The danger of AI is weirder than you think | Janelle Shane

2,816,115 views

2019-11-13 ใƒป TED


New videos

The danger of AI is weirder than you think | Janelle Shane

2,816,115 views ใƒป 2019-11-13

TED


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

๋ฒˆ์—ญ: Chaelim Lee ๊ฒ€ํ† : DK Kim
00:01
So, artificial intelligence
0
1765
3000
์ธ๊ณต์ง€๋Šฅ์€
00:04
is known for disrupting all kinds of industries.
1
4789
3529
์˜จ๊ฐ– ์ข…๋ฅ˜์˜ ์‚ฐ์—…์„ ์™€ํ•ด์‹œํ‚ค๋Š” ๊ฒƒ์œผ๋กœ ์•Œ๋ ค์ ธ ์žˆ์ฃ .
00:08
What about ice cream?
2
8961
2043
์•„์ด์Šคํฌ๋ฆผ ์‹œ์žฅ์€ ์–ด๋–จ๊นŒ์š”?
00:11
What kind of mind-blowing new flavors could we generate
3
11879
3639
์–ด๋–ค ์ƒˆ๋กญ๊ณ  ๋†€๋ผ์šด ๋ง›๋“ค์„
๊ณ ๋„์˜ ์ธ๊ณต์ง€๋Šฅ์œผ๋กœ ๋งŒ๋“ค ์ˆ˜ ์žˆ์„๊นŒ์š”?
00:15
with the power of an advanced artificial intelligence?
4
15542
2976
00:19
So I teamed up with a group of coders from Kealing Middle School
5
19011
4161
๊ทธ๋ž˜์„œ ์ €๋Š” ํ‚ฌ๋ง ์ค‘ํ•™๊ต ํ•™์ƒ๋“ค๊ณผ
00:23
to find out the answer to this question.
6
23196
2241
์ด ๋ฌธ์ œ์˜ ๋‹ต์„ ์ฐพ์•„ ๋ณด๊ณ ์ž ํ–ˆ์Šต๋‹ˆ๋‹ค.
00:25
They collected over 1,600 existing ice cream flavors,
7
25461
5081
ํ˜„์žฌ ์กด์žฌํ•˜๋Š” 1600๊ฐ€์ง€๊ฐ€ ๋„˜๋Š” ์•„์ด์Šคํฌ๋ฆผ ๋ง›๋“ค์„ ๋ชจ์œผ๊ณ ,
00:30
and together, we fed them to an algorithm to see what it would generate.
8
30566
5522
์–ด๋–ค ๊ฒƒ์„ ๋งŒ๋“ค ์ˆ˜ ์žˆ๋Š”์ง€ ๋ณด๊ธฐ ์œ„ํ•ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.
00:36
And here are some of the flavors that the AI came up with.
9
36112
3753
๊ทธ๋ฆฌ๊ณ  ์—ฌ๊ธฐ AI๊ฐ€ ๋งŒ๋“ค์–ด๋‚ธ ๋ง›๋“ค์ด ์žˆ์Šต๋‹ˆ๋‹ค.
00:40
[Pumpkin Trash Break]
10
40444
1471
[ํ˜ธ๋ฐ• ์“ฐ๋ ˆ๊ธฐ ๋ธŒ๋ ˆ์ดํฌ]
00:41
(Laughter)
11
41939
1402
(์›ƒ์Œ)
00:43
[Peanut Butter Slime]
12
43365
2469
[๋•…์ฝฉ ๋ฒ„ํ„ฐ ์Šฌ๋ผ์ž„]
00:46
[Strawberry Cream Disease]
13
46822
1343
[๋”ธ๊ธฐ ํฌ๋ฆผ ์งˆ๋ณ‘]
00:48
(Laughter)
14
48189
2126
(์›ƒ์Œ)
00:50
These flavors are not delicious, as we might have hoped they would be.
15
50339
4597
์ด ๋ง›๋“ค์€ ์ €ํฌ๊ฐ€ ๋ฐ”๋ž๋˜ ๊ฒƒ ๋งŒํผ ๋ง›์žˆ์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
00:54
So the question is: What happened?
16
54960
1864
๋ฌด์Šจ ์ผ์ด ์ผ์–ด๋‚œ ๊ฒƒ์ผ๊นŒ์š”? ๋ญ๊ฐ€ ์ž˜๋ชป๋œ ๊ฑฐ์ฃ ?
00:56
What went wrong?
17
56848
1394
00:58
Is the AI trying to kill us?
18
58266
1959
AI๊ฐ€ ์šฐ๋ฆฌ๋ฅผ ์ฃฝ์ด๋ ค๊ณ  ํ•˜๋Š” ๊ฒƒ์ผ๊นŒ์š”?
01:01
Or is it trying to do what we asked, and there was a problem?
19
61027
4310
์•„๋‹ˆ๋ฉด ์šฐ๋ฆฌ๊ฐ€ ์š”์ฒญํ–ˆ๋˜ ๊ฑธ ํ•˜๋ ค ํ–ˆ์ง€๋งŒ ๋ฌธ์ œ๊ฐ€ ์žˆ์—ˆ๋˜ ๊ฑธ๊นŒ์š”?
01:06
In movies, when something goes wrong with AI,
20
66567
2464
์˜ํ™”์—์„œ AI์™€ ๊ด€๋ จํ•ด ๋ญ”๊ฐ€ ์ž˜๋ชป๋˜๋ฉด,
01:09
it's usually because the AI has decided
21
69055
2712
๋ณดํ†ต์€ AI๊ฐ€ ์ธ๊ฐ„์—๊ฒŒ ๋” ์ด์ƒ ๋ณต์ข…ํ•˜๊ธฐ ์‹ซ๋‹ค๊ณ  ๊ฒฐ์ •ํ•˜๊ณ 
01:11
that it doesn't want to obey the humans anymore,
22
71791
2272
01:14
and it's got its own goals, thank you very much.
23
74087
2623
AI ์Šค์Šค๋กœ์˜ ๋ชฉํ‘œ๋ฅผ ๊ฐ€์ง€๊ธฐ ๋•Œ๋ฌธ์ด์ฃ , ์•„์ฃผ ๊ฐ์‚ฌํ•˜๊ฒŒ๋„์š”.
01:17
In real life, though, the AI that we actually have
24
77266
3216
์‹ค์ œ ์ƒํ™ฉ์—์„œ๋Š” ๊ทธ๋Ÿฌ๋‚˜, ์‹ค์ œ AI๋Š”
01:20
is not nearly smart enough for that.
