑ ๋ณ€ํ™˜ ์‹œ์Šคํ…œ์„ ์กฐ์ ˆํ•ด์„œ
04:43
the standard text-to-speech system so that it would sound like me.
91
283820
4544
์ œ ๋ชฉ์†Œ๋ฆฌ์ฒ˜๋Ÿผ ๋‚˜๋„๋ก ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.
04:48
Again, the result's not perfect.
92
288364
2540
์—ญ์‹œ ๊ฒฐ๊ณผ๋Š” ์™„๋ฒฝํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
04:50
There are in fact quite a few errors.
93
290904
2648
์‚ฌ์‹ค ์˜ค๋ฅ˜๊ฐ€ ์ƒ๋‹นํžˆ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค.
04:53
(In Chinese)
94
293552
2484
(์ค‘๊ตญ์–ด)
04:56
(Applause)
95
296036
3367
(๋ฐ•์ˆ˜)
05:01
There's much work to be done in this area.
96
301446
3576
์•„์ง ๋งŽ์€ ์ž‘์—…์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
05:05
(In Chinese)
97
305022
3645
(์ค‘๊ตญ์–ด)
05:08
(Applause)
98
308667
3433
(๋ฐ•์ˆ˜)
05:13
Jeremy Howard: Well, that was at a machine learning conference in China.
99
313345
3399
์ œ๋ ˆ๋ฏธ ํ•˜์›Œ๋“œ : ์ค‘๊ตญ์—์„œ ์—ด๋ฆฐ ๊ธฐ๊ณ„ ํ•™์Šต ํšŒ์˜์˜€์Šต๋‹ˆ๋‹ค.
05:16
It's not often, actually, at academic conferences
100
316744
2370
ํ•™์ˆ  ํšŒ์˜์—์„œ ์‹ค์ œ๋กœ
05:19
that you do hear spontaneous applause,
101
319114
1897
์ฆ‰ํฅ์ ์ธ ๋ฐ•์ˆ˜๋ฅผ ๋“ฃ๊ธฐ๋Š” ์‰ฝ์ง€ ์•Š์ฃ .
05:21
although of course sometimes at TEDx conferences, feel free.
102
321011
3676
๊ทธ๋ž˜๋„ TEDx ํšŒ์˜์—์„œ๋Š” ์ž์œ ๋กญ๊ฒŒ ํ•˜์„ธ์š”.
05:24
Everything you saw there was happening with deep learning.
103
324687
2795
๊ฑฐ๊ธฐ์„œ ๋ณธ ๋ชจ๋“  ๊ฒƒ์ด ์‹ฌํ™” ํ•™์Šต์œผ๋กœ ์ผ์–ด๋‚ฌ์Šต๋‹ˆ๋‹ค.
05:27
(Applause) Thank you.
104
327482
1525
(๋ฐ•์ˆ˜) ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
05:29
The transcription in English was deep learning.
105
329007
2282
์˜์–ด๋กœ ์˜ฎ๊ฒจ์“ฐ๊ธฐ๋Š” ์‹ฌํ™” ํ•™์Šต์ด์—ˆ์ฃ .
05:31
The translation to Chinese and the text in the top right, deep learning,
106
331289
3412
์ค‘๊ตญ์–ด ๋ฒˆ์—ญ๊ณผ ์˜ค๋ฅธ์ชฝ ์œ„์˜ ๊ธ€์ž๋„ ์‹ฌํ™” ํ•™์Šต์ด์—ˆ๊ณ 
05:34
and the construction of the voice was deep learning as well.
107
334701
3307
๋ชฉ์†Œ๋ฆฌ๋กœ ์žฌ์ƒํ•˜๋Š” ๊ฒƒ ์—ญ์‹œ ์‹ฌํ™” ํ•™์Šต์ด์—ˆ์Šต๋‹ˆ๋‹ค.
05:38
So deep learning is this extraordinary thing.
108
338008
3234
๊ทธ๋ž˜์„œ ์‹ฌํ™” ํ•™์Šต์€ ๋†€๋ผ์šด ๊ฒƒ์ž…๋‹ˆ๋‹ค.
05:41
It's a single algorithm that can seem to do almost anything,
109
341242
3099
ํ•˜๋‚˜์˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ธ๋ฐ ๊ฑฐ์˜ ๋ชจ๋“  ์ผ์„ ํ•  ์ˆ˜ ์žˆ์–ด ๋ณด์ž…๋‹ˆ๋‹ค.
05:44
and I discovered that a year earlier, it had also learned to see.
110
344341
3111
์ œ๊ฐ€ 1๋…„ ์ „์— ๋ฐœ๊ฒฌํ–ˆ๋Š”๋ฐ ๋ณด๋Š” ๋ฒ•๋„ ๋ฐฐ์› ์Šต๋‹ˆ๋‹ค.
05:47
In this obscure competition from Germany
111
347452
2176
๋…์ผ์˜ ์• ๋งคํ•œ ๋Œ€ํšŒ์ธ
05:49
called the German Traffic Sign Recognition Benchmark,
112
349628
2597
๋…์ผ ๊ตํ†ต ์‹ ํ˜ธ ์ธ์‹ ์„ฑ๋Šฅํ‰๊ฐ€์—์„œ
05:52
deep learning had learned to recognize traffic signs like this one.
113
352225
3393
์‹ฌํ™” ํ•™์Šต์€ ์ด๋Ÿฐ ๊ตํ†ต ์‹ ํ˜ธ๋ฅผ ์ธ์‹ํ•˜๋Š” ๋ฒ•์„ ๋ฐฐ์› ์Šต๋‹ˆ๋‹ค.
05:55
Not only could it recognize the traffic signs
114
355618
2094
๊ตํ†ต ์‹ ํ˜ธ๋ฅผ ์ธ์‹ํ•  ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ
05:57
better than any other algorithm,
115
357712
1758
์–ด๋–ค ์•Œ๊ณ ๋ฆฌ์ฆ˜๋ณด๋‹ค ๋‚ซ๊ณ 
05:59
the leaderboard actually showed it was better than people,
116
359470
2719
์„ฑ์ ์ด ์‚ฌ๋žŒ๋ณด๋‹ค 2๋ฐฐ ์ •๋„
06:02
about twice as good as people.
117
362189
1852
๋‚˜์€ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์˜€์Šต๋‹ˆ๋‹ค.
06:04
So by 2011, we had the first example
118
364041
1996
2011๋…„ ์šฐ๋ฆฌ๋Š” ์‚ฌ๋žŒ๋ณด๋‹ค
06:06
of computers that can see better than people.
119
366037
3405
์ž˜ ๋ณผ ์ˆ˜ ์žˆ๋Š” ์ปดํ“จํ„ฐ์˜ ์ฒซ๋ฒˆ์งธ ์˜ˆ๋ฅผ ๊ฐ€์กŒ์Šต๋‹ˆ๋‹ค.
06:09
Since that time, a lot has happened.
120
369442
2049
๊ทธํ›„๋กœ ๋งŽ์€ ์ผ์ด ์ผ์–ด๋‚ฌ์ฃ .
06:11
In 2012, Google announced that they had a deep learning algorithm
121
371491
3514
2012๋…„ ๊ตฌ๊ธ€์€ ์‹ฌํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๋งŒ๋“ค์—ˆ๋‹ค๊ณ  ๋ฐœํ‘œํ–ˆ์Šต๋‹ˆ๋‹ค.
06:15
watch YouTube videos
122
375005
1415
์œ ํŠœ๋ธŒ ๋™์˜์ƒ์„ ๋ณด๊ณ 
06:16
and crunched the data on 16,000 computers for a month,
123
376420
3437
ํ•œ ๋‹ฌ์— 1๋งŒ6์ฒœ ๋Œ€์˜ ์ปดํ“จํ„ฐ ๋ฐ์ดํ„ฐ๋ฅผ ์ฒ˜๋ฆฌํ•ด์„œ
06:19
and the computer independently learned about concepts such as people and cats
124
379857
4361
์ปดํ“จํ„ฐ๋Š” ๊ทธ๋ƒฅ ๋™์˜์ƒ์„ ๋ณด๋Š” ๊ฒƒ๋งŒ์œผ๋กœ ์‚ฌ๋žŒ๊ณผ ๊ณ ์–‘์ด ๊ฐ™์€ ๊ฐœ๋…์„
06:24
just by watching the videos.
125
384218
1809
์Šค์Šค๋กœ ํ•™์Šตํ–ˆ์Šต๋‹ˆ๋‹ค.
06:26
This is much like the way that humans learn.
126
386027
2352
์‚ฌ๋žŒ์ด ๋ฐฐ์šฐ๋Š” ๋ฐฉ๋ฒ•๊ณผ ๋น„์Šทํ•˜์ฃ .
06:28
Humans don't learn by being told what they see,
127
388379
2740
์‚ฌ๋žŒ๋“ค์€ ๋ณด๋Š” ๊ฒƒ์„ ์•Œ๋ ค์ค˜์„œ ๋ฐฐ์šฐ๋Š” ๊ฒŒ ์•„๋‹ˆ๋ผ
06:31
but by learning for themselves what these things are.
128
391119
3331
๊ทธ๊ฒƒ์ด ๋ญ”์ง€ ์Šค์Šค๋กœ ๋ฐฐ์›๋‹ˆ๋‹ค.
06:34
Also in 2012, Geoffrey Hinton, who we saw earlier,
129
394450
3369
๋˜ํ•œ 2012๋…„ ์šฐ๋ฆฌ๊ฐ€ ์•ž์„œ ๋ดค๋˜ ์ œํ”„๋ฆฌ ํžŒํŠผ์€
06:37
won the very popular ImageNet competition,
130
397819
2858
์•„์ฃผ ์œ ๋ช…ํ•œ ์ด๋ฏธ์ง€๋„ท ๋Œ€ํšŒ์—์„œ ์šฐ์Šนํ–ˆ๋Š”๋ฐ
06:40
looking to try to figure out from one and a half million images
131
400677
4141
1๋ฐฑ๋งŒ 5์ฒœ์žฅ์˜ ์‚ฌ์ง„์„ ๋ณด๊ณ  ๊ทธ๊ฒŒ ์–ด๋–ค ์‚ฌ์ง„์ธ์ง€
06:44
what they're pictures of.
