Hod Lipson: Robots that are "self-aware"

117,444 views ใƒป 2007-10-13

TED


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

๋ฒˆ์—ญ: Seyoung Yoon ๊ฒ€ํ† : Yenah Lee
00:25
So, where are the robots?
0
25000
2000
์ง€๊ธˆ ๋กœ๋ด‡๋“ค์€ ์–ด๋”” ์žˆ๋‚˜์š”?
00:27
We've been told for 40 years already that they're coming soon.
1
27000
3000
์ง€๋‚œ 40๋…„๊ฐ„ ์šฐ๋ฆฌ๋Š” ๋จธ์ง€์•Š์•„ ๋กœ๋ด‡๋“ค์„ ๋ณผ ์ˆ˜ ์žˆ์„ ๊ฒƒ์ด๋ผ๊ณ  ๋“ค์–ด ์™”์Šต๋‹ˆ๋‹ค.
00:30
Very soon they'll be doing everything for us.
2
30000
3000
๊ณง ์žˆ์œผ๋ฉด ๋กœ๋ด‡๋“ค์ด ๋Œ€์‹  ์š”๋ฆฌ๋ฅผ ํ•˜๊ณ  ๋ฐฉ ์ฒญ์†Œ๋ฅผ ํ•˜๊ณ  ๋ฌผํ’ˆ์„ ๊ตฌ์ž…ํ•˜๊ณ 
00:33
They'll be cooking, cleaning, buying things, shopping, building. But they aren't here.
3
33000
5000
์‡ผํ•‘์„ ํ•˜๊ณ  ๊ฑด์„ค ๋…ธ๋™์„ ํ•ด ์ค„ ๊ฒƒ์ด๋ผ๊ณ  ๋“ค์—ˆ์ฃ . ํ•˜์ง€๋งŒ ์•„์ง๊นŒ์ง€๋„ ์ƒ์ƒ ์†์˜ ์ด์•ผ๊ธฐ์ผ ๋ฟ์ž…๋‹ˆ๋‹ค.
00:38
Meanwhile, we have illegal immigrants doing all the work,
4
38000
4000
์•„์ง๊นŒ์ง€ ๋กœ๋ด‡๋“ค์˜ ๋…ธ๋™์€ ๋„์ž…๋˜์ง€๋„ ์•Š๊ณ , ๋Œ€์‹  ๋ถˆ๋ฒ• ์ด๋ฏผ์ž๋“ค์ด
00:42
but we don't have any robots.
5
42000
2000
์˜จ๊ฐ– ๋…ธ๋™์„ ๋งก์•„์„œ ํ•˜๊ณ  ์žˆ์ฃ .
00:44
So what can we do about that? What can we say?
6
44000
4000
์ด ๋ฌธ์ œ์— ๋Œ€ํ•ด ๋…ผ์˜๋ฅผ ํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค.
00:48
So I want to give a little bit of a different perspective
7
48000
4000
๋จผ์ € ์ด ๋ฌธ์ œ๋ฅผ ์ข€ ๋‹ค๋ฅธ ๊ด€์ ์—์„œ ์ ‘๊ทผํ•ด
00:52
of how we can perhaps look at these things in a little bit of a different way.
8
52000
6000
์—ฌ๋Ÿฌ๋ถ„๋“ค์ด ๋กœ๋ด‡์ด๋ผ๋Š” ๊ฒƒ์˜ ์ƒˆ๋กœ์šด ๋ฉด์„ ๋ณผ ์ˆ˜ ์žˆ๋„๋ก ํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
00:58
And this is an x-ray picture
9
58000
2000
์ง€๊ธˆ ๋ณด์‹œ๋Š” ๊ฒƒ์€ 88๋…„๋„์— ์ฐ์€
01:00
of a real beetle, and a Swiss watch, back from '88. You look at that --
10
60000
5000
๋”ฑ์ •๋ฒŒ๋ ˆ์™€ ์Šค์œ„์Šค ์‹œ๊ณ„์˜ X์„  ์‚ฌ์ง„์ž…๋‹ˆ๋‹ค.
01:05
what was true then is certainly true today.
11
65000
2000
๊ทธ ๋•Œ๋‚˜ ์ง€๊ธˆ์ด๋‚˜ ๋ณ€ํ•œ ๊ฒƒ์€ ๋ณ„๋กœ ์—†์Šต๋‹ˆ๋‹ค.