25
80506
1863
์ „ํ˜€ ๊ทธ๋ ‡๊ฒŒ ๋˜‘๋˜‘ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
01:22
It has the approximate computing power
26
82781
2982
AI์˜ ์—ฐ์‚ฐ ๋Šฅ๋ ฅ์€ ๋Œ€๋žต ์ง€๋ ์ด ์ •๋„,
01:25
of an earthworm,
27
85787
1276
01:27
or maybe at most a single honeybee,
28
87087
3403
์•„๋งˆ ๊ธฐ๊ปํ•ด๋ด์•ผ ๊ฟ€๋ฒŒ ํ•œ๋งˆ๋ฆฌ,
01:30
and actually, probably maybe less.
29
90514
2215
์‚ฌ์‹ค, ์•„๋งˆ ๊ทธ๊ฒƒ๋„ ์•ˆ ๋  ๊ฒ๋‹ˆ๋‹ค.
01:32
Like, we're constantly learning new things about brains
30
92753
2594
์šฐ๋ฆฌ๋Š” ๋Š์ž„์—†์ด ๋‘๋‡Œ์— ๋Œ€ํ•ด ์ƒˆ๋กœ์šด ๊ฒƒ์„ ๋ฐฐ์›Œ์„œ
01:35
that make it clear how much our AIs don't measure up to real brains.
31
95371
4360
AI๊ฐ€ ์‹ค์ œ ๋‘๋‡Œ์— ์–ผ๋งˆ๋‚˜ ๋ชป๋ฏธ์น˜๋Š”์ง€ ๋ถ„๋ช…ํžˆ ์•Œ ์ˆ˜ ์žˆ์ฃ .
01:39
So today's AI can do a task like identify a pedestrian in a picture,
32
99755
5663
์˜ค๋Š˜๋‚ ์˜ AI๋Š” ์‚ฌ์ง„ ์†์˜ ํ–‰์ธ์„ ์‹๋ณ„ํ•˜๊ธฐ ๊ฐ™์€ ์ผ์„ ํ•  ์ˆ˜ ์žˆ์ฃ .
01:45
but it doesn't have a concept of what the pedestrian is
33
105442
2983
๊ทธ๋Ÿฌ๋‚˜ AI๋Š” ํ–‰์ธ์ด ๋ฌด์—‡์ด๋ผ๋Š” ๊ฐœ๋…์€ ๊ฐ€์ง€๊ณ  ์žˆ์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
01:48
beyond that it's a collection of lines and textures and things.
34
108449
4824
๊ทธ๊ฒƒ์€ ์„ ๊ณผ ์งˆ๊ฐ๊ฐ™์€ ๊ฒƒ๋“ค์˜ ๋ฉ์–ด๋ฆฌ ๋„ˆ๋จธ์˜ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
01:53
It doesn't know what a human actually is.
35
113792
2521
AI๋Š” ์‹ค์ œ ์ธ๊ฐ„์ด๋ผ๋Š” ๊ฒŒ ๋ฌด์—‡์ธ์ง€ ์•Œ์ง€ ๋ชปํ•ฉ๋‹ˆ๋‹ค.
01:56
So will today's AI do what we ask it to do?
36
116822
3282
๊ทธ๋ž˜์„œ ์˜ค๋Š˜๋‚ ์˜ AI๋Š” ์šฐ๋ฆฌ๊ฐ€ ์š”์ฒญํ•œ ๊ฒƒ์„ ์ˆ˜ํ–‰ํ• ๊นŒ์š”?
02:00
It will if it can,
37
120128
1594
ํ•  ์ˆ˜ ์žˆ๋‹ค๋ฉด ํ•˜๊ฒ ์ฃ ,
02:01
but it might not do what we actually want.
38
121746
2726
๊ทธ๋Ÿฌ๋‚˜ ์šฐ๋ฆฌ๊ฐ€ ์ง„์งœ ์›ํ•˜๋Š” ๊ฒƒ์„ ํ•˜์ง€ ์•Š์„ ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค.
02:04
So let's say that you were trying to get an AI
39
124496
2415
๊ฐ€๋ น ์—ฌ๋Ÿฌ๋ถ„์ด AI๋ฅผ ์ด์šฉํ•ด์„œ ์ด ๋กœ๋ด‡ ๋ถ€ํ’ˆ๋“ค๋กœ
02:06
to take this collection of robot parts
40
126935
2619
02:09
and assemble them into some kind of robot to get from Point A to Point B.
41
129578
4197
๋กœ๋ด‡์„ ์กฐ๋ฆฝํ•ด์„œ A์—์„œ B๋กœ ๊ฐ„๋‹ค๊ณ  ์ƒ๊ฐํ•ด ๋ด…์‹œ๋‹ค.
02:13
Now, if you were going to try and solve this problem
42
133799
2481
๋งŒ์•ฝ ์—ฌ๋Ÿฌ๋ถ„์ด ์ด ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด
์ „ํ†ต์ ์ธ ๋ฐฉ์‹์˜ ์ปดํ“จํ„ฐ ํ”„๋กœ๊ทธ๋ž˜๋ฐ์„ ์ด์šฉํ•œ๋‹ค๋ฉด,
02:16
by writing a traditional-style computer program,
43
136304
2351
02:18
you would give the program step-by-step instructions
44
138679
3417
์ด ํ”„๋กœ๊ทธ๋žจ์— ๋‹จ๊ณ„๋ณ„ ์ง€์‹œ๋ฅผ ์ฃผ๊ฒ ์ฃ .
02:22
on how to take these parts,
45
142120
1329
๋ถ€ํ’ˆ๋“ค์€ ๋‹ค๋ฃจ๋Š” ๋ฐฉ๋ฒ•์ด๋ผ๋“ ๊ฐ€, ๋‹ค๋ฆฌ๊ฐ€ ์žˆ๋Š” ๋กœ๋ด‡์œผ๋กœ ๋งŒ๋“œ๋Š” ๋ฐฉ๋ฒ•๊ณผ,
02:23
how to assemble them into a robot with legs
46
143473
2407
02:25
and then how to use those legs to walk to Point B.
47
145904
2942
๊ทธ ๋‹ค๋ฆฌ๋กœ B๊นŒ์ง€ ๊ฑธ์–ด๊ฐ€๋Š” ๋ฐฉ๋ฒ•์„์š”.
02:29
But when you're using AI to solve the problem,
48
149441
2340
๊ทธ๋Ÿฌ๋‚˜ AI๋ฅผ ์ด์šฉํ•ด์„œ ์ด ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•œ๋‹ค๋ฉด,
02:31
it goes differently.
49
151805
1174
๋‹ค๋ฅธ ์ด์•ผ๊ธฐ๊ฐ€ ๋ฉ๋‹ˆ๋‹ค.