132
404818
1438
๋งž์ถ”๋Š” ๋‚ด์šฉ์ด์ฃ .
06:46
As of 2014, we're now down to a six percent error rate
133
406256
3533
2014๋…„ ์ด์ œ ์˜์ƒ ์ธ์‹์—์„œ 6%์˜ ์˜ค์ฐจ์œจ๊นŒ์ง€
06:49
in image recognition.
134
409789
1453
๋‚ด๋ ค๊ฐ”์Šต๋‹ˆ๋‹ค.
06:51
This is better than people, again.
135
411242
2026
์ด๊ฒƒ๋„ ์‚ฌ๋žŒ๋ณด๋‹ค ๋‚ซ์Šต๋‹ˆ๋‹ค.
06:53
So machines really are doing an extraordinarily good job of this,
136
413268
3769
๊ธฐ๊ณ„๋Š” ์ •๋ง ๋†€๋ผ์šธ๋งŒํผ ์ผ์„ ์ž˜ํ•˜๊ณ  ์žˆ๊ณ 
06:57
and it is now being used in industry.
137
417037
2269
์ด์ œ ์‚ฐ์—…์—์„œ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค.
06:59
For example, Google announced last year
138
419306
3042
์˜ˆ๋ฅผ ๋“ค์–ด, ๊ตฌ๊ธ€์€ ์ž‘๋…„์—
07:02
that they had mapped every single location in France in two hours,
139
422348
4585
ํ”„๋ž‘์Šค์˜ ๊ตฌ์„๊ตฌ์„์„ 2์‹œ๊ฐ„ ์•ˆ์— ์ง€๋„๋กœ ๋งŒ๋“ค์—ˆ๋‹ค๊ณ  ๋ฐœํ‘œํ–ˆ๋Š”๋ฐ
07:06
and the way they did it was that they fed street view images
140
426933
3447
๊ทธ๋“ค์ด ํ•œ ๋ฐฉ๋ฒ•์€ ๊ธธ๊ฑฐ๋ฆฌ์—์„œ ์ฐ์€ ์‚ฌ์ง„์„
07:10
into a deep learning algorithm to recognize and read street numbers.
141
430380
4319
์‹ฌํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ์ž…๋ ฅํ•ด์„œ ์ฃผ์†Œ๋ฅผ ์ธ์‹ํ•˜๊ณ  ์ฝ๊ฒŒ ํ–ˆ์Šต๋‹ˆ๋‹ค.
07:14
Imagine how long it would have taken before:
142
434699
2220
์ด์ „์—๋Š” ์–ผ๋งˆ๋‚˜ ์˜ค๋ž˜ ๊ฑธ๋ ธ์„์ง€ ์ƒ๊ฐํ•ด๋ณด์„ธ์š”.
07:16
dozens of people, many years.
143
436919
3355
์ˆ˜์‹ญ๋ช…์˜ ์‚ฌ๋žŒ๋“ค์ด ๋ช‡ ๋…„๋™์•ˆ ํ–ˆ๊ฒ ์ฃ .
07:20
This is also happening in China.
144
440274
1911
์ด๊ฒƒ์€ ์ค‘๊ตญ์—์„œ๋„ ์ผ์–ด๋‚˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
07:22
Baidu is kind of the Chinese Google, I guess,
145
442185
4036
๋ฐ”์ด๋‘๋Š” ์ค‘๊ตญํŒ ๊ตฌ๊ธ€์ด๋ผ๊ณ  ์ œ๊ฐ€ ์ถ”์ธกํ•˜๋Š”๋ฐ
07:26
and what you see here in the top left
146
446221
2283
์™ผ์ชฝ ์œ„์—์„œ ๋ณด๋Š” ๊ฒƒ์€
07:28
is an example of a picture that I uploaded to Baidu's deep learning system,
147
448504
3974
๋ฐ”์ด๋‘์˜ ์‹ฌํ™” ํ•™์Šต ์‹œ์Šคํ…œ์— ์ œ๊ฐ€ ์˜ฌ๋ฆฐ ์‚ฌ์ง„์˜ ์˜ˆ์ด๊ณ 
07:32
and underneath you can see that the system has understood what that picture is
148
452478
3769
๊ทธ ์•„๋ž˜์— ๊ทธ ์‚ฌ์ง„์ด ๋ญ”์ง€๋ฅผ ์‹œ์Šคํ…œ์ด ์ดํ•ดํ•˜๊ณ 
07:36
and found similar images.
149
456247
2236
๋น„์Šทํ•œ ์‚ฌ์ง„๋“ค์„ ์ฐพ์•„๋†“์€ ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์ฃ .
07:38
The similar images actually have similar backgrounds,
150
458483
2736
๋น„์Šทํ•œ ์‚ฌ์ง„๋“ค์€ ์‹ค์ œ๋กœ ๋น„์Šทํ•œ ๋ฐฐ๊ฒฝ๊ณผ
07:41
similar directions of the faces,
151
461219
1658
๋น„์Šทํ•œ ์–ผ๊ตด ๋ฐฉํ–ฅ์„ ๊ฐ–๊ณ  ์žˆ๊ณ 
07:42
even some with their tongue out.
152
462877
1788
ํ˜€๋ฅผ ๋‚ด๋ฏผ ๋ชจ์Šต๋„ ๋น„์Šทํ•˜์ฃ .
07:44
This is not clearly looking at the text of a web page.
153
464665
3030
์ด๊ฒƒ์€ ์›นํŽ˜์ด์ง€์˜ ๊ธ€์ž๋ฅผ ์ฐพ์€ ๊ฒŒ ์•„๋‹™๋‹ˆ๋‹ค.
07:47
All I uploaded was an image.
154
467695
1412
์ œ๊ฐ€ ์˜ฌ๋ฆฐ ๊ฒƒ์€ ์‚ฌ์ง„์ด์—ˆ์ฃ .
07:49
So we now have computers which really understand what they see
155
469107
4021
์ด์ œ ์ปดํ“จํ„ฐ๊ฐ€ ๋ณธ ๊ฒƒ์„ ์ •๋ง ์ดํ•ดํ•ด์„œ
07:53
and can therefore search databases
156
473128
1624
์ˆ˜์ฒœ๋งŒ ์žฅ์˜ ์‚ฌ์ง„์ด ๋“ 
07:54
of hundreds of millions of images in real time.
157
474752
3554
๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋ฅผ ์‹ค์‹œ๊ฐ„์œผ๋กœ ์ฐพ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
07:58
So what does it mean now that computers can see?
158
478306
3230
์ปดํ“จํ„ฐ๊ฐ€ ๋ณผ ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒŒ ๋ฌด์Šจ ์˜๋ฏธ์ผ๊นŒ์š”?
08:01
Well, it's not just that computers can see.
159
481536
2017
์ปดํ“จํ„ฐ๊ฐ€ ๋ณผ ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ๋งŒ์ด ์•„๋‹ˆ๋ผ
08:03
In fact, deep learning has done more than that.
160
483553
2069
์‚ฌ์‹ค ์‹ฌํ™” ํ•™์Šต์€ ๋” ๋งŽ์€ ์ผ์„ ํ–ˆ์Šต๋‹ˆ๋‹ค.
08:05
Complex, nuanced sentences like this one
161
485622
2948
์ด๋ ‡๊ฒŒ ๋ณต์žกํ•˜๊ณ  ๋ฏธ๋ฌ˜ํ•œ ๋ฌธ์žฅ์€
08:08
are now understandable with deep learning algorithms.
162
488570
2824
์ด์ œ ์‹ฌํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ์ดํ•ดํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
08:11
As you can see here,
163
491394
1303
์—ฌ๊ธฐ์„œ ๋ณด๋“ฏ์ด
08:12
this Stanford-based system showing the red dot at the top
164
492697
2768
์œ„์— ์žˆ๋Š” ๋นจ๊ฐ„์ ์„ ๋ณด์—ฌ์ฃผ๋Š” ์Šคํƒ ํฌ๋“œ์— ์žˆ๋Š” ์‹œ์Šคํ…œ์€
08:15
has figured out that this sentence is expressing negative sentiment.
165
495465
3919
์ด ๋ฌธ์žฅ์ด ๋ถ€์ •์ ์ธ ๋Š๋‚Œ์„ ํ‘œํ˜„ํ•˜๋Š” ๊ฒƒ์„ ์•Œ์•„๋ƒˆ์Šต๋‹ˆ๋‹ค.
08:19
Deep learning now in fact is near human performance
166
499384
3406
์‹ฌํ™” ํ•™์Šต์€ ์ด์ œ ์‚ฌ์‹ค ์‚ฌ๋žŒ์— ๊ฐ€๊น๊ฒŒ
08:22
at understanding what sentences are about and what it is saying about those things.
167
502802
5121
๋ฌธ์žฅ์„ ์ดํ•ดํ•˜๊ณ  ๊ทธ๊ฒŒ ์–ด๋–ค ๋ง์„ ํ•˜๋Š”์ง€ ์••๋‹ˆ๋‹ค.
08:27
Also, deep learning has been used to read Chinese,
168
507923
2728
์‹ฌํ™” ํ•™์Šต์€ ๋˜ํ•œ ์ค‘๊ตญ์–ด๋ฅผ ์ฝ๋Š”๋ฐ ์‚ฌ์šฉ๋˜์—ˆ๊ณ 
08:30
again at about native Chinese speaker level.
169
510651
3156
์ค‘๊ตญ์–ด ์›์–ด๋ฏผ ์ˆ˜์ค€์ž…๋‹ˆ๋‹ค.
08:33
This algorithm developed out of Switzerland
170
513807
2168
์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์Šค์œ„์Šค์—์„œ ๊ฐœ๋ฐœ๋˜์—ˆ๋Š”๋ฐ
08:35
by people, none of whom speak or understand any Chinese.
171
515975
3356
๊ฐœ๋ฐœ์ž ์ค‘ ์ค‘๊ตญ์–ด๋ฅผ ํ•  ์ˆ˜ ์žˆ๋Š” ์‚ฌ๋žŒ์ด ์•„๋ฌด๋„ ์—†์—ˆ์Šต๋‹ˆ๋‹ค.