01:07
We can still make the pieces. We can make the right pieces.
12
67000
3000
์šฐ๋ฆฌ๋Š” ์—ฌ์ „ํžˆ ๊ธฐ๊ณ„ ๋ถ€ํ’ˆ๋“ค์„ ๋งŒ๋“ค๊ณ 
01:10
We can make the circuitry of the right computational power,
13
70000
3000
์—ฐ์‚ฐ ๋Šฅ๋ ฅ์„ ๊ฐ€์ง„ ์ „์ž ํšŒ๋กœ๋ฅผ ๋งŒ๋“ญ๋‹ˆ๋‹ค.
01:13
but we can't actually put them together to make something
14
73000
3000
ํ•˜์ง€๋งŒ ์—ฌ์ „ํžˆ ๋”ฑ์ •๋ฒŒ๋ ˆ์ฒ˜๋Ÿผ ๋ฌด์—‡์ธ๊ฐ€๋ฅผ ํ•˜๋Š” ๋™์‹œ์—
01:16
that will actually work and be as adaptive as these systems.
15
76000
5000
์ฃผ๋ณ€ ํ™˜๊ฒฝ์— ์ ์‘ํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ๊ณ„๋ฅผ ๋งŒ๋“ค์ง€๋Š” ๋ชปํ•˜์ฃ .
01:21
So let's try to look at it from a different perspective.
16
81000
2000
์ด์ œ ๋‹ค๋ฅธ ๊ด€์ ์—์„œ ํ•œ ๋ฒˆ ์‚ดํŽด๋ด…์‹œ๋‹ค.
01:23
Let's summon the best designer, the mother of all designers.
17
83000
4000
๊ฐ€์žฅ ๊ฐ•๋ ฅํ•œ ๋””์ž์ด๋„ˆ์ด์ž ๋ชจ๋“  ๋””์ž์ด๋„ˆ๋“ค์˜ ์Šค์Šน์ธ
01:27
Let's see what evolution can do for us.
18
87000
3000
'์ง„ํ™”'๋ฅผ ํ†ตํ•ด ํ•œ ๋ฒˆ ์‚ดํŽด๋ณด์ฃ .
01:30
So we threw in -- we created a primordial soup
19
90000
4000
์šฐ๋ฆฌ๋Š” ์›์‹œ ์ง€๊ตฌ์˜ ์ƒํƒœ๋ฅผ ๋ชจ๋ฐฉํ•ด ์‡  ๋ง‰๋Œ€๊ธฐ,
01:34
with lots of pieces of robots -- with bars, with motors, with neurons.
20
94000
4000
์ „๋™๊ธฐ์™€ ๊ฐ™์€ ๋กœ๋ด‡ ๋ถ€ํ’ˆ๋“ค๋กœ ์ด๋ฃจ์–ด์ง„ ์›์ƒ์•ก์„ ๋งŒ๋“ค์–ด๋ดค์Šต๋‹ˆ๋‹ค.
01:38
Put them all together, and put all this under kind of natural selection,
21
98000
4000
๊ทธ ๋ถ€ํ’ˆ๋“ค๋กœ ๋กœ๋ด‡๋“ค์ด ๋งŒ๋“ค์–ด์ง€๋ฉด ์ž์—ฐ ์„ ํƒ์„ ์ ์šฉ์‹œํ‚ค๊ณ 
01:42
under mutation, and rewarded things for how well they can move forward.
22
102000
4000
๋Œ์—ฐ๋ณ€์ด๋„ ๋งŒ๋“ค๊ณ  ์•ž์œผ๋กœ ์–ผ๋งˆ๋‚˜ ์›€์ง์ด๋Š”์ง€์— ๋”ฐ๋ผ ๋ณด์ƒ๋„ ํ–ˆ์ฃ .
01:46
A very simple task, and it's interesting to see what kind of things came out of that.
23
106000
6000
๊ต‰์žฅํžˆ ๊ฐ„๋‹จํ•œ ๊ฒƒ์ด์ง€๋งŒ ๊ฒฐ๊ณผ๋Š” ์•„์ฃผ ํฅ๋ฏธ๋กœ์› ์Šต๋‹ˆ๋‹ค.