02:33
You don't tell it how to solve the problem,
50
153003
2382
์–ด๋–ป๊ฒŒ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š”์ง€์— ๋Œ€ํ•ด AI์—๊ฒŒ ์•Œ๋ ค์ฃผ์ง€ ์•Š๊ณ 
02:35
you just give it the goal,
51
155409
1479
์—ฌ๋Ÿฌ๋ถ„์€ ๊ทธ๋ƒฅ ๋ชฉํ‘œ๋ฅผ ์ค๋‹ˆ๋‹ค.
02:36
and it has to figure out for itself via trial and error
52
156912
3262
๊ทธ๋ฆฌ๊ณ  ๊ทธ๊ฒƒ์€ ์‹œํ–‰์ฐฉ์˜ค๋ฅผ ํ†ตํ•ด ์Šค์Šค๋กœ ๋ชฉํ‘œ์— ๋„๋‹ฌํ•  ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋‚ด๋Š” ๊ฒƒ์ด์ฃ .
02:40
how to reach that goal.
53
160198
1484
02:42
And it turns out that the way AI tends to solve this particular problem
54
162254
4102
AI๊ฐ€ ์ด ํŠน์ •ํ•œ ๋ฌธ์ œ๋ฅผ ํ‘ธ๋Š” ๋ฐฉ์‹์€
02:46
is by doing this:
55
166380
1484
์ด๋ ‡์Šต๋‹ˆ๋‹ค.
02:47
it assembles itself into a tower and then falls over
56
167888
3367
์Šค์Šค๋กœ๋ฅผ ํƒ‘์œผ๋กœ ์กฐ๋ฆฝํ•œ ๋‹ค์Œ์— ์“ฐ๋Ÿฌ์ ธ์„œ B์— ๋–จ์–ด์ง€๋Š” ๊ฒƒ์ด์ฃ .
02:51
and lands at Point B.
57
171279
1827
02:53
And technically, this solves the problem.
58
173130
2829
๊ทธ๋ฆฌ๊ณ , ๋”ฐ์ง€๊ณ  ๋ณด๋ฉด, ์ด๊ฑด ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ธด ํ•ฉ๋‹ˆ๋‹ค.
02:55
Technically, it got to Point B.
59
175983
1639
๊ธฐ์ˆ ์ ์œผ๋กœ๋Š”, B๊นŒ์ง€ ๋„๋‹ฌํ•œ ๊ฒƒ์ด์ฃ .
02:57
The danger of AI is not that it's going to rebel against us,
60
177646
4265
AI์˜ ์œ„ํ—˜์€ ๊ทธ๊ฒƒ์ด ์šฐ๋ฆฌ์—๊ฒŒ ๋งž์„ค ๊ฒƒ์ด๋ผ๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ,
03:01
it's that it's going to do exactly what we ask it to do.
61
181935
4274
์šฐ๋ฆฌ๊ฐ€ ์š”์ฒญํ•œ ๊ฒƒ์„ ์•„์ฃผ ๊ทธ๋Œ€๋กœ ํ•  ๊ฒƒ์ด๋ผ๋Š” ๊ฒ๋‹ˆ๋‹ค.
03:06
So then the trick of working with AI becomes:
62
186876
2498
๋”ฐ๋ผ์„œ AI๋ฅผ ์‚ฌ์šฉํ•  ๋•Œ์˜ ์š”์ ์€
03:09
How do we set up the problem so that it actually does what we want?
63
189398
3828
AI๊ฐ€ ์šฐ๋ฆฌ๊ฐ€ ์›ํ•˜๋Š” ๊ฒฐ๊ณผ๋ฅผ ๋งŒ๋“ค๋„๋ก ์–ด๋–ป๊ฒŒ ๋ฌธ์ œ๋ฅผ ์„ค์ •ํ•˜๋Š๋ƒ์ž…๋‹ˆ๋‹ค.
03:14
So this little robot here is being controlled by an AI.
64
194726
3306
์—ฌ๊ธฐ ์ด ์กฐ๊ทธ๋งŒ ๋กœ๋ด‡์€ AI๊ฐ€ ์กฐ์ข…ํ•ฉ๋‹ˆ๋‹ค.
03:18
The AI came up with a design for the robot legs
65
198056
2814
AI๋Š” ๋กœ๋ด‡๋‹ค๋ฆฌ์˜ ๋””์ž์ธ์„ ์ƒ๊ฐํ•ด๋ƒˆ๊ณ 
03:20
and then figured out how to use them to get past all these obstacles.
66
200894
4078
๋ชจ๋“  ์žฅ์• ๋ฌผ๋“ค์„ ์ง€๋‚˜๊ฐ€๊ธฐ ์œ„ํ•ด ๋‹ค๋ฆฌ๋ฅผ ์ด์šฉํ•  ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋ƒˆ์Šต๋‹ˆ๋‹ค.
03:24
But when David Ha set up this experiment,
67
204996
2741
๊ทธ๋Ÿฌ๋‚˜ ๋ฐ์ด๋น„๋“œ ํ•˜์”จ๊ฐ€ ์ด ์‹คํ—˜์„ ๊ณ ์•ˆํ•  ๋•Œ,
03:27
he had to set it up with very, very strict limits
68
207761
2856
๊ทธ๋Š” ์•„์ฃผ, ์•„์ฃผ ์—„๊ฒฉํ•œ ์ œํ•œ์„ ์„ค์ •ํ•ด์•ผ๋งŒ ํ–ˆ์Šต๋‹ˆ๋‹ค.
03:30
on how big the AI was allowed to make the legs,
69
210641
3292
AI๊ฐ€ ๋งŒ๋“ค ์ˆ˜ ์žˆ๋Š” ๋‹ค๋ฆฌ์˜ ํฌ๊ธฐ์—์š”.
03:33
because otherwise ...
70
213957
1550
๊ทธ๋ ‡์ง€ ์•Š์•˜๋‹ค๋ฉด...
03:43
(Laughter)
71
223058
3931
(์›ƒ์Œ)
03:48
And technically, it got to the end of that obstacle course.
72
228563
3745
๊ทธ๋ฆฌ๊ณ  ์—„๋ฐ€ํžˆ ๋งํ•˜๋ฉด, ์ด๊ฒƒ์€ ์žฅ์• ๋ฌผ ์ฝ”์Šค๋ฅผ ํ†ต๊ณผํ–ˆ์Šต๋‹ˆ๋‹ค.
03:52
So you see how hard it is to get AI to do something as simple as just walk.
73
232332
4942
์ด์ฒ˜๋Ÿผ ๊ทธ๋ƒฅ ๊ฑท๋Š” ๊ฒƒ ๊ฐ™์€ ๊ฐ„๋‹จํ•œ ์ผ๋„ AI์—๊ฒŒ๋Š” ์–ด๋ ต์Šต๋‹ˆ๋‹ค.