08:39
As I say, using deep learning
172
519331
2051
์‹ฌํ™” ํ•™์Šต์„ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์€
08:41
is about the best system in the world for this,
173
521382
2219
์‚ฌ๋žŒ์˜ ์ดํ•ด์— ๋น„ํ•ด์„œ๋„
08:43
even compared to native human understanding.
174
523601
5117
์„ธ๊ณ„ ์ตœ๊ณ ์˜ ์‹œ์Šคํ…œ์— ๊ด€ํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
08:48
This is a system that we put together at my company
175
528718
2964
์ด๊ฒƒ์€ ์šฐ๋ฆฌ๊ฐ€ ํšŒ์‚ฌ์—์„œ
08:51
which shows putting all this stuff together.
176
531682
2046
๋ชจ๋“  ๊ฒƒ์„ ๋‹ค ํ†ตํ•ฉํ•ด์„œ ๋งŒ๋“  ์‹œ์Šคํ…œ์ž…๋‹ˆ๋‹ค.
08:53
These are pictures which have no text attached,
177
533728
2461
์ด๊ฒƒ๋“ค์€ ๊ธ€์ž๊ฐ€ ์—†๋Š” ์‚ฌ์ง„๋“ค๋กœ์„œ
08:56
and as I'm typing in here sentences,
178
536189
2352
์ œ๊ฐ€ ๋ฌธ์žฅ์„ ์ž…๋ ฅํ•˜๋ฉด
08:58
in real time it's understanding these pictures
179
538541
2969
์‹ค์‹œ๊ฐ„์œผ๋กœ ๊ทธ ์‚ฌ์ง„๋“ค์„ ์ดํ•ดํ•ด์„œ
09:01
and figuring out what they're about
180
541510
1679
๊ทธ๊ฒŒ ์–ด๋–ค ์‚ฌ์ง„์ธ์ง€ ์•Œ๊ณ 
09:03
and finding pictures that are similar to the text that I'm writing.
181
543189
3163
์ œ๊ฐ€ ์“ฐ๋Š” ๊ธ€์— ๋Œ€ํ•ด ๋น„์Šทํ•œ ์‚ฌ์ง„์„ ์ฐพ์•„์ค๋‹ˆ๋‹ค.
09:06
So you can see, it's actually understanding my sentences
182
546352
2756
๋ณด๋‹ค์‹œํ”ผ ์ œ๊ฐ€ ์“ด ๊ธ€์„ ์ดํ•ดํ•˜๊ณ 
09:09
and actually understanding these pictures.
183
549108
2224
์ด ์‚ฌ์ง„๋“ค์„ ์‹ค์ œ๋กœ ์ดํ•ดํ•ฉ๋‹ˆ๋‹ค.
09:11
I know that you've seen something like this on Google,
184
551332
2559
์—ฌ๋Ÿฌ๋ถ„์€ ๊ตฌ๊ธ€์—์„œ ์ด์™€ ๋น„์Šทํ•œ ๊ฒƒ์„ ๋ดค์„ ํ…๋ฐ
09:13
where you can type in things and it will show you pictures,
185
553891
2775
์—ฌ๋Ÿฌ๋ถ„์ด ๊ธ€์ž๋ฅผ ์ž…๋ ฅํ•˜๋ฉด ์‚ฌ์ง„์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
09:16
but actually what it's doing is it's searching the webpage for the text.
186
556666
3424
ํ•˜์ง€๋งŒ ์‹ค์ œ๋กœ๋Š” ๊ทธ ๊ธ€์ž๊ฐ€ ์žˆ๋Š” ์›นํŽ˜์ด์ง€๋ฅผ ์ฐพ๋Š” ๊ฑฐ์ฃ .
09:20
This is very different from actually understanding the images.
187
560090
3001
์ด๊ฒƒ์€ ์‚ฌ์ง„์„ ์‹ค์ œ๋กœ ์ดํ•ดํ•˜๋Š” ๊ฒƒ๊ณผ ์•„์ฃผ ๋‹ค๋ฆ…๋‹ˆ๋‹ค.
09:23
This is something that computers have only been able to do
188
563091
2752
์ด๊ฒƒ์€ ์ปดํ“จํ„ฐ๊ฐ€ ์ง€๋‚œ ๋ช‡ ๋‹ฌ๋™์•ˆ
09:25
for the first time in the last few months.
189
565843
3248
์ฒ˜์Œ์œผ๋กœ ํ•  ์ˆ˜ ์žˆ์—ˆ๋˜ ์ผ์ž…๋‹ˆ๋‹ค.
09:29
So we can see now that computers can not only see but they can also read,
190
569091
4091
์ด์ œ ์ปดํ“จํ„ฐ๋Š” ๋ณผ ์ˆ˜ ์žˆ์„ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์ฝ์„ ์ˆ˜๋„ ์žˆ๊ณ 
09:33
and, of course, we've shown that they can understand what they hear.
191
573182
3765
๋ฌผ๋ก  ๋“ค์€ ๊ฒƒ๋„ ์ดํ•ดํ•  ์ˆ˜ ์žˆ์Œ์„ ๋ดค์Šต๋‹ˆ๋‹ค.
09:36
Perhaps not surprising now that I'm going to tell you they can write.
192
576947
3442
์ปดํ“จํ„ฐ๊ฐ€ ์“ธ ์ค„ ์•ˆ๋‹ค๊ณ  ์–˜๊ธฐํ•ด๋„ ์ด์ œ๋Š” ๋†€๋ผ์ง€ ์•Š์œผ์‹ค ๊ฑฐ์—์š”.
09:40
Here is some text that I generated using a deep learning algorithm yesterday.
193
580389
4783
์ด๊ฒƒ์€ ์‹ฌํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์‚ฌ์šฉํ•ด์„œ ์–ด์ œ ์ œ๊ฐ€ ๋งŒ๋“  ๊ธ€์ž…๋‹ˆ๋‹ค.
09:45
And here is some text that an algorithm out of Stanford generated.
194
585172
3924
์ด๊ฒƒ์€ ์Šคํƒ ํฌ๋“œ์—์„œ ๋งŒ๋“  ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ๋งŒ๋“  ๊ธ€์ž…๋‹ˆ๋‹ค.
09:49
Each of these sentences was generated
195
589096
1764
์ด ๊ธ€์€ ๊ฐ๊ฐ์˜ ์‚ฌ์ง„์„
09:50
by a deep learning algorithm to describe each of those pictures.
196
590860
4249
์„ค๋ช…ํ•˜๊ธฐ ์œ„ํ•ด ์‹ฌํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.
09:55
This algorithm before has never seen a man in a black shirt playing a guitar.
197
595109
4472
์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๊ฒ€์€์ƒ‰ ์…”์ธ ๋ฅผ ์ž…๊ณ  ๊ธฐํƒ€๋ฅผ ์น˜๋Š” ๋‚จ์ž๋ฅผ ๋ณธ ์ ์ด ์—†์Šต๋‹ˆ๋‹ค.
09:59
It's seen a man before, it's seen black before,
198
599581
2220
๋‚จ์ž๋ฅผ ๋ณธ ์ ์ด ์žˆ๊ณ  ๊ฒ€์€ ์ƒ‰์„ ๋ณธ ์ ์ด ์žˆ๊ณ 
10:01
it's seen a guitar before,
199
601801
1599
๊ธฐํƒ€๋ฅผ ๋ณธ ์ ์€ ์žˆ์–ด์š”.
10:03
but it has independently generated this novel description of this picture.
200
603400
4294
๊ทธ๋Ÿฐ๋ฐ ์Šค์Šค๋กœ ์ด ์‚ฌ์ง„์„ ํ›Œ๋ฅญํ•˜๊ฒŒ ์„ค๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.
10:07
We're still not quite at human performance here, but we're close.
201
607694
3502
์•„์ง๋„ ์‚ฌ๋žŒ๋ณด๋‹ค๋Š” ๋ชปํ•˜์ง€๋งŒ ๊ฝค ๊ฐ€๊นŒ์ด ์™”์Šต๋‹ˆ๋‹ค.
10:11
In tests, humans prefer the computer-generated caption
202
611196
4068
์‹คํ—˜์—์„œ ์‚ฌ๋žŒ๋“ค์€ ์ปดํ“จํ„ฐ๊ฐ€ ๋งŒ๋“ค์–ด๋‚ธ ์บก์…˜์„
10:15
one out of four times.
203
615264
1527
4ํšŒ๋‹น 1ํšŒ ๊ผด๋กœ ์ข‹์•„ํ–ˆ์Šต๋‹ˆ๋‹ค.
10:16
Now this system is now only two weeks old,
204
616791
2064
์ด ์‹œ์Šคํ…œ์€ ์ด์ œ 2์ฃผ๊ฐ€ ๋˜์—ˆ๋Š”๋ฐ
10:18
so probably within the next year,
205
618855
1846
์•„๋งˆ๋„ ๋‚ด๋…„ ์•ˆ์œผ๋กœ
10:20
the computer algorithm will be well past human performance
206
620701
2801
์ง€๊ธˆ ์ง„ํ–‰๋˜๋Š” ์†๋„๋กœ ๋ด์„œ ์ปดํ“จํ„ฐ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด
10:23
at the rate things are going.
207
623502
1862
์‚ฌ๋žŒ์„ ์•ž์ง€๋ฅผ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
10:25
So computers can also write.
208
625364
3049
์ปดํ“จํ„ฐ๋Š” ์“ธ ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค.
10:28
So we put all this together and it leads to very exciting opportunities.
209
628413
3475
๊ทธ๋ž˜์„œ ์ด ๋ชจ๋“  ๊ธฐ๋Šฅ์„ ํ•ฉํ•˜๋ฉด ์•„์ฃผ ํฅ๋ฏธ๋กœ์šด ๊ธฐํšŒ๊ฐ€ ์ƒ๊ธฐ๊ฒ ์ฃ .