01:52
So if you look, you can see a lot of different machines
24
112000
3000
์—ฌ๊ธฐ ๋ณด์‹œ๋ฉด ์ด ๊ณผ์ •์„ ํ†ตํ•ด ์–ผ๋งˆ๋‚˜ ๋งŽ์€ ์ข…๋ฅ˜์˜ ๊ธฐ๊ณ„๋“ค์ด
01:55
come out of this. They all move around.
25
115000
2000
ํƒ„์ƒํ–ˆ๋Š”์ง€ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ชจ๋‘ ๋‹ค
01:57
They all crawl in different ways, and you can see on the right,
26
117000
4000
์ด๋ฆฌ์ €๋ฆฌ ์›€์ง์ด๊ณ  ์žˆ์ฃ . ์˜ค๋ฅธ์ชฝ์—๋Š” ํ˜„์‹ค์—์„œ ์‹ค์ œ๋กœ
02:01
that we actually made a couple of these things,
27
121000
2000
์›€์ง์ด๋Š” ๋กœ๋ด‡๋“ค๋„ ๋ณด์ž…๋‹ˆ๋‹ค.
02:03
and they work in reality. These are not very fantastic robots,
28
123000
3000
๊ทธ๋‹ค์ง€ ๋ฉ‹์žˆ๊ฑฐ๋‚˜ ํ™˜์ƒ์ ์ด์ง„ ์•Š์ง€๋งŒ
02:06
but they evolved to do exactly what we reward them for:
29
126000
4000
์ง„ํ™”์‹œํ‚จ ๋ชฉ์ ์„ ์ถฉ๋ถ„ํžˆ ๋‹ฌ์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ฐ”๋กœ ์•ž์œผ๋กœ ์›€์ง์ด๋Š” ๊ฒƒ์ด์ฃ .
02:10
for moving forward. So that was all done in simulation,
30
130000
3000
์ด๊ฑด ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ํ˜„์‹ค ์† ์‹ค์ œ
02:13
but we can also do that on a real machine.
31
133000
2000
๊ธฐ๊ณ„์—๋„ ์ ์šฉ ๊ฐ€๋Šฅํ•œ ์ด์•ผ๊ธฐ์ž…๋‹ˆ๋‹ค.
02:15
Here's a physical robot that we actually
32
135000
5000
์—ฌ๊ธฐ ์šฐ๋ฆฌ๊ฐ€ ์‹ค์ œ๋กœ ๋งŒ๋“  ๋กœ๋ด‡์ด ์žˆ์Šต๋‹ˆ๋‹ค.
02:20
have a population of brains,
33
140000
3000
๋กœ๋ด‡์— ์กด์žฌํ•˜๋Š” ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋‘๋‡Œ๋“ค์ด ์„œ๋กœ ๊ฒฝ์Ÿํ•˜๊ณ 
02:23
competing, or evolving on the machine.
34
143000
2000
์ง„ํ™”ํ•˜๋Š” ์ค‘์ด์ฃ . ์•ฝ๊ฐ„ ๋กœ๋ฐ์˜ค์™€
02:25
It's like a rodeo show. They all get a ride on the machine,
35
145000
3000
๋น„์Šทํ•ฉ๋‹ˆ๋‹ค. ๊ฐ๊ฐ์˜ ๋‘๋‡Œ๋Š” ์ด ๋กœ๋ด‡์„ ์กฐ์ •ํ• 
02:28
and they get rewarded for how fast or how far
36
148000
3000
๊ธฐํšŒ๋ฅผ ๊ฐ–๊ฒŒ ๋˜๋Š”๋ฐ ๊ฐ€์žฅ ๋น ๋ฅธ ์‹œ๊ฐ„ ๋‚ด์— ๊ฐ€์žฅ ๋ฉ€๋ฆฌ
02:31
they can make the machine move forward.
37
151000
2000
์ด ๊ธฐ๊ณ„๋ฅผ ์•ž์œผ๋กœ ์›€์ง์ผ ๊ฒฝ์šฐ ๊ฐ€์žฅ ํฐ ๋ณด์ƒ์„ ๋ฐ›์ฃ .
02:33
And you can see these robots are not ready
38
153000
2000
์•„์ง ์„ธ๊ณ„๋ฅผ ์ œํŒจํ•  ์ˆ˜์ค€์€ ์•„๋‹ˆ์ง€๋งŒ
02:35
to take over the world yet, but
39
155000
3000
์‹œ๊ฐ„์ด ์ง€๋‚ ์ˆ˜๋ก ์ ์  ์•ž์œผ๋กœ ์›€์ง์ด๋Š”
02:38
they gradually learn how to move forward,
40
158000
2000
๋ฐฉ๋ฒ•์„ ํ„ฐ๋“ํ•˜๊ณ 
02:40
and they do this autonomously.