03:57
So seeing the AI do this, you may say, OK, no fair,
74
237298
3820
๊ทธ๋ž˜์„œ AI๊ฐ€ ์ด๋Ÿฌ๋Š” ๊ฒƒ์„ ๋ณด๋ฉด, ์—ฌ๋Ÿฌ๋ถ„์€ ์•„๋งˆ ์ด๋ ‡๊ฒŒ ๋งํ•  ๊ฒ๋‹ˆ๋‹ค.
๊ทœ์น™์œ„๋ฐ˜์ด์•ผ, ๊ทธ๋ƒฅ ํฐ ํƒ‘์ด ๋ผ์„œ ๋„˜์–ด์ง€๋ฉด ์•ˆ๋ผ.
04:01
you can't just be a tall tower and fall over,
75
241142
2580
04:03
you have to actually, like, use legs to walk.
76
243746
3435
๊ฑธ์œผ๋ ค๋ฉด ๋‹ค๋ฆฌ๊ฐ™์€ ๊ฑธ ์จ์•ผ์ง€.
04:07
And it turns out, that doesn't always work, either.
77
247205
2759
๊ทธ๋Ÿฐ๋ฐ ๊ทธ ๋ฐฉ๋ฒ•๋„ ํ•ญ์ƒ ๋˜์ง€๋Š” ์•Š์ฃ .
04:09
This AI's job was to move fast.
78
249988
2759
์ด AI์˜ ๋ชฉํ‘œ๋Š” ๋น ๋ฅด๊ฒŒ ์›€์ง์ด๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
04:13
They didn't tell it that it had to run facing forward
79
253115
3593
๊ทธ๋“ค์€ AI์—๊ฒŒ ์•ž์œผ๋กœ ๋‹ฌ๋ ค์•ผ ํ•˜๊ณ ,
04:16
or that it couldn't use its arms.
80
256732
2258
ํŒ”์„ ์‚ฌ์šฉํ•˜๋ฉด ์•ˆ๋œ๋‹ค๊ณ  ์•Œ๋ ค์ฃผ์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค.
04:19
So this is what you get when you train AI to move fast,
81
259487
4618
AI์—๊ฒŒ ๋นจ๋ฆฌ ์›€์ง์ด๋Š” ๊ฒƒ์„ ํ›ˆ๋ จ์‹œํ‚ค๋ฉด ์ด๋Ÿฐ ๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
04:24
you get things like somersaulting and silly walks.
82
264129
3534
๊ณต์ค‘์ œ๋น„๋ฅผ ํ•˜๊ฑฐ๋‚˜ ๋ฐ”๋ณด๊ฐ™์€ ๊ฑธ์Œ๊ฐ™์€ ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์ฃ .
04:27
It's really common.
83
267687
1400
์ด๊ฑด ์•„์ฃผ ํ”ํ•ฉ๋‹ˆ๋‹ค.
04:29
So is twitching along the floor in a heap.
84
269667
3179
๋ฐ”๋‹ฅ์—์„œ ์›…ํฌ๋ฆฌ๊ณ  ์”ฐ๋ฃฉ๊ฑฐ๋ฆฌ๋Š” ๊ฒƒ๋„์š”.
04:32
(Laughter)
85
272870
1150
(์›ƒ์Œ)
04:35
So in my opinion, you know what should have been a whole lot weirder
86
275241
3254
๊ทธ๋ž˜์„œ ์ œ ์ƒ๊ฐ์—๋Š”, ๋” ์ด์ƒํ–ˆ์–ด์•ผ ํ–ˆ๋˜ ๊ฒƒ์€
04:38
is the "Terminator" robots.
87
278519
1396
โ€œํ„ฐ๋ฏธ๋„ค์ดํ„ฐโ€ ๋กœ๋ด‡์ž…๋‹ˆ๋‹ค.
04:40
Hacking "The Matrix" is another thing that AI will do if you give it a chance.
88
280256
3755
โ€œ๋งคํŠธ๋ฆญ์Šคโ€๋ฅผ ํ•ดํ‚นํ•˜๋Š” ๊ฒƒ์€ ๊ธฐํšŒ๋ฅผ ์ฃผ๋ฉด AI๊ฐ€ ํ•  ๋˜๋‹ค๋ฅธ ์ผ์ด์ฃ .
04:44
So if you train an AI in a simulation,
89
284035
2517
๋งŒ์•ฝ ์—ฌ๋Ÿฌ๋ถ„์ด ์‹œ๋ฎฌ๋ ˆ์ด์…˜์—์„œ AI๋ฅผ ํ›ˆ๋ จ์‹œํ‚จ๋‹ค๋ฉด,
04:46
it will learn how to do things like hack into the simulation's math errors
90
286576
4113
์‹œ๋ฎฌ๋ ˆ์ด์…˜์˜ ์ˆ˜ํ•™์  ์˜ค๋ฅ˜๋“ค์„ ํ•ดํ‚นํ•˜๋Š” ๊ฒƒ๊ณผ ๊ฐ™์€ ๋ฐฉ๋ฒ•์„ ๋ฐฐ์›Œ์„œ
04:50
and harvest them for energy.
91
290713
2207
๊ทธ๊ฒƒ๋“ค์„ ํ†ตํ•ด ์—๋„ˆ์ง€๋ฅผ ์–ป์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
04:52
Or it will figure out how to move faster by glitching repeatedly into the floor.
92
292944
5475
์•„๋‹ˆ๋ฉด ๋ฐ˜๋ณต์ ์œผ๋กœ ๋ฐ”๋‹ฅ์— ๊ณ ์žฅ์„ ๋‚ด ๋” ๋นจ๋ฆฌ ์›€์ง์ด๋Š” ๋ฐฉ๋ฒ•์„ ์•Œ์•„๋‚ด๊ฒ ์ฃ .
04:58
When you're working with AI,
93
298443
1585
AI์™€ ์ผํ•˜๋Š” ๊ฒƒ์€ ์‚ฌ๋žŒ๊ณผ ์ผํ•˜๋Š” ๊ฒƒ๊ณผ ๋‹ค๋ฅด๊ณ ,
05:00
it's less like working with another human
94
300052
2389
05:02
and a lot more like working with some kind of weird force of nature.
95
302465
3629
์ž์—ฐ์˜ ์–ด๋–ค ์ด์ƒํ•œ ํž˜๊ณผ ์ผํ•˜๋Š” ๊ฒƒ๊ณผ ๋” ์œ ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
05:06
And it's really easy to accidentally give AI the wrong problem to solve,
96
306562
4623
์‹ค์ˆ˜๋กœ AI์—๊ฒŒ ์ž˜๋ชป๋œ ๋ฌธ์ œ๋ฅผ ์ฃผ๋Š” ๊ฒƒ๋„ ์‰ฌ์šด ์ผ์ž…๋‹ˆ๋‹ค.