10:31
For example, in medicine,
210
631888
1492
์˜ˆ๋ฅผ ๋“ค์–ด ์˜ํ•™์—์„œ
10:33
a team in Boston announced that they had discovered
211
633380
2525
๋ณด์Šคํ„ด์˜ ํŒ€์€ ์ข…์–‘์—์„œ
10:35
dozens of new clinically relevant features
212
635905
2949
์ž„์ƒ์ ์œผ๋กœ ๊ด€๋ จ๋œ ์ˆ˜์‹ญ๊ฐ€์ง€์˜ ํŠน์ง•์„ ์ƒˆ๋กญ๊ฒŒ ๋ฐœ๊ฒฌํ–ˆ๋Š”๋ฐ
10:38
of tumors which help doctors make a prognosis of a cancer.
213
638854
4266
์ด๊ฒƒ์œผ๋กœ ์˜์‚ฌ๋“ค์ด ์•”์„ ์˜ˆ์ธกํ•˜๋Š”๋ฐ ๋„์›€์„ ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
10:44
Very similarly, in Stanford,
214
644220
2296
์Šคํƒ ํฌ๋“œ์—์„œ๋„ ๋น„์Šทํ•˜๊ฒŒ
10:46
a group there announced that, looking at tissues under magnification,
215
646516
3663
ํ•œ ๊ทธ๋ฃน์ด ์กฐ์ง์„ ํ™•๋Œ€๊ฒฝ์œผ๋กœ ๋ณด๋Š”
10:50
they've developed a machine learning-based system
216
650179
2381
๊ธฐ๊ณ„ ํ•™์Šต์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ์‹œ์Šคํ…œ์„ ๊ฐœ๋ฐœํ–ˆ๋Š”๋ฐ
10:52
which in fact is better than human pathologists
217
652560
2582
์‚ฌ์‹ค ์•” ํ™˜์ž์˜ ์ƒ์กด์œจ์„ ์˜ˆ์ธกํ•˜๋Š”๋ฐ
10:55
at predicting survival rates for cancer sufferers.
218
655142
4377
๋ณ‘๋ฆฌํ•™์ž๋ณด๋‹ค ๋‚ซ๋‹ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค.
10:59
In both of these cases, not only were the predictions more accurate,
219
659519
3245
๋‘ ๊ฒฝ์šฐ ๋ชจ๋‘ ์˜ˆ์ธก์ด ๋” ์ •ํ™•ํ•  ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ
11:02
but they generated new insightful science.
220
662764
2502
ํ†ต์ฐฐ๋ ฅ์žˆ๋Š” ๊ณผํ•™์„ ์ƒˆ๋กœ ๋งŒ๋“ค์–ด๋ƒˆ์Šต๋‹ˆ๋‹ค.
11:05
In the radiology case,
221
665276
1505
๋ฐฉ์‚ฌ์„ ํ•™์˜ ๊ฒฝ์šฐ
11:06
they were new clinical indicators that humans can understand.
222
666781
3095
์‚ฌ๋žŒ์ด ์ดํ•ดํ•  ์ˆ˜ ์žˆ๋Š” ์ƒˆ๋กœ์šด ์ž„์ƒ ์ง•ํ›„๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค.
11:09
In this pathology case,
223
669876
1792
๋ณ‘๋ฆฌํ•™์˜ ๊ฒฝ์šฐ
11:11
the computer system actually discovered that the cells around the cancer
224
671668
4500
์ปดํ“จํ„ฐ ์‹œ์Šคํ…œ์€ ์ง„๋‹จ์„ ํ•˜๋Š”๋ฐ
11:16
are as important as the cancer cells themselves
225
676168
3340
์‹ค์ œ๋กœ ์•”์ฃผ๋ณ€์˜ ์„ธํฌ๊ฐ€ ์•” ์„ธํฌ ๋งŒํผ์ด๋‚˜
11:19
in making a diagnosis.
226
679508
1752
์ค‘์š”ํ•˜๋‹ค๋Š” ์‚ฌ์‹ค์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค.
11:21
This is the opposite of what pathologists had been taught for decades.
227
681260
5361
์ด๋Š” ๋ณ‘๋ฆฌํ•™์ž๊ฐ€ ์ˆ˜์‹ญ๋…„๋™์•ˆ ๊ฐ€๋ฅด์นœ ์‚ฌ์‹ค๊ณผ ๋ฐ˜๋Œ€๋ฉ๋‹ˆ๋‹ค.
11:26
In each of those two cases, they were systems developed
228
686621
3292
๊ฐ๊ฐ์˜ ๊ฒฝ์šฐ์—์„œ ์‹œ์Šคํ…œ์€
11:29
by a combination of medical experts and machine learning experts,
229
689913
3621
์˜ํ•™ ์ „๋ฌธ๊ณผ์™€ ๊ธฐ๊ณ„ ํ•™์Šต ์ „๋ฌธ๊ฐ€๊ฐ€ ํ•จ๊ป˜ ๊ฐœ๋ฐœํ–ˆ์ง€๋งŒ
11:33
but as of last year, we're now beyond that too.
230
693534
2741
์ž‘๋…„์— ๊ทธ๊ฑธ ๋›ฐ์–ด๋„˜์—ˆ์Šต๋‹ˆ๋‹ค.
11:36
This is an example of identifying cancerous areas
231
696275
3549
์ด๊ฒƒ์€ ํ˜„๋ฏธ๊ฒฝ์œผ๋กœ ์‚ฌ๋žŒ์˜ ์กฐ์ง์—์„œ
11:39
of human tissue under a microscope.
232
699824
2530
์•” ์กฐ์ง์„ ๋ฐํžˆ๋Š” ์˜ˆ์ž…๋‹ˆ๋‹ค.
11:42
The system being shown here can identify those areas more accurately,
233
702354
4613
์—ฌ๊ธฐ์„œ ๋ณด๋Š” ์‹œ์Šคํ…œ์€ ์•” ์กฐ์ง์„ ๋” ์ •ํ™•ํžˆ ํŒ๋ณ„ํ•  ์ˆ˜ ์žˆ๊ณ 
11:46
or about as accurately, as human pathologists,
234
706967
2775
๋ณ‘๋ฆฌํ•™์ž๋งŒํผ์ด๋‚˜ ์ •ํ™•ํ•˜๊ฒŒ ํŒ๋ณ„ํ•  ์ˆ˜ ์žˆ์ง€๋งŒ
11:49
but was built entirely with deep learning using no medical expertise
235
709742
3392
์˜ํ•™ ์ „๋ฌธ๊ฐ€๋ฅผ ์“ฐ์ง€ ์•Š๊ณ  ๊ทธ ๋ถ„์•ผ์— ์ง€์‹์ด ์ „ํ˜€ ์—†๋Š” ์‚ฌ๋žŒ๋“ค์ด
11:53
by people who have no background in the field.
236
713134
2526
์‹ฌํ™” ํ•™์Šต ๋งŒ์œผ๋กœ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.
11:56
Similarly, here, this neuron segmentation.
237
716730
2555
๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ์—ฌ๊ธฐ ์‹ ๊ฒฝ ๋ถ„ํ• ์ธ๋ฐ
11:59
We can now segment neurons about as accurately as humans can,
238
719285
3668
์‚ฌ๋žŒ๋งŒํผ์ด๋‚˜ ์ •ํ™•ํ•˜๊ฒŒ ์‹ ๊ฒฝ์„ ๋ถ„ํ• ํ•  ์ˆ˜ ์žˆ์ง€๋งŒ
12:02
but this system was developed with deep learning
239
722953
2717
์ด ์‹œ์Šคํ…œ์€ ์˜ํ•™์— ๋ฐฐ๊ฒฝ์ง€์‹์ด ์—†๋Š” ์‚ฌ๋žŒ๋“ค์ด
12:05
using people with no previous background in medicine.
240
725670
3251
์‹ฌํ™” ํ•™์Šต์„ ์ด์šฉํ•ด์„œ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.
12:08
So myself, as somebody with no previous background in medicine,
241
728921
3227
๊ทธ๋ž˜์„œ ์ €์ฒ˜๋Ÿผ ์˜ํ•™์— ๋ฐฐ๊ฒฝ์ง€์‹์ด ์—†๋Š” ์‚ฌ๋žŒ์ด
12:12
I seem to be entirely well qualified to start a new medical company,
242
732148
3727
์ƒˆ๋กœ์šด ์˜๋ฃŒ ํšŒ์‚ฌ๋ฅผ ์‹œ์ž‘ํ•˜๋Š”๋ฐ ์•„์ฃผ ์ ํ•ฉํ•œ ์‚ฌ๋žŒ์ฒ˜๋Ÿผ ๋ณด์—ฌ์„œ
12:15
which I did.
243
735875
2146
์‹ค์ œ๋กœ ๊ทธ๋ ‡๊ฒŒ ํ–ˆ์ฃ .
12:18
I was kind of terrified of doing it,
244
738021
1740
๊ณตํฌ๋ฅผ ๋Š๊ผˆ์ง€๋งŒ
12:19
but the theory seemed to suggest that it ought to be possible
245
739761
2889
์ด๋ก ์€ ์ด๋Ÿฐ ๋ฐ์ดํ„ฐ ๋ถ„์„๊ธฐ๋ฒ•์„ ์ด์šฉํ•ด์„œ
12:22
to do very useful medicine using just these data analytic techniques.
246
742650
5492
์•„์ฃผ ์œ ์šฉํ•œ ์˜ํ•™์ด ๊ฐ€๋Šฅํ•จ์„ ์ œ์‹œํ•ด์ฃผ๊ณ  ์žˆ์—ˆ์ฃ .
12:28
And thankfully, the feedback has been fantastic,
247
748142
2480
๊ทธ๋ฆฌ๊ณ  ๊ฐ์‚ฌํ•˜๊ฒŒ๋„ ํ‰๊ฐ€๋Š” ์ข‹์•˜์Šต๋‹ˆ๋‹ค.
12:30
not just from the media but from the medical community,
248
750622
2356
๋ฏธ๋””์–ด ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์˜ํ•™๊ณ„์—์„œ๋„
12:32
who have been very supportive.
249
752978
2344
์•„์ฃผ ๊ธ์ •์ ์ด์—ˆ์Šต๋‹ˆ๋‹ค.