41
160000
3000
๊ทธ ๋ชจ๋“  ๊ณผ์ •์„ ์Šค์Šค๋กœ ํ•ด๋‚ด๋Š” ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
02:43
So in these two examples, we had basically
42
163000
4000
์ง€๊ธˆ๊นŒ์ง€๋Š” ์‹œ๋ฎฌ๋ ˆ์ด์…˜์—์„œ ํ˜น์€
02:47
machines that learned how to walk in simulation,
43
167000
3000
ํ˜„์‹ค ์„ธ๊ณ„์—์„œ ์›€์ง์ด๋Š” ๋ฐฉ๋ฒ•์„ ๋ฐฐ์šฐ๋Š”
02:50
and also machines that learned how to walk in reality.
44
170000
2000
๋กœ๋ด‡๋“ค๋งŒ ๋ณด์—ฌ๋“œ๋ ธ์Šต๋‹ˆ๋‹ค.
02:52
But I want to show you a different approach,
45
172000
2000
์ด์ œ ์•ฝ๊ฐ„ ๋‹ค๋ฅธ ์ข…๋ฅ˜์˜ ๋กœ๋ด‡์„ ์†Œ๊ฐœํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.
02:54
and this is this robot over here, which has four legs.
46
174000
6000
์—ฌ๊ธฐ ๋„ค ๊ฐœ์˜ ๋‹ค๋ฆฌ๊ฐ€ ๋‹ฌ๋ฆฐ ์ด ๋กœ๋ด‡์€ 8๊ฐœ์˜ ์ „๋™๊ธฐ๊ฐ€ ๋‹ฌ๋ ค ์žˆ์Šต๋‹ˆ๋‹ค.
03:00
It has eight motors, four on the knees and four on the hip.
47
180000
2000
๋„ค ๊ฐœ๋Š” ๋ฌด๋ฆŽ์—, ๋‹ค๋ฅธ ๋„ค ๊ฐœ๋Š” ์—‰๋ฉ์ด์ชฝ์— ์žˆ์ฃ .
03:02
It has also two tilt sensors that tell the machine
48
182000
3000
๊ธฐ๊ณ„๊ฐ€ ์–ด๋А ์ชฝ์œผ๋กœ ๊ธฐ์šธ์–ด์ ธ ์žˆ๋Š”์ง€๋ฅผ
03:05
which way it's tilting.
49
185000
3000
์•Œ๋ ค์ฃผ๋Š” ์„ผ์„œ๋„ ์žฅ์ฐฉ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
03:08
But this machine doesn't know what it looks like.
50
188000
2000
๊ทธ๋Ÿฌ๋‚˜ ์ด ๊ธฐ๊ณ„๋Š” ์ž๊ธฐ ์ž์‹ ์ด ์–ด๋–ค ๊ตฌ์กฐ์ธ์ง€
03:10
You look at it and you see it has four legs,
51
190000
2000
์ „ํ˜€ ๋ชจ๋ฆ…๋‹ˆ๋‹ค. ์—ฌ๋Ÿฌ๋ถ„์€ ์ด ๋กœ๋ด‡์„ '๋ณผ' ์ˆ˜ ์žˆ๋Š”๋ฐ ๋น„ํ•ด
03:12
the machine doesn't know if it's a snake, if it's a tree,
52
192000
2000
์ด ๊ธฐ๊ณ„๋Š” ์ž์‹ ์ด ๋ฑ€์ธ์ง€
03:14
it doesn't have any idea what it looks like,
53
194000
3000
๋‚˜๋ฌด์ธ์ง€ ์ „ํ˜€ ์•Œ์ง€ ๋ชปํ•˜์ฃ . ๋•Œ๋ฌธ์— ์ด ๊ธฐ๊ณ„๋Š” ์ž์‹ ์ด
03:17
but it's going to try to find that out.
54
197000
2000
๋ฌด์—‡์ธ์ง€ ํƒ์ง€ํ•˜๋ ค๊ณ  ํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
03:19
Initially, it does some random motion,
55
199000
2000
๊ฐ€์žฅ ๋จผ์ € ๋ฌด์ž‘์œ„๋กœ ์›€์ง์ž„์„ ํ•˜๋‚˜
03:21
and then it tries to figure out what it might look like.