05:11
and often we don't realize that until something has actually gone wrong.
97
311209
4538
๋ณดํ†ต ์‹ค์ œ๋กœ ์ผ์ด ์ž˜๋ชป๋  ๋•Œ๊นŒ์ง€ ์šฐ๋ฆฌ๋Š” ๊ทธ๊ฑธ ์•Œ์•„์ฑ„์ง€ ๋ชปํ•˜์ฃ .
05:16
So here's an experiment I did,
98
316242
2080
์ œ๊ฐ€ ์ง„ํ–‰ํ•œ ์‹คํ—˜์ด ํ•˜๋‚˜ ์žˆ์Šต๋‹ˆ๋‹ค.
05:18
where I wanted the AI to copy paint colors,
99
318346
3182
์ €๋Š” AI๊ฐ€ ๋ฌผ๊ฐ ์ƒ‰๋“ค์„ ๋ณต์‚ฌํ•ด์„œ
05:21
to invent new paint colors,
100
321552
1746
์ƒˆ๋กœ์šด ๋ฌผ๊ฐ์„ ๋งŒ๋“ค์–ด๋‚ด๊ธฐ๋ฅผ ์›ํ–ˆ์Šต๋‹ˆ๋‹ค.
05:23
given the list like the ones here on the left.
101
323322
2987
์—ฌ๊ธฐ ์™ผ์ชฝ์— ์žˆ๋Š” ๋ชฉ๋ก์„ ์ฃผ๊ณ ์š”.
05:26
And here's what the AI actually came up with.
102
326798
3004
์ด๊ฒŒ AI๊ฐ€ ์‹ค์ œ๋กœ ๋งŒ๋“ค์–ด๋‚ธ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
05:29
[Sindis Poop, Turdly, Suffer, Gray Pubic]
103
329826
3143
[์‹ ๋””์Šค ๋˜ฅ, ๋˜ฅ๋ฉ์–ด๋ฆฌ๊ฐ™์€, ๊ณ ์ƒํ•˜๋‹ค, ํšŒ์ƒ‰ ์Œ๋ถ€]
05:32
(Laughter)
104
332993
4230
(์›ƒ์Œ)
05:39
So technically,
105
339177
1886
๊ทธ๋ž˜์„œ ์—„๋ฐ€ํžˆ ๋งํ•˜๋ฉด,
05:41
it did what I asked it to.
106
341087
1864
์ œ๊ฐ€ ์š”์ฒญํ•œ ๊ฒƒ์„ ํ•˜๊ธด ํ–ˆ์Šต๋‹ˆ๋‹ค.
05:42
I thought I was asking it for, like, nice paint color names,
107
342975
3308
์ €๋Š” ์ œ๊ฐ€ ๋ฉ‹์ง„ ๋ฌผ๊ฐ ์ด๋ฆ„๋“ค์„ ์š”์ฒญํ–ˆ๋‹ค๊ณ  ์ƒ๊ฐํ–ˆ๋Š”๋ฐ
05:46
but what I was actually asking it to do
108
346307
2307
์‹ค์ œ๋กœ ์ œ๊ฐ€ ์š”์ฒญํ–ˆ๋˜ ๊ฒƒ์€
05:48
was just imitate the kinds of letter combinations
109
348638
3086
์›๋ž˜์— ์žˆ๋˜ ๊ฒƒ๋“ค์— ๋ณด์ด๋Š” ๋ฌธ์ž์กฐํ•ฉ์„ ๊ทธ๋ƒฅ ๋ชจ๋ฐฉํ•˜๋Š” ๊ฒƒ์ด์—ˆ์Šต๋‹ˆ๋‹ค.
05:51
that it had seen in the original.
110
351748
1905
05:53
And I didn't tell it anything about what words mean,
111
353677
3098
๊ทธ ๋‹จ์–ด๋“ค์˜ ๋œป์ด ๋ฌด์—‡์ธ์ง€์— ๋Œ€ํ•ด ์•Œ๋ ค์ฃผ์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค.
05:56
or that there are maybe some words
112
356799
2560
ํ˜น์€ ๋ฌผ๊ฐ์— ์‚ฌ์šฉํ•˜๋ฉด ์•ˆ๋˜๋Š” ๋‹จ์–ด๋„ ์žˆ์„ ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ๋„์š”.
05:59
that it should avoid using in these paint colors.
113
359383
2889
06:03
So its entire world is the data that I gave it.
114
363141
3494
AI๊ฐ€ ์•Œ๊ณ  ์žˆ๋Š” ์„ธ๊ณ„๋Š” ์ œ๊ฐ€ ์ค€ ๋ฐ์ดํ„ฐ๊ฐ€ ์ „๋ถ€์˜€์ง€์š”.
06:06
Like with the ice cream flavors, it doesn't know about anything else.
115
366659
4028
์•„์ด์Šคํฌ๋ฆผ ๋ง›์ฒ˜๋Ÿผ, ๋‹ค๋ฅธ ๊ฒƒ์— ๋Œ€ํ•ด์„œ๋Š” ์ „ํ˜€ ์•„๋Š” ๊ฒƒ์ด ์—†์—ˆ์Šต๋‹ˆ๋‹ค.
06:12
So it is through the data
116
372491
1638
๊ทธ๋ž˜์„œ ์ด ๋ฐ์ดํ„ฐ๋ฅผ ํ†ตํ•ด์„œ
06:14
that we often accidentally tell AI to do the wrong thing.
117
374153
4044
์šฐ๋ฆฌ๋Š” AI์—๊ฒŒ ์ž˜๋ชป๋œ ๊ฒƒ์„ ํ•˜๋ผ๊ณ  ์ข…์ข… ์‹ค์ˆ˜๋กœ ๋งํ•ฉ๋‹ˆ๋‹ค.
06:18
This is a fish called a tench.
118
378694
3032
์ด๊ฑด ์ž‰์–ด๋ผ๊ณ  ๋ถˆ๋ฆฌ๋Š” ๋ฌผ๊ณ ๊ธฐ์ž…๋‹ˆ๋‹ค.
06:21
And there was a group of researchers
119
381750
1815
์—ฐ๊ตฌ์ž๋“ค์ด AI๋ฅผ ํ›ˆ๋ จ์‹œ์ผœ ์‚ฌ์ง„์—์„œ ์ž‰์–ด๋ฅผ ์‹๋ณ„ํ•˜๋„๋ก ํ–ˆ์Šต๋‹ˆ๋‹ค.
06:23
who trained an AI to identify this tench in pictures.