12:35
The theory is that we can take the middle part of the medical process
250
755322
4149
๊ทธ ์ด๋ก ์€ ์˜๋ฃŒ ๊ณผ์ •์˜ ์ค‘๊ฐ„ ๋ถ€๋ถ„์„ ์šฐ๋ฆฌ๊ฐ€ ๊ฐ€์ ธ์™€์„œ
12:39
and turn that into data analysis as much as possible,
251
759471
2893
์ตœ๋Œ€ํ•œ ๋ฐ์ดํ„ฐ ๋ถ„์„์„ ํ•œ ๋’ค
12:42
leaving doctors to do what they're best at.
252
762364
3065
์˜์‚ฌ๋“ค์—๊ฒŒ ๊ทธ๋“ค์ด ์ž˜ํ•˜๋Š” ์ผ์„ ๋งก๊ธฐ๋Š” ๊ฑฐ์ฃ .
12:45
I want to give you an example.
253
765429
1602
์˜ˆ๋ฅผ ๋ณด์—ฌ๋“œ๋ฆฌ๊ฒ ์Šต๋‹ˆ๋‹ค.
12:47
It now takes us about 15 minutes to generate a new medical diagnostic test
254
767031
4944
์ƒˆ๋กœ์šด ์˜๋ฃŒ ์ง„๋‹จ ์‹คํ—˜์„ ํ•˜๋Š”๋ฐ 15๋ถ„์ฏค ๊ฑธ๋ฆฌ๋Š”๋ฐ
12:51
and I'll show you that in real time now,
255
771975
1954
์ด์ œ ์‹ค์‹œ๊ฐ„์œผ๋กœ ๋ณด์—ฌ๋“œ๋ฆฌ์ฃ .
12:53
but I've compressed it down to three minutes by cutting some pieces out.
256
773929
3487
๋ช‡ ๋‹จ๊ณ„๋ฅผ ์ƒ๋žตํ•ด์„œ 3๋ถ„์œผ๋กœ ์ค„์˜€์Šต๋‹ˆ๋‹ค.
12:57
Rather than showing you creating a medical diagnostic test,
257
777416
3061
์˜๋ฃŒ ์ง„๋‹จ ์‹คํ—˜์„ ํ•˜๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ฃผ๋Š” ๋Œ€์‹ 
13:00
I'm going to show you a diagnostic test of car images,
258
780477
3369
์ž๋™์ฐจ ์‚ฌ์ง„์˜ ์ง„๋‹จ ์‹คํ—˜์„ ๋ณด์—ฌ๋“œ๋ฆฌ๊ฒ ์Šต๋‹ˆ๋‹ค.
13:03
because that's something we can all understand.
259
783846
2222
์™œ๋ƒํ•˜๋ฉด ์šฐ๋ฆฌ ๋ชจ๋‘ ์ดํ•ดํ•  ์ˆ˜ ์žˆ๋Š” ๊ฑฐ๋‹ˆ๊นŒ์š”.
13:06
So here we're starting with about 1.5 million car images,
260
786068
3201
์—ฌ๊ธฐ์„œ 150๋งŒ ๊ฐœ์˜ ์ž๋™์ฐจ ์‚ฌ์ง„์œผ๋กœ ์‹œ์ž‘ํ•˜์ฃ .
13:09
and I want to create something that can split them into the angle
261
789269
3206
์‚ฌ์ง„์„ ์ฐ์€ ๊ฐ๋„๋กœ
13:12
of the photo that's being taken.
262
792475
2223
๋ถ„๋ฅ˜ํ•˜๋Š” ๋ญ”๊ฐ€๋ฅผ ๋งŒ๋“ค๊ณ  ์‹ถ์–ด์š”.
13:14
So these images are entirely unlabeled, so I have to start from scratch.
263
794698
3888
์ด ์‚ฌ์ง„๋“ค์€ ๋ชจ๋‘ ์ œ๋ชฉ๋„ ์—†์–ด์„œ ์ฒ˜์Œ๋ถ€ํ„ฐ ์‹œ์ž‘ํ•ด์•ผ ๋ฉ๋‹ˆ๋‹ค.
13:18
With our deep learning algorithm,
264
798586
1865
์‹ฌํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ
13:20
it can automatically identify areas of structure in these images.
265
800451
3707
์ด ์‚ฌ์ง„๋“ค์˜ ๊ตฌ์กฐ๋ฅผ ์ž๋™์œผ๋กœ ๊ตฌ๋ณ„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
13:24
So the nice thing is that the human and the computer can now work together.
266
804158
3620
์ข‹์€ ์ ์€ ์‚ฌ๋žŒ๊ณผ ์ปดํ“จํ„ฐ๊ฐ€ ํ•จ๊ป˜ ์ผํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฑฐ์ฃ .
13:27
So the human, as you can see here,
267
807778
2178
์‚ฌ๋žŒ์€ ์—ฌ๊ธฐ์„œ ๋ณด๋‹ค์‹œํ”ผ
13:29
is telling the computer about areas of interest
268
809956
2675
์ปดํ“จํ„ฐํ•œํ…Œ ๊ด€์‹ฌ๋ถ„์•ผ๋ฅผ ๋งํ•˜๊ณ 
13:32
which it wants the computer then to try and use to improve its algorithm.
269
812631
4650
์ปดํ“จํ„ฐ๊ฐ€ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๊ฐœ์„ ํ•˜์ฃ .
13:37
Now, these deep learning systems actually are in 16,000-dimensional space,
270
817281
4296
์ž, ์ด ์‹ฌํ™” ํ•™์Šต ์‹œ์Šคํ…œ์€ ์‹ค์ œ๋กœ 1๋งŒ6์ฒœ ์ฐจ์›์˜ ๊ณต๊ฐ„์„ ๊ฐ€์ง‘๋‹ˆ๋‹ค.
13:41
so you can see here the computer rotating this through that space,
271
821577
3432
์ปดํ“จํ„ฐ๊ฐ€ ์ด๊ฒƒ์„ ๊ทธ ๊ณต๊ฐ„์‚ฌ์ด๋กœ ํšŒ์ „ํ•˜๋Š” ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
13:45
trying to find new areas of structure.
272
825009
1992
์ƒˆ๋กœ์šด ๊ตฌ์กฐ๋ฅผ ๋ฐœ๊ฒฌํ•˜๋ ค๋Š” ๊ฑฐ์ฃ .
13:47
And when it does so successfully,
273
827001
1781
์ปดํ“จํ„ฐ๊ฐ€ ์„ฑ๊ณต์ ์œผ๋กœ ๋๋‚ด๋ฉด
13:48
the human who is driving it can then point out the areas that are interesting.
274
828782
4004
๊ทธ๊ฑธ ์ž‘๋™ํ•˜๋Š” ์‚ฌ๋žŒ์€ ๊ด€์‹ฌ์žˆ๋Š” ๋ถ„์•ผ๋ฅผ ๊ฐ€๋ฆฌํ‚ต๋‹ˆ๋‹ค.
13:52
So here, the computer has successfully found areas,
275
832786
2422
์—ฌ๊ธฐ์„œ ์ปดํ“จํ„ฐ๋Š” ๊ทธ ๋ถ„์•ผ๋ฅผ ์„ฑ๊ณต์ ์œผ๋กœ ์ฐพ์•„๋ƒˆ๋Š”๋ฐ
13:55
for example, angles.
276
835208
2562
์ด ๊ฒฝ์šฐ๋Š” ๊ฐ๋„์ด์ฃ .
13:57
So as we go through this process,
277
837770
1606
์šฐ๋ฆฌ๊ฐ€ ์ด ๊ณผ์ •์„ ๊ฑฐ์ณ๊ฐ€๋ฉด์„œ
13:59
we're gradually telling the computer more and more
278
839376
2340
์ปดํ“จํ„ฐํ•œํ…Œ ์šฐ๋ฆฌ๊ฐ€ ์ฐพ๊ณ  ์žˆ๋Š” ๊ตฌ์กฐ์— ๋Œ€ํ•ด์„œ
14:01
about the kinds of structures we're looking for.
279
841716
2428
๋‹จ๊ณ„์ ์œผ๋กœ ๋” ๋งŽ์ด ๋งํ•ด์ค๋‹ˆ๋‹ค.
14:04
You can imagine in a diagnostic test
280
844144
1772
์ง„๋‹จ ์‹คํ—˜์—์„œ
14:05
this would be a pathologist identifying areas of pathosis, for example,
281
845916
3350
๋ณ‘๋ฆฌํ•™์ž๊ฐ€ ๋ณ‘์  ์ƒํƒœ์ธ ๊ณณ์„ ๋ฐํ˜€๋‚ด๊ฑฐ๋‚˜
14:09
or a radiologist indicating potentially troublesome nodules.
282
849266
5026
๋ฐฉ์‚ฌ์„ ์˜๊ฐ€ ๋ฌธ์ œ๊ฐ€ ์žˆ์„ ์ˆ˜ ์žˆ๋Š” ํ˜น์„ ๊ฐ€๋ฅดํ‚ค๋Š” ๊ฒƒ์„ ์ƒ์ƒํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
14:14
And sometimes it can be difficult for the algorithm.
283
854292
2559
์•Œ๊ณ ๋ฆฌ์ฆ˜์—์„œ ์–ด๋ ค์šด ๋ถ€๋ถ„๋„ ์žˆ์Šต๋‹ˆ๋‹ค.
14:16
In this case, it got kind of confused.
284
856851
1964
์ด ๊ฒฝ์šฐ ์•ฝ๊ฐ„ ํ—ท๊ฐˆ๋ ธ์–ด์š”.
14:18
The fronts and the backs of the cars are all mixed up.
285
858815
2550
์ž๋™์ฐจ์˜ ์•ž๊ณผ ๋’ค๊ฐ€ ๋ชจ๋‘ ์„ž์—ฌ๋ฒ„๋ ธ์ฃ .