56
201000
3000
์‹œ๋„ํ•ด๋ด…๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ ๋Š” ์Šค์Šค๋กœ๊ฐ€ ์–ด๋–ค ๊ตฌ์กฐ๋กœ ๋ผ ์žˆ๋Š”์ง€
03:24
And you're seeing a lot of things passing through its minds,
57
204000
2000
์•Œ์•„๋‚ด๋ ค๊ณ  ํ•˜์ฃ . ํ™”๋ฉด์„ ๋ณด์‹œ๋ฉด
03:26
a lot of self-models that try to explain the relationship
58
206000
4000
๋กœ๋ด‡์ด ์‹ค์ œ๋กœ ์›€์ง์ธ ๊ฒƒ๊ณผ ์„ผ์„œ๋กœ ๋А๋‚€ ์ •๋ณด์˜ ๊ด€๊ณ„๋ฅผ ์„ค๋ช…ํ•˜๊ธฐ ์œ„ํ•ด
03:30
between actuation and sensing. It then tries to do
59
210000
3000
ํ˜„์žฌ ์–ด๋–ค ๊ตฌ์กฐ๋“ค์„ ์ƒ๊ฐ ์ค‘์ธ์ง€ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
03:33
a second action that creates the most disagreement
60
213000
4000
๊ทธ ๋‹ค์Œ์—๋Š” ํ›„๋ณด๊ตฐ ์‚ฌ์ด์—์„œ ๊ฒฐ๊ณผ๊ฐ’์ด ์„œ๋กœ ๋‹ค๋ฅด๊ฒŒ ๋‚˜์˜ฌ ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒ๋˜๋Š” ์›€์ง์ž„์„
03:37
among predictions of these alternative models,
61
217000
2000
์‹œ๋„ํ•ด๋ด…๋‹ˆ๋‹ค. ์‹คํ—˜์‹ค ์•ˆ์˜ ๊ณผํ•™์ž์ฒ˜๋Ÿผ์š”.
03:39
like a scientist in a lab. Then it does that
62
219000
2000
๊ทธ ๊ฒฐ๊ณผ๋ฅผ ๋‚˜๋ฆ„๋Œ€๋กœ ํ•ด์„ํ•˜๋ฉด์„œ
03:41
and tries to explain that, and prune out its self-models.
63
221000
4000
ํ›„๋ณด ๊ตฌ์กฐ์˜ ์ข…๋ฅ˜๋ฅผ ์ขํ˜€ ๋‚˜๊ฐ‘๋‹ˆ๋‹ค.
03:45
This is the last cycle, and you can see it's pretty much
64
225000
3000
์ด๊ฑด ๋งˆ์ง€๋ง‰ ๋‹จ๊ณ„์ธ๋ฐ ์ด์ œ ์Šค์Šค๋กœ๊ฐ€ ์–ด๋–ค ๊ตฌ์กฐ์ธ์ง€
03:48
figured out what its self looks like. And once it has a self-model,
65
228000
4000
์ฐพ์•„๋ƒˆ๋‹ค๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด์ฒ˜๋Ÿผ ์ž๊ฐ€ ๋ชจํ˜•์„ ์™„์„ฑ์‹œํ‚ค๋ฉด
03:52
it can use that to derive a pattern of locomotion.
66
232000
4000
๊ทธ๊ฒƒ์„ ์ด์šฉํ•ด ์•ž์œผ๋กœ ์›€์ง์ผ ์ผ๋ จ์˜ ๊ณผ์ •์„ ์œ ๋„ํ•ด๋ƒ…๋‹ˆ๋‹ค.
03:56
So what you're seeing here are a couple of machines --
67
236000
2000
์—ฌ๊ธฐ ๋กœ๋ด‡์ด ์œ ๋„ํ•ด๋‚ธ
03:58
a pattern of locomotion.
68
238000
2000
์›€์ง์ž„์˜ ๊ณผ์ •์„ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
04:00
We were hoping that it wass going to have a kind of evil, spidery walk,
69
240000
4000
๊ฑฐ๋ฏธ์ฒ˜๋Ÿผ ์›€์ง์ด๊ธฐ๋ฅผ ๋ฐ”๋ž์ง€๋งŒ
04:04
but instead it created this pretty lame way of moving forward.