120
383589
3874
06:27
But then when they asked it
121
387487
1296
๊ทธ๋Ÿฌ๋‚˜ ๊ทธ๋“ค์ด AI์—๊ฒŒ ์‚ฌ์ง„์˜ ์–ด๋–ค ๋ถ€๋ถ„์„
06:28
what part of the picture it was actually using to identify the fish,
122
388807
3426
๋ฌผ๊ณ ๊ธฐ๋ฅผ ์‹๋ณ„ํ•˜๋Š”๋ฐ ์ผ๋Š”์ง€ ๋ฌผ์–ด๋ณด์ž
06:32
here's what it highlighted.
123
392257
1358
์ด๊ฒƒ์ด ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค.
06:35
Yes, those are human fingers.
124
395203
2189
๋„ค, ์ €๊ฒƒ๋“ค์€ ์‚ฌ๋žŒ์˜ ์†๊ฐ€๋ฝ์ž…๋‹ˆ๋‹ค.
06:37
Why would it be looking for human fingers
125
397416
2059
์™œ ๋ฌผ๊ณ ๊ธฐ๋ฅผ ์‹๋ณ„ํ•˜๋Š”๋ฐ ์‚ฌ๋žŒ์˜ ์†๊ฐ€๋ฝ์„ ์ฐพ๊ณ  ์žˆ์„๊นŒ์š”?
06:39
if it's trying to identify a fish?
126
399499
1921
06:42
Well, it turns out that the tench is a trophy fish,
127
402126
3164
์ž‰์–ด๋Š” ๊ธฐ๋…์‚ฌ์ง„์œผ๋กœ ๋‚จ๊ธธ๋งŒํ•œ ๋ฌผ๊ณ ๊ธฐ์—ฌ์„œ,
06:45
and so in a lot of pictures that the AI had seen of this fish
128
405314
3811
AI๊ฐ€ ํ›ˆ๋ จ๋™์•ˆ ๋ณด์•˜๋˜ ์ด ๋ฌผ๊ณ ๊ธฐ์˜ ์‚ฌ์ง„๋“ค์€
06:49
during training,
129
409149
1151
06:50
the fish looked like this.
130
410324
1490
์ด๋Ÿฌํ–ˆ์Šต๋‹ˆ๋‹ค.
06:51
(Laughter)
131
411838
1635
(์›ƒ์Œ)
06:53
And it didn't know that the fingers aren't part of the fish.
132
413497
3330
์†๊ฐ€๋ฝ์ด ๋ฌผ๊ณ ๊ธฐ์˜ ์ผ๋ถ€๊ฐ€ ์•„๋‹ˆ๋ผ๋Š” ๊ฒƒ์„ ๋ชฐ๋ž์ฃ .
06:58
So you see why it is so hard to design an AI
133
418808
4120
๊ทธ๋ž˜์„œ ์—ฌ๋Ÿฌ๋ถ„์€ ์ง„์งœ๋กœ ๋ฌด์—‡์„ ์ฐพ๊ณ  ์žˆ๋Š”์ง€ ์ดํ•ดํ•  ์ˆ˜ ์žˆ๋Š”
07:02
that actually can understand what it's looking at.
134
422952
3319
AI๋ฅผ ๋””์ž์ธํ•˜๋Š” ๊ฒƒ์ด ์™œ ์–ด๋ ค์šด์ง€ ์•Œ ์ˆ˜ ์žˆ์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
07:06
And this is why designing the image recognition
135
426295
2862
๊ทธ๋ฆฌ๊ณ  ์ด๊ฒƒ์ด ์ž์œจ์ฃผํ–‰ ์ž๋™์ฐจ์—์„œ์˜ ์ด๋ฏธ์ง€ ์ธ์‹์„
07:09
in self-driving cars is so hard,
136
429181
2067
๋””์ž์ธํ•˜๋Š” ๊ฒƒ์ด ์™œ ๊ทธ๋ ‡๊ฒŒ ํž˜๋“  ์ผ์ธ ์ง€์— ๋Œ€ํ•œ ์ด์œ ์ด๊ณ 
07:11
and why so many self-driving car failures
137
431272
2205
๋งŽ์€ ์ž์œจ์ฃผํ–‰ ์ž๋™์ฐจ์˜ ์‹คํŒจ๋“ค์˜ ์ด์œ ๋Š” AI๊ฐ€ ํ˜ผ๋ž€์Šค๋Ÿฌ์›Œํ–ˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.
07:13
are because the AI got confused.
138
433501
2885
07:16
I want to talk about an example from 2016.
139
436410
4008
2016๋…„์— ์žˆ์—ˆ๋˜ ํ•œ ์˜ˆ์— ๋Œ€ํ•ด ์ด์•ผ๊ธฐํ•ด๋ณด๊ณ  ์‹ถ์Šต๋‹ˆ๋‹ค.
07:20
There was a fatal accident when somebody was using Tesla's autopilot AI,
140
440442
4455
ํ…Œ์Šฌ๋ผ ์ž๋™์กฐ์ข… AI๋ฅผ ์ด์šฉํ•˜๋˜ ์‚ฌ๋žŒ์ด ์•„์ฃผ ์น˜๋ช…์ ์ธ ์‚ฌ๊ณ ๋ฅผ ๋‹นํ–ˆ์Šต๋‹ˆ๋‹ค.
07:24
but instead of using it on the highway like it was designed for,
141
444921
3414
์›๋ž˜ ๊ณ ์†๋„๋กœ์—์„œ ์‚ฌ์šฉํ•˜๋„๋ก ๋””์ž์ธ๋˜์—ˆ๋Š”๋ฐ
07:28
they used it on city streets.
142
448359
2205
๋„์‹œ ๋„๋กœ์—์„œ ์‚ฌ์šฉ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
07:31
And what happened was,
143
451239
1175
๊ทธ๋ฆฌ๊ณ  ์–ด๋–ค ์ผ์ด ์ผ์–ด๋‚ฌ๋ƒ๋ฉด,
07:32
a truck drove out in front of the car and the car failed to brake.
144
452438
3396
์–ด๋–ค ํŠธ๋Ÿญ์ด ์ฐจ ์•ž์œผ๋กœ ๋‚˜์™”๋Š”๋ฐ, ๊ทธ ์ฐจ๋Š” ์„œ์ง€ ๋ชปํ–ˆ์Šต๋‹ˆ๋‹ค.
07:36
Now, the AI definitely was trained to recognize trucks in pictures.