14:21
So here we have to be a bit more careful,
286
861365
2072
๊ทธ๋ž˜์„œ ์—ฌ๊ธฐ์„œ ์ข€๋” ์ฃผ์˜ํ•ด์„œ
14:23
manually selecting these fronts as opposed to the backs,
287
863437
3232
๋’ค๊ฐ€ ์•„๋‹ˆ๋ผ ์•ž์„ ์ˆ˜๋™์œผ๋กœ ์„ ํƒํ•ด์„œ
14:26
then telling the computer that this is a type of group
288
866669
5506
์ปดํ“จํ„ฐ์—๊ฒŒ ์šฐ๋ฆฌ๊ฐ€ ๊ด€์‹ฌ์žˆ๋Š” ๋ถ€๋ถ„์ด ์ด ๋ถ€๋ถ„์ด๋ผ๊ณ 
14:32
that we're interested in.
289
872175
1348
์–˜๊ธฐ๋ฅผ ํ•ด์•ผํ•ฉ๋‹ˆ๋‹ค.
14:33
So we do that for a while, we skip over a little bit,
290
873523
2677
๊ทธ๋ž˜์„œ ํ•œ๋™์•ˆ ๊ทธ ์ž‘์—…์„ ํ•˜๊ณ  ์ข€ ๋” ๊ฑด๋„ˆ๋›ฐ๋ฉด
14:36
and then we train the machine learning algorithm
291
876200
2246
์ด๋Ÿฐ ์ˆ˜๋ฐฑ ๊ฐ€์ง€ ์ผ์„ ๋ฐ”ํƒ•์œผ๋กœ
14:38
based on these couple of hundred things,
292
878446
1974
๊ธฐ๊ณ„ ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ํ›ˆ๋ จ์‹œ์ผœ
14:40
and we hope that it's gotten a lot better.
293
880420
2025
์•ž์œผ๋กœ ๋” ๋‚˜์•„์ง€๊ธฐ๋ฅผ ๋ฐ”๋ž๋‹ˆ๋‹ค.
14:42
You can see, it's now started to fade some of these pictures out,
294
882445
3073
๋ณด๋‹ค์‹œํ”ผ ์‹œ์Šคํ…œ์€ ์‚ฌ์ง„๋“ค ์ผ๋ถ€๋ฅผ ์‚ฌ๋ผ์ง€๊ฒŒ ๋งŒ๋“ค๋ฉด์„œ
14:45
showing us that it already is recognizing how to understand some of these itself.
295
885518
4708
์ด ์‚ฌ์ง„๋“ค์„ ์ดํ•ดํ•˜๋Š” ๋ฒ•์„ ์ด๋ฏธ ์ธ์‹ํ•˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
14:50
We can then use this concept of similar images,
296
890226
2902
์šฐ๋ฆฌ๋Š” ๋น„์Šทํ•œ ์‚ฌ์ง„์˜ ๊ฐœ๋…์„ ์ด์šฉํ•ด์„œ
14:53
and using similar images, you can now see,
297
893128
2094
์ด์ œ ์—ฌ๋Ÿฌ๋ถ„์ด ๋ณด๋Š” ๊ฒƒ๊ณผ ๊ฐ™์ด
14:55
the computer at this point is able to entirely find just the fronts of cars.
298
895222
4019
์ด ์‹œ์ ์—์„œ ์ปดํ“จํ„ฐ๋Š” ์ž๋™์ฐจ์˜ ์•ž๋งŒ ์ฐพ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
14:59
So at this point, the human can tell the computer,
299
899241
2948
์ด ์‹œ์ ์—์„œ ์‚ฌ๋žŒ์€ ์ปดํ“จํ„ฐ์—๊ฒŒ
15:02
okay, yes, you've done a good job of that.
300
902189
2293
์ข‹์•„, ์ž˜ ํ–ˆ์–ด. ๋ผ๊ณ  ๋งํ•  ์ˆ˜ ์žˆ์ฃ .
15:05
Sometimes, of course, even at this point
301
905652
2185
๋ฌผ๋ก  ์–ด๋–ค ๊ฒฝ์šฐ๋Š” ์ด ์‹œ์ ์—๋„
15:07
it's still difficult to separate out groups.
302
907837
3674
๊ทธ๋ฃน์œผ๋กœ ๋‚˜๋ˆ„๊ธฐ๊ฐ€ ์–ด๋ ต์Šต๋‹ˆ๋‹ค.
15:11
In this case, even after we let the computer try to rotate this for a while,
303
911511
3884
์ด ๊ฒฝ์šฐ ์ปดํ“จํ„ฐ๊ฐ€ ํ•œ๋™์•ˆ ์ด๊ฒƒ์„ ํšŒ์ „ํ•˜๊ฒŒ ๋‚ด๋ฒ„๋ ค๋‘ฌ๋„
15:15
we still find that the left sides and the right sides pictures
304
915399
3345
์™ผ์ชฝ๊ณผ ์˜ค๋ฅธ์ชฝ์ด ๋’ค์„ž์ธ ๊ฒƒ์„
15:18
are all mixed up together.
305
918744
1478
๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
15:20
So we can again give the computer some hints,
306
920222
2140
๊ทธ๋ž˜์„œ ์ปดํ“จํ„ฐํ•œํ…Œ ๋‹ค์‹œ ํžŒํŠธ๋ฅผ ์ค˜์„œ
15:22
and we say, okay, try and find a projection that separates out
307
922362
2976
์‹ฌํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ด์šฉํ•ด์„œ
15:25
the left sides and the right sides as much as possible
308
925338
2607
์™ผ์ชฝ๊ณผ ์˜ค๋ฅธ์ชฝ์„ ๊ฐ€๋Šฅํ•œ ๋ถ„๋ฆฌ์‹œํ‚ค๋Š”
15:27
using this deep learning algorithm.
309
927945
2122
ํˆฌ์‚ฌ๋„๋ฅผ ์ฐพ์•„๋ผ๊ณ  ํ•ฉ๋‹ˆ๋‹ค.
15:30
And giving it that hint -- ah, okay, it's been successful.
310
930067
2942
๊ทธ ํžŒํŠธ๋ฅผ ์ฃผ๋ฉด ์„ฑ๊ณต์ž…๋‹ˆ๋‹ค.
15:33
It's managed to find a way of thinking about these objects
311
933009
2882
์ด๋“ค ๋ฌผ์ฒด๋“ค์„ ๋ถ„๋ฆฌํ•ด๋‚ด๋Š”
15:35
that's separated out these together.
312
935891
2380
๋ฐฉ๋ฒ•์„ ์Šค์Šค๋กœ ์ฐพ์€ ๊ฑฐ์ฃ .
15:38
So you get the idea here.
313
938271
2438
์—ฌ๊ธฐ์„œ ์ƒ๊ฐ์„ ์–ป์„ ์ˆ˜ ์žˆ์ฃ .
15:40
This is a case not where the human is being replaced by a computer,
314
940709
8197
์‚ฌ๋žŒ์ด ์ปดํ“จํ„ฐ๋กœ ๋Œ€์ฒด๋˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์•„๋‹ˆ๋ผ
15:48
but where they're working together.
315
948906
2640
ํ•จ๊ป˜ ์ผํ•ฉ๋‹ˆ๋‹ค.
15:51
What we're doing here is we're replacing something that used to take a team
316
951546
3550
์šฐ๋ฆฌ๊ฐ€ ์—ฌ๊ธฐ์„œ ํ•˜๋Š” ์ผ์€
15:55
of five or six people about seven years
317
955096
2002
5-6๋ช…์˜ ํŒ€์ด 7๋…„์ฏค ๊ฑธ๋ฆฌ๋Š” ์ผ์„
15:57
and replacing it with something that takes 15 minutes
318
957098
2605
ํ•œ ์‚ฌ๋žŒ์ด 15๋ถ„ ๊ฑธ๋ ค์„œ
15:59
for one person acting alone.
319
959703
2505
ํ•˜๋Š” ์ผ๋กœ ๋Œ€์ฒดํ•ฉ๋‹ˆ๋‹ค.
16:02
So this process takes about four or five iterations.
320
962208
3950
์ด ๊ณผ์ •์€ 4 - 5 ๋ฒˆ ๋ฐ˜๋ณตํ•ฉ๋‹ˆ๋‹ค.
16:06
You can see we now have 62 percent
321
966158
1859
๋ณด๋‹ค์‹œํ”ผ 150๋งŒ ์žฅ์˜ ์‚ฌ์ง„์˜
16:08
of our 1.5 million images classified correctly.
322
968017
2959
62%๊ฐ€ ์ œ๋Œ€๋กœ ๋ถ„๋ฅ˜๋œ ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์ฃ .
16:10
And at this point, we can start to quite quickly
323
970976
2472
์ด ์‹œ์ ์—์„œ ์šฐ๋ฆฌ๋Š”
16:13
grab whole big sections,
324
973448
1297
์ „์ฒด๋ฅผ ๋น ๋ฅด๊ฒŒ ์žก์•„์„œ
16:14
check through them to make sure that there's no mistakes.
325
974745
2919
์‹ค์ˆ˜๊ฐ€ ์—†๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.
16:17
Where there are mistakes, we can let the computer know about them.
326
977664
3952
์‹ค์ˆ˜๊ฐ€ ์žˆ์œผ๋ฉด ์ปดํ“จํ„ฐ์—๊ฒŒ ์•Œ๋ฆฌ์ฃ .
16:21
And using this kind of process for each of the different groups,
327
981616
3045
๊ฐ๊ฐ์˜ ๋‹ค๋ฅธ ๊ทธ๋ฃน์—์„œ ์ด๋Ÿฐ ๊ณผ์ •์„ ํ†ตํ•ด
16:24
we are now up to an 80 percent success rate
328
984661
2487
150๋งŒ ์žฅ์˜ ์‚ฌ์ง„์„ ๋ถ„๋ฅ˜ํ•˜๋Š”๋ฐ
16:27
in classifying the 1.5 million images.
329
987148
2415
80%์˜ ์„ฑ๊ณต์œจ์„ ๋ณด์ž…๋‹ˆ๋‹ค.
16:29
And at this point, it's just a case
330
989563
2078
์ด ์‹œ์ ์—์„œ๋Š”
16:31
of finding the small number that aren't classified correctly,
331
991641
3579
๋ฐ”๋ฅด๊ฒŒ ๋ถ„๋ฅ˜๋˜์ง€ ์•Š์€ ์ ์€ ์ˆซ์ž๋ฅผ ์ฐพ์•„
16:35
and trying to understand why.