70
244000
4000
๋ณด์‹œ๋Š” ๊ฒƒ์ฒ˜๋Ÿผ ์šฐ์Šค๊ฝ์Šค๋Ÿฌ์šด ๋ชจ์Šต์œผ๋กœ ์›€์ง์ด๋”๊ตฐ์š”.
04:08
But when you look at that, you have to remember
71
248000
3000
ํ•˜์ง€๋งŒ ์—ฌ๊ธฐ์„œ ์ฃผ๋ชฉํ•  ๊ฒƒ์€ ์ด ๋กœ๋ด‡์ด
04:11
that this machine did not do any physical trials on how to move forward,
72
251000
6000
์•ž์œผ๋กœ ์›€์ง์ด๊ธฐ ์œ„ํ•ด ์•„๋ฌด๊ฑฐ๋‚˜ ๋งˆ๊ตฌ ์‹œ๋„ํ•œ ๊ฒƒ์ด ์•„๋‹ˆ๊ณ 
04:17
nor did it have a model of itself.
73
257000
2000
์ฒ˜์Œ์—” ์ž์‹ ์— ๋Œ€ํ•œ ์ธ์‹์กฐ์ฐจ ์—†์—ˆ๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค.
04:19
It kind of figured out what it looks like, and how to move forward,
74
259000
3000
์ฆ‰ ์•„๋ฌด๊ฒƒ๋„ ์—†๋Š” ์ƒํƒœ์—์„œ ์Šค์Šค๋กœ๊ฐ€ ์–ด๋–ค ๊ตฌ์กฐ๋กœ ๋ผ ์žˆ๋Š”์ง€
04:22
and then actually tried that out.
75
262000
4000
์•Œ์•„๋‚ด๊ณ  ์•ž์œผ๋กœ ์›€์ง์ด๋Š” ๋ฒ•์„ ํ„ฐ๋“ํ•œ ํ›„ ์‹ค์ œ๋กœ ์•ž์œผ๋กœ ์›€์ง์˜€๋‹ค๋Š” ๊ฒƒ์ด์ฃ .
04:26
(Applause)
76
266000
5000
(๋ฐ•์ˆ˜)
04:31
So, we'll move forward to a different idea.
77
271000
4000
์ด์ œ ๋˜ ๋‹ค๋ฅธ ์•„์ด๋””์–ด๋กœ ๋„˜์–ด๊ฐ‘์‹œ๋‹ค.
04:35
So that was what happened when we had a couple of --
78
275000
5000
๋ฐฉ๊ธˆ ์ „๊นŒ์ง€ ๋ณด์—ฌ๋“œ๋ฆฐ ๊ฒƒ์€--
04:40
that's what happened when you had a couple of -- OK, OK, OK --
79
280000
4000
๋ฐฉ๊ธˆ ์ „๊นŒ์ง€ ๋ณด์—ฌ๋“œ๋ฆฐ ๊ฒƒ์€-- ์•Œ์•˜์–ด. ์•Œ์•˜์–ด. ๊ทธ๋งŒํ•ด.
04:44
(Laughter)
80
284000
2000
(์›ƒ์Œ)
04:46
-- they don't like each other. So
81
286000
2000
์„œ๋กœ ์‹ซ์–ดํ•˜๋Š” ๊ฒƒ ๊ฐ™๊ตฐ์š”.
04:48
there's a different robot.
82
288000
3000
์—ฌ๊ธฐ ๋˜ ๋‹ค๋ฅธ ๋กœ๋ด‡์ด ์žˆ์Šต๋‹ˆ๋‹ค.
04:51
That's what happened when the robots actually
83
291000
2000
๋ฐฉ๊ธˆ ์ „๊นŒ์ง€ ๋ณด์—ฌ๋“œ๋ฆฐ ๊ฒƒ์€ ๊ฒฐ๊ณผ๊ฐ€ ์ข‹์„ ๋•Œ
04:53
are rewarded for doing something.
84
293000
2000
๋ณด์ƒ์„ ํ•œ ๊ฒฝ์šฐ์˜€์Šต๋‹ˆ๋‹ค.
04:55
What happens if you don't reward them for anything, you just throw them in?