145
456507
4762
AI๋Š” ๋ถ„๋ช…ํžˆ ์‚ฌ์ง„๋“ค์—์„œ ํŠธ๋Ÿญ์„ ์ธ์‹ํ•˜๋„๋ก ํ›ˆ๋ จ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
07:41
But what it looks like happened is
146
461293
2145
๊ทธ๋Ÿฌ๋‚˜ ์ข€๋” ๋“ค์—ฌ๋‹ค ๋ณด๋ฉด
07:43
the AI was trained to recognize trucks on highway driving,
147
463462
2931
AI๊ฐ€ ๊ณ ์†๋„๋กœ์— ์žˆ๋Š” ํŠธ๋Ÿญ๋“ค์„ ์ธ์‹ํ•˜๋„๋ก ํ›ˆ๋ จ๋œ ๊ฑฐ ๊ฐ™์•„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค.
07:46
where you would expect to see trucks from behind.
148
466417
2899
ํŠธ๋Ÿญ์˜ ๋’ท๋ชจ์Šต์„ ๋ณผ ๊ฒƒ์ด๋ผ๊ณ  ์ƒ๊ฐํ•˜๋„๋ก ๋ง์ด์ฃ .
07:49
Trucks on the side is not supposed to happen on a highway,
149
469340
3420
ํŠธ๋Ÿญ์˜ ์˜†๋ชจ์Šต์„ ๋ณด๋Š” ๊ฒƒ์€ ๊ณ ์†๋„๋กœ์—๋Š” ์ผ์–ด๋‚˜์ง€ ์•Š๋Š” ์ผ์ด์—ˆ๊ณ ,
07:52
and so when the AI saw this truck,
150
472784
3455
์ด AI๊ฐ€ ์ด ํŠธ๋Ÿญ์„ ๋ดค์„ ๋•,
07:56
it looks like the AI recognized it as most likely to be a road sign
151
476263
4827
์•„๋งˆ๋„ ํŠธ๋Ÿญ์„ ๋„๋กœ ํ‘œ์ง€ํŒ์œผ๋กœ ์ธ์‹ํ•˜๊ณ 
08:01
and therefore, safe to drive underneath.
152
481114
2273
๋”ฐ๋ผ์„œ ์šด์ „์„ ํ•ด๋„ ์•ˆ์ „ํ•˜๋‹ค๊ณ  ํŒ๋‹จํ•œ ๊ฒƒ์ด์ฃ .
08:04
Here's an AI misstep from a different field.
153
484114
2580
์—ฌ๊ธฐ, ๋‹ค๋ฅธ ๋ถ„์•ผ์—์„œ AI์˜ ์‹ค์ˆ˜๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
08:06
Amazon recently had to give up on a rรฉsumรฉ-sorting algorithm
154
486718
3460
์•„๋งˆ์กด์€ ์ด๋ ฅ์„œ๋ถ„๋ฅ˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ํฌ๊ธฐํ•ด์•ผ๋งŒ ํ–ˆ์Šต๋‹ˆ๋‹ค.
08:10
that they were working on
155
490202
1220
์‹ค์ œ ์ ์šฉ์—์„œ ์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ์—ฌ์„ฑ์„ ์ฐจ๋ณ„ํ•œ๋‹ค๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ•œ ๊ฒƒ์ด์ฃ .
08:11
when they discovered that the algorithm had learned to discriminate against women.
156
491446
3908
๊ทธ๋“ค์ด AI ํ›ˆ๋ จ์šฉ์œผ๋กœ ์‚ฌ์šฉํ•œ ์ด๋ ฅ์„œ๋Š”
08:15
What happened is they had trained it on example rรฉsumรฉs
157
495378
2716
08:18
of people who they had hired in the past.
158
498118
2242
๊ณผ๊ฑฐ์— ๊ณ ์šฉํ•œ ์‚ฌ๋žŒ๋“ค์˜ ๊ฒƒ์ด์—ˆ์Šต๋‹ˆ๋‹ค.
08:20
And from these examples, the AI learned to avoid the rรฉsumรฉs of people
159
500384
4023
๊ทธ ์˜ˆ์‹œ๋“ค๋กœ๋ถ€ํ„ฐ, AI๋Š” ์ด๋Ÿฐ ์‚ฌ๋žŒ๋“ค์„ ๊ธฐํ”ผํ•˜๋Š” ๊ฒƒ์„ ๋ฐฐ์› ์Šต๋‹ˆ๋‹ค.
08:24
who had gone to women's colleges
160
504431
2026
์—ฌ๋Œ€๋ฅผ ๋‚˜์˜จ ์‚ฌ๋žŒ๋“ค,
08:26
or who had the word "women" somewhere in their resume,
161
506481
2806
์ด๋ ฅ์„œ ์–ด๋”˜๊ฐ€์— โ€˜์—ฌ์„ฑโ€™์ด๋ผ๋Š” ๋‹จ์–ด๊ฐ€ ์žˆ๋Š” ์‚ฌ๋žŒ๋“ค,
08:29
as in, "women's soccer team" or "Society of Women Engineers."
162
509311
4576
์˜ˆ๋ฅผ ๋“ค์–ด '์—ฌ์ž ์ถ•๊ตฌํŒ€', '์—ฌ์„ฑ๊ณตํ•™์žํ˜‘ํšŒ'๊ฐ™์€ ๋‹จ์–ด๋ง์ด์ฃ .
08:33
The AI didn't know that it wasn't supposed to copy this particular thing
163
513911
3974
AI๋Š” ์ด ํŠน์ •ํ•œ ํ–‰๋™์„ ๋”ฐ๋ผ ํ•ด์„  ์•ˆ๋œ๋‹ค๋Š” ๊ฒƒ์„ ๋ชจ๋ฅด๊ณ  ์žˆ์—ˆ์ฃ .
08:37
that it had seen the humans do.
164
517909
1978
์ธ๊ฐ„๋“ค์ด ํ•˜๋Š” ๊ฑธ ๋ดค๋”๋ผ๋„์š”.
08:39
And technically, it did what they asked it to do.
165
519911
3177
๊ทธ๋ฆฌ๊ณ  ์—„๋ฐ€ํžˆ ๋งํ•˜์ž๋ฉด, AI๋Š” ์•„๋งˆ์กด์ด ์š”์ฒญํ•œ ๊ฒƒ์„ ํ–ˆ์Šต๋‹ˆ๋‹ค.
08:43
They just accidentally asked it to do the wrong thing.
166
523112
2797
๊ทธ๋“ค์€ ๊ทธ์ € ์‹ค์ˆ˜๋กœ ์ž˜๋ชป๋œ ์ผ์„ ์‹œํ‚จ ๊ฒƒ์ด์ฃ .
08:46
And this happens all the time with AI.
167
526653
2895
AI์—๊ฒŒ ์ด๋Ÿฐ ์ผ์€ ํ•ญ์ƒ ์ผ์–ด๋‚ฉ๋‹ˆ๋‹ค.
08:50
AI can be really destructive and not know it.