332
995220
2888
์ด์œ ๋ฅผ ์•Œ์•„๋‚ด๋Š” ๊ณผ์ •์ž…๋‹ˆ๋‹ค.
16:38
And using that approach,
333
998108
1743
๊ทธ๋Ÿฐ ๋ฐฉ์‹์œผ๋กœ
16:39
by 15 minutes we get to 97 percent classification rates.
334
999851
4121
15๋ถ„ ์•ˆ์— ์šฐ๋ฆฌ๋Š” 97%์˜ ๋ถ„๋ฅ˜์œจ์„ ์–ป์Šต๋‹ˆ๋‹ค.
16:43
So this kind of technique could allow us to fix a major problem,
335
1003972
4600
์ด๋Ÿฐ ๊ธฐ์ˆ ์€ ์šฐ๋ฆฌ๊ฐ€ ์ค‘์š”ํ•œ ๋ฌธ์ œ๋ฅผ ๊ณ ์น  ์ˆ˜ ์žˆ๊ฒŒ ํ•˜๋Š”๋ฐ
16:48
which is that there's a lack of medical expertise in the world.
336
1008578
3036
๊ทธ๊ฒƒ์€ ์„ธ๊ณ„์—์„œ ์˜๋ฃŒ ์ „๋ฌธ๊ฐ€๊ฐ€ ๋ถ€์กฑํ•˜๋‹ค๋Š” ์‚ฌ์‹ค์ž…๋‹ˆ๋‹ค.
16:51
The World Economic Forum says that there's between a 10x and a 20x
337
1011614
3489
์„ธ๊ณ„ ๊ฒฝ์ œ ํฌ๋Ÿผ์€ ๊ฐœ๋ฐœ๋„์ƒ๊ตญ์—์„œ
16:55
shortage of physicians in the developing world,
338
1015103
2624
10๋ฐฐ์—์„œ 20๋ฐฐ์˜ ์˜์‚ฌ๊ฐ€ ๋ถ€์กฑํ•˜๋‹ค๊ณ  ๋งํ–ˆ๋Š”๋ฐ
16:57
and it would take about 300 years
339
1017727
2113
๊ทธ ๋ฌธ์ œ๋ฅผ ๊ณ ์น˜๊ธฐ ์œ„ํ•ด
16:59
to train enough people to fix that problem.
340
1019840
2894
์ถฉ๋ถ„ํ•œ ์ธ์›์„ ๊ต์œก์‹œํ‚ค๋ ค๋ฉด 300๋…„์ด ๊ฑธ๋ฆฝ๋‹ˆ๋‹ค.
17:02
So imagine if we can help enhance their efficiency
341
1022734
2885
์ด๋Ÿฐ ์‹ฌํ™” ํ•™์Šต ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•ด์„œ
17:05
using these deep learning approaches?
342
1025619
2839
๊ทธ๋“ค์˜ ํšจ์œจ์„ ๋†’์ผ ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ƒ์ƒํ•ด๋ณด์„ธ์š”.
17:08
So I'm very excited about the opportunities.
343
1028458
2232
์ €๋Š” ๊ทธ๋Ÿฐ ๊ธฐํšŒ์— ๋Œ€ํ•ด ์•„์ฃผ ํฅ๋ถ„ํ–ˆ์Šต๋‹ˆ๋‹ค.
17:10
I'm also concerned about the problems.
344
1030690
2589
์ €๋Š” ๊ทธ ๋ฌธ์ œ๋„ ๊ฑฑ์ •ํ•ฉ๋‹ˆ๋‹ค.
17:13
The problem here is that every area in blue on this map
345
1033279
3124
์—ฌ๊ธฐ์„œ ๋ฌธ์ œ๋Š” ์ด ์ง€๋„์—์„œ ํŒŒ๋ž€์ƒ‰์œผ๋กœ ํ‘œ์‹œ๋œ ๊ณณ์€
17:16
is somewhere where services are over 80 percent of employment.
346
1036403
3769
์„œ๋น„์Šค๊ฐ€ ๊ณ ์šฉ์˜ 80% ์ด์ƒ์„ ์ฐจ์ง€ํ•ฉ๋‹ˆ๋‹ค.
17:20
What are services?
347
1040172
1787
๋ฌด์Šจ ์„œ๋น„์Šค์ผ๊นŒ์š”?
17:21
These are services.
348
1041959
1514
์ด๋Ÿฐ ์„œ๋น„์Šค์ž…๋‹ˆ๋‹ค.
17:23
These are also the exact things that computers have just learned how to do.
349
1043473
4154
์ด๊ฒƒ๋“ค์€ ์ปดํ“จํ„ฐ๊ฐ€ ๋ฐฉ๊ธˆ ๋ฐฐ์šด ๊ฒƒ๊ณผ ๋˜‘๊ฐ™์Šต๋‹ˆ๋‹ค.
17:27
So 80 percent of the world's employment in the developed world
350
1047627
3804
๊ฐœ๋ฐœ๋œ ์„ธ์ƒ์—์„œ ๊ณ ์šฉ์˜ 80%๊ฐ€
17:31
is stuff that computers have just learned how to do.
351
1051431
2532
์ปดํ“จํ„ฐ๊ฐ€ ๋ฐฉ๊ธˆ ๋ฐฐ์šด ๊ฒƒ์ž…๋‹ˆ๋‹ค.
17:33
What does that mean?
352
1053963
1440
๊ทธ๊ฒŒ ๋ญ˜ ๋œปํ•ฉ๋‹ˆ๊นŒ?
17:35
Well, it'll be fine. They'll be replaced by other jobs.
353
1055403
2583
๊ธ€์Ž„, ๊ดœ์ฐฎ์„๊ฑฐ์—์š”. ๋‹ค๋ฅธ ์ผ์ž๋ฆฌ๋กœ ๋Œ€์ฒด๋˜๊ฒ ์ฃ .
17:37
For example, there will be more jobs for data scientists.
354
1057986
2707
์˜ˆ๋ฅผ ๋“ค๋ฉด, ๋ฐ์ดํ„ฐ ๊ณผํ•™์žํ•œํ…Œ ๋” ๋งŽ์€ ์ผ์ด ์žˆ์„ ๊ฒ๋‹ˆ๋‹ค.
17:40
Well, not really.
355
1060693
817
๊ทธ๋ ‡์ง€ ์•Š์•„์š”.
17:41
It doesn't take data scientists very long to build these things.
356
1061510
3118
๋ฐ์ดํ„ฐ ๊ณผํ•™์ž๊ฐ€ ์ด๋Ÿฐ ๊ฒƒ์„ ๋งŒ๋“œ๋Š”๋ฐ ์˜ค๋ž˜ ๊ฑธ๋ฆฌ์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
17:44
For example, these four algorithms were all built by the same guy.
357
1064628
3252
์˜ˆ๋ฅผ ๋“ค์–ด, 4๊ฐ€์ง€ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๋ชจ๋‘ ํ•œ ์‚ฌ๋žŒ์ด ๋งŒ๋“ค์—ˆ์ฃ .
17:47
So if you think, oh, it's all happened before,
358
1067880
2438
์—ฌ๋Ÿฌ๋ถ„์ด ์ด์ „์—๋„ ์ด๋Ÿฐ ์ผ์ด ๋ฒŒ์–ด์กŒ๋‹ค๊ณ  ์ƒ๊ฐํ•œ๋‹ค๋ฉด
17:50
we've seen the results in the past of when new things come along
359
1070318
3808
๊ณผ๊ฑฐ์— ์ƒˆ๋กœ์šด ๊ฒƒ์ด ๋‚˜ํƒ€๋‚ฌ์„ ๋•Œ ๊ทธ ๊ฒฐ๊ณผ๋ฅผ ๋ณธ ์ ์ด ์žˆ์ฃ .
17:54
and they get replaced by new jobs,
360
1074126
2252
์ƒˆ๋กœ์šด ์ผ์ž๋ฆฌ๋กœ ๋Œ€์ฒด๋˜์—ˆ๊ณ 
17:56
what are these new jobs going to be?
361
1076378
2116
์ƒˆ๋กœ์šด ์ผ์ž๋ฆฌ๋Š” ์–ด๋–ค ๊ฒƒ์ผ๊นŒ์š”?
17:58
It's very hard for us to estimate this,
362
1078494
1871
์ด๊ฒƒ์„ ์˜ˆ์ธกํ•˜๊ธฐ๊ฐ€ ์ •๋ง ์–ด๋ ต์Šต๋‹ˆ๋‹ค.
18:00
because human performance grows at this gradual rate,
363
1080365
2739
์™œ๋ƒํ•˜๋ฉด ์‚ฌ๋žŒ์˜ ์„ฑ๊ณผ๋Š” ์ด๋ ‡๊ฒŒ ์ ์ง„์ ์ธ๋ฐ
18:03
but we now have a system, deep learning,
364
1083104
2562
์‹ฌํ™” ํ•™์Šต ์‹œ์Šคํ…œ์€
18:05
that we know actually grows in capability exponentially.
365
1085666
3227
๋Šฅ๋ ฅ์ด ๊ธฐํ•˜๊ธ‰์ˆ˜์ ์œผ๋กœ ์ฆ๊ฐ€ํ•˜๋Š” ๊ฒƒ์„ ์••๋‹ˆ๋‹ค.
18:08
And we're here.
366
1088893
1605
์šฐ๋ฆฌ๋Š” ์—ฌ๊ธฐ์— ์žˆ์ฃ .
18:10
So currently, we see the things around us
367
1090498
2061
ํ˜„์žฌ ์šฐ๋ฆฌ๋Š” ์ฃผ๋ณ€์„ ๋ณด๋ฉด์„œ ๋งํ•ด์š”.
18:12
and we say, "Oh, computers are still pretty dumb." Right?
368
1092559
2676
"์ปดํ“จํ„ฐ๋Š” ์ •๋ง ๋ฐ”๋ณด์•ผ." ๊ทธ๋ ‡์ง€?
18:15
But in five years' time, computers will be off this chart.