85
295000
3000
๊ทธ๋Ÿฐ๋ฐ ๋ณด์ƒ์„ ํ•˜์ง€ ์•Š์œผ๋ฉด ์–ด๋–ป๊ฒŒ ๋ ๊นŒ์š”?
04:58
So we have these cubes, like the diagram showed here.
86
298000
3000
์—ฌ๊ธฐ ๊ทธ๋ฆผ๊ณผ ๊ฐ™์€ ์ •์œก๋ฉด์ฒด๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
05:01
The cube can swivel, or flip on its side,
87
301000
2000
์ด ์ •์œก๋ฉด์ฒด๋Š” ํšŒ์ „ํ•˜๊ฑฐ๋‚˜ ์˜†์œผ๋กœ ๋ˆ„์šธ์ˆ˜๋„ ์žˆ์ฃ .
05:04
and we just throw 1,000 of these cubes into a soup --
88
304000
4000
์ด ์ •์œก๋ฉด์ฒด 1,000๊ฐœ๋ฅผ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์—์„œ ๋งŒ๋“  ๋‹ค์Œ
05:08
this is in simulation --and don't reward them for anything,
89
308000
2000
์–ด๋– ํ•œ ๋ณด์ƒ๋„ ํ•˜์ง€ ์•Š์€ ์ƒํƒœ์—์„œ
05:10
we just let them flip. We pump energy into this
90
310000
3000
๊ณ„์†ํ•ด์„œ ์˜†์œผ๋กœ ๋ˆ•๋„๋ก ํ–ˆ์Šต๋‹ˆ๋‹ค.
05:13
and see what happens in a couple of mutations.
91
313000
3000
๋Œ์—ฐ๋ณ€์ด ๋ช‡ ๊ฐœ๊ฐ€ ๋‚˜ํƒ€๋‚œ ํ›„ ์–ด๋–ป๊ฒŒ ๋˜๋Š”์ง€ ๋ณด์ฃ .
05:16
So, initially nothing happens, they're just flipping around there.
92
316000
3000
์ฒ˜์Œ์—” ์•„๋ฌด๊ฒƒ๋„ ์ผ์–ด๋‚˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
05:19
But after a very short while, you can see these blue things
93
319000
4000
ํ•˜์ง€๋งŒ ์‹œ๊ฐ„์ด ์ข€ ํ๋ฅด์ž ์˜ค๋ฅธ์ชฝ ํŒŒ๋ž€์ƒ‰์˜ ๊ฐœ์ฒด์ˆ˜๊ฐ€
05:23
on the right there begin to take over.
94
323000
2000
๋Š˜์–ด๋‚˜๋Š” ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
05:25
They begin to self-replicate. So in absence of any reward,
95
325000
4000
์ž๊ฐ€ ๋ณต์ œ๋ฅผ ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์™ธ๋ถ€ ๋ณด์ƒ์ด ์—†์„ ๊ฒฝ์šฐ
05:29
the intrinsic reward is self-replication.
96
329000
3000
์ž๊ฐ€ ๋ณต์ œ๋ฅผ ํ†ตํ•ด ์Šค์Šค๋กœ ๋ณด์ƒํ•˜๋Š” ๊ฒƒ์ด์ฃ .
05:32
And we've actually built a couple of these,
97
332000
1000
๋น„์Šทํ•œ ๊ฒƒ์„ ์‹ค์ œ๋กœ ๋งŒ๋“ค์–ด๋ดค์Šต๋‹ˆ๋‹ค.
05:33
and this is part of a larger robot made out of these cubes.
98
333000
4000
์ด๊ฑด ๋” ํฐ ๋กœ๋ด‡์˜ ์ผ๋ถ€๋ถ„์ผ ๋ฟ์ž…๋‹ˆ๋‹ค.
05:37
It's an accelerated view, where you can see the robot actually
99
337000
3000
๋น ๋ฅธ ์†๋„๋กœ ์žฌ์ƒ์‹œํ‚ค๋ฉด ๋กœ๋ด‡์ด ์ž๊ฐ€ ๋ณต์ œํ•˜๋Š”
05:40
carrying out some of its replication process.
100
340000
2000
๋ชจ์Šต์„ ๋ณผ ์ˆ˜ ์žˆ์ฃ .
05:42
So you're feeding it with more material -- cubes in this case --
101
342000
4000
์Šค์Šค๋กœ๋ฅผ ๋ณต์ œ์‹œํ‚ฌ ์›๋ฃŒ์™€ ์—๋„ˆ์ง€๋งŒ ๊ณต๊ธ‰ํ•ด์ฃผ๋ฉด
05:46
and more energy, and it can make another robot.