168
530120
3591
AI๋Š” ์•„์ฃผ ํ•ด๋กœ์šด ์ผ์„ ํ•˜๋Š” ์™€์ค‘์—, ํ•ด๋กญ๋‹ค๋Š” ๊ฒƒ์„ ๋ชจ๋ฅผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
08:53
So the AIs that recommend new content in Facebook, in YouTube,
169
533735
5078
ํŽ˜์ด์Šค๋ถ, ์œ ํŠœ๋ธŒ์—์„œ ์ƒˆ๋กœ์šด ์ฝ˜ํ…์ธ ๋ฅผ ์ถ”์ฒœํ•ด์ฃผ๋Š” AI๋“ค์€
08:58
they're optimized to increase the number of clicks and views.
170
538837
3539
ํด๋ฆญ ์ˆ˜์™€ ์กฐํšŒ ์ˆ˜๋ฅผ ๋Š˜๋ฆฌ๋„๋ก ์ตœ์ ํ™”๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
09:02
And unfortunately, one way that they have found of doing this
171
542400
3436
๊ทธ๋ฆฌ๊ณ  ๋ถˆํ–‰ํ•˜๊ฒŒ๋„, ๊ทธ๋“ค์ด ์ฐพ์€ ๋ฐฉ๋ฒ•์€
09:05
is to recommend the content of conspiracy theories or bigotry.
172
545860
4503
์Œ๋ชจ๋ก ์ด๋‚˜ ์‹ฌํ•œ ํŽธ๊ฒฌ์ด ์žˆ๋Š” ์ฝ˜ํ…์ธ ๋ฅผ ์ถ”์ฒœํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
09:10
The AIs themselves don't have any concept of what this content actually is,
173
550902
5302
AI๋“ค ์ž์ฒด์—๋Š” ์ด ์ฝ˜ํ…์ธ ๋“ค์ด ์‹ค์ œ๋กœ ๋ฌด์—‡์ธ์ง€์— ๋Œ€ํ•œ ๊ฐœ๋…์ด ์—†์Šต๋‹ˆ๋‹ค.
09:16
and they don't have any concept of what the consequences might be
174
556228
3395
๊ทธ๋ฆฌ๊ณ  ๊ทธ๋“ค์€ ๊ฒฐ๊ณผ๊ฐ€ ์–ด๋–จ ๊ฒƒ์ธ์ง€์— ๋Œ€ํ•œ ๊ฐœ๋…๋„ ์—†์Šต๋‹ˆ๋‹ค.
09:19
of recommending this content.
175
559647
2109
์ด๋Ÿฌํ•œ ์ฝ˜ํ…์ธ ๋ฅผ ์ถ”์ฒœํ•ด์คŒ์œผ๋กœ์จ ๋ฐœ์ƒ๋  ๊ฒฐ๊ณผ์š”.
09:22
So, when we're working with AI,
176
562296
2011
๊ทธ๋ž˜์„œ ์šฐ๋ฆฌ๊ฐ€ AI์™€ ์ผํ•  ๋•Œ,
09:24
it's up to us to avoid problems.
177
564331
4182
๋ฌธ์ œ๋“ค์„ ํ”ผํ•˜๋Š” ๊ฒƒ์€ ์šฐ๋ฆฌ์—๊ฒŒ ๋‹ฌ๋ ค์žˆ์Šต๋‹ˆ๋‹ค.
09:28
And avoiding things going wrong,
178
568537
2323
์ผ๋“ค์ด ์ž˜๋ชป๋˜๋Š” ๊ฒƒ์„ ํ”ผํ•˜๋Š” ๊ฒƒ์€,
09:30
that may come down to the age-old problem of communication,
179
570884
4526
์•„์ฃผ ์˜ค๋ž˜๋œ ์†Œํ†ต์˜ ๋ฌธ์ œ๋กœ ์ด์–ด์ง‘๋‹ˆ๋‹ค.
09:35
where we as humans have to learn how to communicate with AI.
180
575434
3745
์ธ๊ฐ„์ธ ์šฐ๋ฆฌ๊ฐ€ AI์™€ ์†Œํ†ตํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๋ฐฐ์›Œ์•ผํ•˜๋Š” ๊ฑฐ์ฃ .
09:39
We have to learn what AI is capable of doing and what it's not,
181
579203
4039
AI๊ฐ€ ํ•  ์ˆ˜ ์žˆ๋Š” ์ผ๊ณผ ์—†๋Š” ์ผ์ด ๋ฌด์—‡์ธ์ง€ ์•Œ์•„์•ผ ํ•ฉ๋‹ˆ๋‹ค.
09:43
and to understand that, with its tiny little worm brain,
182
583266
3086
๋˜ํ•œ AI๋Š” ๊ทธ ์กฐ๊ทธ๋งŒ ์ง€๋ ์ด๊ฐ™์€ ๋‡Œ๋กœ
09:46
AI doesn't really understand what we're trying to ask it to do.
183
586376
4013
์šฐ๋ฆฌ๊ฐ€ ์–ด๋–ค ๊ฒƒ์„ ์š”์ฒญํ•˜๋ ค๊ณ  ํ•˜๋Š”์ง€ ์ดํ•ดํ•˜์ง€ ๋ชปํ•œ๋‹ค๋Š” ๊ฒƒ์„ ์•Œ์•„์•ผํ•ฉ๋‹ˆ๋‹ค.
09:51
So in other words, we have to be prepared to work with AI
184
591148
3321
๋‹ค์‹œ ๋งํ•ด, ์šฐ๋ฆฌ๋Š” ์ค€๋น„ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
09:54
that's not the super-competent, all-knowing AI of science fiction.
185
594493
5258
์šฐ๋ฆฌ๊ฐ€ ์‚ฌ์šฉํ•  AI๋Š” ๊ณต์ƒ๊ณผํ•™์—๋‚˜ ์žˆ๋Š” ์ „์ง€์ „๋Šฅํ•œ AI๊ฐ€ ์•„๋‹™๋‹ˆ๋‹ค.
09:59
We have to be prepared to work with an AI
186
599775
2862
ํ˜„์žฌ ์šฐ๋ฆฌ๊ฐ€ ์‹ค์ œ๋กœ ๊ฐ–๊ณ  ์žˆ๋Š” AI์™€ ํ•จ๊ป˜ ์ผํ•˜๋„๋ก ์ค€๋น„ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
10:02
that's the one that we actually have in the present day.
187
602661
2938
10:05
And present-day AI is plenty weird enough.
188
605623
4205
๊ทธ๋ฆฌ๊ณ  ์˜ค๋Š˜๋‚ ์˜ AI๋Š” ๋Œ€๋‹จํžˆ ์ด์ƒํ•ฉ๋‹ˆ๋‹ค.
10:09
Thank you.
189
609852
1190
๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
10:11
(Applause)
190
611066
5225
(๋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