369
1095235
3429
ํ•˜์ง€๋งŒ 5๋…„ ์•ˆ์— ์ปดํ“จํ„ฐ๋Š” ์ด ๋„ํ‘œ๋ฐ–์œผ๋กœ ๋‚˜๊ฐˆ ๊ฒ๋‹ˆ๋‹ค.
18:18
So we need to be starting to think about this capability right now.
370
1098664
3865
๊ทธ๋ž˜์„œ ์ด ๋Šฅ๋ ฅ์„ ์ง€๊ธˆ ๋‹น์žฅ ์ƒ๊ฐํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
18:22
We have seen this once before, of course.
371
1102529
2050
๋ฌผ๋ก  ์ „์—๋„ ์ด๊ฑธ ๋ณธ ์ ์ด ์žˆ์Šต๋‹ˆ๋‹ค.
18:24
In the Industrial Revolution,
372
1104579
1387
์‚ฐ์—… ํ˜๋ช…์—์„œ
18:25
we saw a step change in capability thanks to engines.
373
1105966
2851
์—”์ง„ ๋•๋ถ„์— ๋Šฅ๋ ฅ์ด ํ•œ ๋‹จ๊ณ„ ๋‹ฌ๋ผ์กŒ์ฃ .
18:29
The thing is, though, that after a while, things flattened out.
374
1109667
3138
ํ•˜์ง€๋งŒ ์‹œ๊ฐ„์ด ์ข€ ํ๋ฅธ ๋’ค ์˜ค๋ฆ„์„ธ๊ฐ€ ๋ฉˆ์ท„์Šต๋‹ˆ๋‹ค.
18:32
There was social disruption,
375
1112805
1702
์‚ฌํšŒ์  ๋ถ„์—ด์ด ์žˆ์—ˆ์ง€๋งŒ
18:34
but once engines were used to generate power in all the situations,
376
1114507
3439
์—”์ง„์„ ์‚ฌ์šฉํ•ด์„œ ๋ชจ๋“  ์ƒํ™ฉ์—์„œ ๋™๋ ฅ์„ ๋งŒ๋“ค์–ด๋‚ด์ž
18:37
things really settled down.
377
1117946
2354
๋ชจ๋“ ๊ฒŒ ์•ˆ์ •๋˜์—ˆ์ฃ .
18:40
The Machine Learning Revolution
378
1120300
1473
๊ธฐ๊ณ„ ํ•™์Šต ํ˜๋ช…์€
18:41
is going to be very different from the Industrial Revolution,
379
1121773
2909
์‚ฐ์—… ํ˜๋ช…๊ณผ๋Š” ์•„์ฃผ ๋‹ค๋ฆ…๋‹ˆ๋‹ค.
18:44
because the Machine Learning Revolution, it never settles down.
380
1124682
2950
๊ธฐ๊ณ„ ํ•™์Šต ํ˜๋ช…์€ ์ ˆ๋Œ€ ์•ˆ์ •๋˜์ง€ ์•Š์„ ๊ฑฐ๋‹ˆ๊นŒ์š”.
18:47
The better computers get at intellectual activities,
381
1127632
2982
์ปดํ“จํ„ฐ์˜ ์ง€๋Šฅํ™œ๋™์ด ๋” ๋‚˜์„์ˆ˜๋ก
18:50
the more they can build better computers to be better at intellectual capabilities,
382
1130614
4248
๋” ๋‚˜์€ ์ปดํ“จํ„ฐ๋ฅผ ๋งŒ๋“คํ…Œ๊ณ  ๊ทธ ์ปดํ“จํ„ฐ๋Š” ์ง€์  ๋Šฅ๋ ฅ์ด ๋” ๋›ฐ์–ด๋‚˜๊ฒ ์ฃ .
18:54
so this is going to be a kind of change
383
1134862
1908
๊ทธ๋ž˜์„œ ์ด๊ฒƒ์€ ์„ธ๊ณ„๊ฐ€ ์‹ค์ œ๋กœ
18:56
that the world has actually never experienced before,
384
1136770
2478
๊ฒฝํ—˜ํ•ด๋ณธ ์ ์ด ์—†๋Š” ๋ณ€ํ™”๊ฐ€ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
18:59
so your previous understanding of what's possible is different.
385
1139248
3306
์—ฌ๋Ÿฌ๋ถ„์ด ์ด์ „์— ๊ฐ€๋Šฅํ•˜๋‹ค๊ณ  ์ดํ•ดํ•œ ๊ฒƒ๋“ค์ด ์ด์ œ๋Š” ๋‹ค๋ฆ…๋‹ˆ๋‹ค.
19:02
This is already impacting us.
386
1142974
1780
์ด๊ฒƒ์€ ์ด๋ฏธ ์šฐ๋ฆฌ์—๊ฒŒ ์˜ํ–ฅ์„ ์ฃผ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
19:04
In the last 25 years, as capital productivity has increased,
387
1144754
3630
์ง€๋‚œ 25๋…„๊ฐ„ ์ž๋ณธ ์ƒ์‚ฐ๋Ÿ‰์€ ์ฆ๊ฐ€ํ–ˆ์ง€๋งŒ
19:08
labor productivity has been flat, in fact even a little bit down.
388
1148400
4188
๋…ธ๋™ ์ƒ์‚ฐ๋Ÿ‰์€ ๋ณ€ํ™”๊ฐ€ ์—†์—ˆ๊ณ  ์‚ฌ์‹ค ์กฐ๊ธˆ ๊ฐ์†Œํ–ˆ์Šต๋‹ˆ๋‹ค.
19:13
So I want us to start having this discussion now.
389
1153408
2741
๊ทธ๋ž˜์„œ ์ด๋Ÿฐ ํ† ๋ก ์„ ์ง€๊ธˆ๋ถ€ํ„ฐ ์‹œ์ž‘ํ•˜๊ณ  ์‹ถ์Šต๋‹ˆ๋‹ค.
19:16
I know that when I often tell people about this situation,
390
1156149
3027
์ œ๊ฐ€ ์ด๋Ÿฐ ์ƒํ™ฉ์„ ์‚ฌ๋žŒ๋“ค์—๊ฒŒ ์ข…์ข… ์–˜๊ธฐํ•˜๋ฉด
19:19
people can be quite dismissive.
391
1159176
1490
์‚ฌ๋žŒ๋“ค์€ ์•„์ฃผ ๋ฌด์‹œํ•ฉ๋‹ˆ๋‹ค.
19:20
Well, computers can't really think,
392
1160666
1673
์ปดํ“จํ„ฐ๋Š” ์ง„์งœ ์ƒ๊ฐํ•  ์ˆ˜ ์—†์–ด.
19:22
they don't emote, they don't understand poetry,
393
1162339
3028
๊ฐ์ •์„ ๋“œ๋Ÿฌ๋‚ด์ง€ ๋ชปํ•˜๊ณ  ์‹œ๋„ ์ดํ•ด๋ฅผ ๋ชปํ•˜์ง€.
19:25
we don't really understand how they work.
394
1165367
2521
์šฐ๋ฆฌ๋Š” ์ปดํ“จํ„ฐ๊ฐ€ ์–ด๋–ป๊ฒŒ ์ž‘๋™ํ•˜๋Š”์ง€ ์ •๋ง ์ดํ•ดํ•  ์ˆ˜ ์—†์–ด.
19:27
So what?
395
1167888
1486
๊ทธ๋Ÿฌ๋‹ˆ ์–ด์ฉŒ๋ผ๊ณ ?
19:29
Computers right now can do the things
396
1169374
1804
์ปดํ“จํ„ฐ๋Š” ์ง€๊ธˆ
19:31
that humans spend most of their time being paid to do,
397
1171178
2719
์‚ฌ๋žŒ๋“ค์ด ๋ˆ๋ฐ›๊ณ  ํ•˜๋Š” ์ผ์„ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
19:33
so now's the time to start thinking
398
1173897
1731
๊ทธ๋ž˜์„œ ์ด์ œ๋Š” ์šฐ๋ฆฌ๊ฐ€
19:35
about how we're going to adjust our social structures and economic structures
399
1175628
4387
์ด๋Ÿฐ ์ƒˆ๋กœ์šด ํ˜„์‹ค์„ ์ธ์‹ํ•˜๋„๋ก ์‚ฌํšŒ์ , ๊ฒฝ์ œ์  ๊ตฌ์กฐ๋ฅผ ์กฐ์ •ํ•˜๋Š” ๋ฒ•์„
19:40
to be aware of this new reality.
400
1180015
1840
์ƒ๊ฐํ•ด๋ด์•ผ ํ•  ๋•Œ์ž…๋‹ˆ๋‹ค.
19:41
Thank you.
401
1181855
1533
๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
19:43
(Applause)
402
1183388
802
(๋ฐ•์ˆ˜)
์ด ์›น์‚ฌ์ดํŠธ ์ •๋ณด

์ด ์‚ฌ์ดํŠธ๋Š” ์˜์–ด ํ•™์Šต์— ์œ ์šฉํ•œ YouTube ๋™์˜์ƒ์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์ „ ์„ธ๊ณ„ ์ตœ๊ณ ์˜ ์„ ์ƒ๋‹˜๋“ค์ด ๊ฐ€๋ฅด์น˜๋Š” ์˜์–ด ์ˆ˜์—…์„ ๋ณด๊ฒŒ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ฐ ๋™์˜์ƒ ํŽ˜์ด์ง€์— ํ‘œ์‹œ๋˜๋Š” ์˜์–ด ์ž๋ง‰์„ ๋”๋ธ” ํด๋ฆญํ•˜๋ฉด ๊ทธ๊ณณ์—์„œ ๋™์˜์ƒ์ด ์žฌ์ƒ๋ฉ๋‹ˆ๋‹ค. ๋น„๋””์˜ค ์žฌ์ƒ์— ๋งž์ถฐ ์ž๋ง‰์ด ์Šคํฌ๋กค๋ฉ๋‹ˆ๋‹ค. ์˜๊ฒฌ์ด๋‚˜ ์š”์ฒญ์ด ์žˆ๋Š” ๊ฒฝ์šฐ ์ด ๋ฌธ์˜ ์–‘์‹์„ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ์˜ํ•˜์‹ญ์‹œ์˜ค.

https://forms.gle/WvT1wiN1qDtmnspy7