102
346000
3000
๋˜ ๋‹ค๋ฅธ ๋กœ๋ด‡์„ ์ €์ ˆ๋กœ ํƒ„์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
05:49
So of course, this is a very crude machine,
103
349000
3000
๋ฌผ๋ก  ์ด๊ฑด ์•„์ฃผ ๋‹จ์ˆœํ•œ ๊ธฐ๊ณ„์— ํ•ด๋‹น๋˜์ง€๋งŒ ํ˜„์žฌ ์šฐ๋ฆฌ๋Š”
05:52
but we're working on a micro-scale version of these,
104
352000
2000
์ด ๊ธฐ๊ณ„๋ฅผ ์ดˆ์†Œํ˜•์œผ๋กœ ์ถ•์†Œ์‹œํ‚ค๋Š” ์ž‘์—…์„ ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
05:54
and hopefully the cubes will be like a powder that you pour in.
105
354000
3000
์ž˜ ๋œ๋‹ค๋ฉด ์ด ์ •์œก๋ฉด์ฒด๋“ค์€ ๋ถ„๋ง ๊ฐ€๋ฃจ์™€ ๊ฐ™์€ ๋ชจ์Šต์„ ์ง€๋‹ˆ๊ฒŒ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
05:57
OK, so what can we learn? These robots are of course
106
357000
5000
๊ทธ๋Ÿผ ์—ฌ๊ธฐ์„œ ๋ฌด์—‡์„ ๋ฐฐ์šธ ์ˆ˜ ์žˆ์„๊นŒ์š”? ์ง€๊ธˆ๊นŒ์ง€ ์†Œ๊ฐœํ•ด๋“œ๋ฆฐ ๋กœ๋ด‡๋“ค์€
06:02
not very useful in themselves, but they might teach us something
107
362000
3000
๊ทธ ์ž์ฒด๋กœ๋Š” ๋ณ„ ์“ธ๋ชจ๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์–ด๋–ป๊ฒŒ ํ•ด์•ผ
06:05
about how we can build better robots,
108
365000
3000
๋” ๋‚˜์€ ๋กœ๋ด‡์„ ๋งŒ๋“ค ์ˆ˜ ์žˆ๋Š”์ง€
06:08
and perhaps how humans, animals, create self-models and learn.
109
368000
5000
๋˜ ์ธ๊ฐ„๊ณผ ๋‹ค๋ฅธ ๋™๋ฌผ๋“ค์ด ์–ด๋–ป๊ฒŒ ์ž์‹ ์— ๋Œ€ํ•ด ์ธ์‹ํ•˜๊ณ  ๋ฐฐ์šฐ๋Š”์ง€์— ๋Œ€ํ•œ ๋‹ต์„ ์ œ์‹œํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค.
06:13
And one of the things that I think is important
110
373000
2000
๋งˆ์ง€๋ง‰์œผ๋กœ ์ œ๊ฐ€ ์ค‘์š”ํ•˜๊ฒŒ ์ƒ๊ฐํ•˜๋Š” ๊ฒƒ ์ค‘ ํ•˜๋‚˜๋Š”
06:15
is that we have to get away from this idea
111
375000
2000
์ธ๊ฐ„์ด ์ˆ˜๋™์ ์œผ๋กœ ๊ธฐ๊ณ„๋ฅผ ์„ค๊ณ„ํ•˜๋Š” ๊ฒƒ์„ ํƒˆํ”ผํ•ด์•ผ
06:17
of designing the machines manually,
112
377000
2000
ํ•œ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋งˆ์น˜ ์•„์ด์ฒ˜๋Ÿผ ๊ธฐ๊ณ„ ์Šค์Šค๋กœ๊ฐ€ ์ง„ํ™”ํ•˜๊ณ 
06:19
but actually let them evolve and learn, like children,
113
379000
3000
๋ฐฐ์šธ ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋ฉด ์•„๋งˆ ๊ทธ๋Ÿฐ ๋‚ ์ด ๋จธ์ง€์•Š์•„ ์˜ฌ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
06:22
and perhaps that's the way we'll get there. Thank you.
114
382000
2000
๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
06:24
(Applause)
115
384000
2000
(๋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