What intelligent machines can learn from a school of fish | Radhika Nagpal

109,666 views ・ 2017-10-06

TED


Please double-click on the English subtitles below to play the video.

00:12
In my early days as a graduate student,
0
12575
2015
00:14
I went on a snorkeling trip off the coast of the Bahamas.
1
14614
3555
00:18
I'd actually never swum in the ocean before,
2
18609
2949
00:21
so it was a bit terrifying.
3
21582
1844
00:23
What I remember the most is, as I put my head in the water
4
23836
3000
00:26
and I was trying really hard to breathe through the snorkel,
5
26860
4250
00:31
this huge group of striped yellow and black fish
6
31134
5366
00:36
came straight at me ...
7
36524
1483
00:38
and I just froze.
8
38637
1397
00:40
And then, as if it had suddenly changed its mind,
9
40795
3613
00:44
came towards me and then swerved to the right
10
44432
2437
00:46
and went right around me.
11
46893
1515
00:48
It was absolutely mesmerizing.
12
48740
1526
00:50
Maybe many of you have had this experience.
13
50290
2182
00:53
Of course, there's the color and the beauty of it,
14
53059
3422
00:56
but there was also just the sheer oneness of it,
15
56505
2928
00:59
as if it wasn't hundreds of fish
16
59457
2343
01:01
but a single entity with a single collective mind
17
61824
3135
01:04
that was making decisions.
18
64983
1507
01:07
When I look back, I think that experience really ended up determining
19
67486
3682
01:11
what I've worked on for most of my career.
20
71192
2222
01:15
I'm a computer scientist,
21
75977
1280
01:17
and the field that I work in is artificial intelligence.
22
77281
2747
01:20
And a key theme in AI
23
80459
1517
01:22
is being able to understand intelligence by creating our own computational systems
24
82000
4443
01:26
that display intelligence the way we see it in nature.
25
86467
3253
01:30
Now, most popular views of AI, of course, come from science fiction and the movies,
26
90287
4438
01:34
and I'm personally a big Star Wars fan.
27
94749
2577
01:38
But that tends to be a very human-centric view of intelligence.
28
98321
3889
01:42
When you think of a fish school,
29
102964
2207
01:45
or when I think of a flock of starlings,
30
105195
2953
01:48
that feels like a really different kind of intelligence.
31
108172
3340
01:52
For starters, any one fish is just so tiny
32
112765
3913
01:56
compared to the sheer size of the collective,
33
116702
2887
01:59
so it seems that any one individual
34
119613
3110
02:02
would have a really limited and myopic view of what's going on,
35
122747
2993
02:05
and intelligence isn't really about the individual
36
125764
2334
02:08
but somehow a property of the group itself.
37
128122
2677
02:11
Secondly, and the thing that I still find most remarkable,
38
131938
3231
02:15
is that we know that there are no leaders supervising this fish school.
39
135193
5032
02:20
Instead, this incredible collective mind behavior
40
140983
3501
02:24
is emerging purely from the interactions of one fish and another.
41
144508
4532
02:29
Somehow, there are these interactions or rules of engagement
42
149064
3968
02:33
between neighboring fish
43
153056
1755
02:34
that make it all work out.
44
154835
1467
02:37
So the question for AI then becomes,
45
157556
2651
02:40
what are those rules of engagement that lead to this kind of intelligence,
46
160231
4158
02:44
and of course, can we create our own?
47
164413
1907
02:46
And that's the primary thing that I work on with my team in my lab.
48
166819
3587
02:50
We work on it through theory,
49
170763
1637
02:52
looking at abstract rule systems
50
172424
2348
02:54
and thinking about the mathematics behind it.
51
174796
2349
02:57
We also do it through biology, working closely with experimentalists.
52
177717
4285
03:02
But mostly, we do it through robotics,
53
182399
1953
03:04
where we try to create our own collective systems
54
184376
3904
03:08
that can do the kinds of things that we see in nature,
55
188304
2707
03:11
or at least try to.
56
191035
1237
03:13
One of our first robotic quests along this line
57
193727
2804
03:16
was to create our very own colony of a thousand robots.
58
196555
4045
03:20
So very simple robots,
59
200960
1334
03:22
but they could be programmed to exhibit collective intelligence,
60
202318
3603
03:25
and that's what we were able to do.
61
205945
1729
03:28
So this is what a single robot looks like.
62
208014
2032
03:30
It's quite small, about the size of a quarter,
63
210070
2523
03:32
and you can program how it moves,
64
212617
2310
03:34
but it can also wirelessly communicate with other robots,
65
214951
3416
03:38
and it can measure distances from them.
66
218391
2167
03:40
And so now we can start to program exactly an interaction,
67
220582
3476
03:44
a rule of engagement between neighbors.
68
224082
2094
03:46
And once we have this system,
69
226533
1894
03:48
we can start to program many different kinds of rules of engagement
70
228451
3416
03:51
that you would see in nature.
71
231891
1506
03:53
So for example, spontaneous synchronization,
72
233421
2976
03:56
how audiences are clapping and suddenly start all clapping together,
73
236421
5238
04:01
the fireflies flashing together.
74
241683
2068
04:06
We can program rules for pattern formation,
75
246739
2691
04:09
how cells in a tissue
76
249454
1786
04:11
determine what role they're going to take on
77
251264
2102
04:13
and set the patterns of our bodies.
78
253390
1706
04:16
We can program rules for migration,
79
256865
2089
04:18
and in this way, we're really learning from nature's rules.
80
258978
2977
04:22
But we can also take it a step further.
81
262415
2647
04:25
We can actually take these rules that we've learned from nature
82
265086
2992
04:28
and combine them and create entirely new collective behaviors
83
268102
3794
04:31
of our very own.
84
271920
1198
04:33
So for example,
85
273780
1478
04:35
imagine that you had two different kinds of rules.
86
275282
2352
04:38
So your first rule is a motion rule
87
278194
2119
04:40
where a moving robot can move around other stationary robots.
88
280337
4341
04:44
And your second rule is a pattern rule
89
284702
1811
04:46
where a robot takes on a color based on its two nearest neighbors.
90
286537
3157
04:50
So if I start with a blob of robots in a little pattern seed,
91
290499
3445
04:53
it turns out that these two rules are sufficient for the group
92
293968
2906
04:56
to be able to self-assemble a simple line pattern.
93
296898
2752
05:00
And if I have more complicated pattern rules,
94
300934
2544
05:03
and I design error correction rules,
95
303502
2317
05:05
we can actually create really, really complicated self assemblies,
96
305843
3097
05:08
and here's what that looks like.
97
308964
1644
05:11
So here, you're going to see a thousand robots
98
311694
2985
05:14
that are working together to self-assemble the letter K.
99
314703
3462
05:18
The K is on its side.
100
318189
1306
05:20
And the important thing is that no one is in charge.
101
320043
2731
05:22
So any single robot is only talking to a small number of robots nearby it,
102
322798
4825
05:27
and it's using its motion rule to move around the half-built structure
103
327647
3937
05:31
just looking for a place to fit in based on its pattern rules.
104
331608
3007
05:35
And even though no robot is doing anything perfectly,
105
335614
4398
05:40
the rules are such that we can get the collective to do its goal
106
340036
3660
05:43
robustly together.
107
343720
1473
05:45
And the illusion becomes almost so perfect, you know --
108
345853
2982
05:48
you just start to not even notice that they're individual robots at all,
109
348859
3416
05:52
and it becomes a single entity,
110
352299
1683
05:54
kind of like the school of fish.
111
354006
1721
05:59
So these are robots and rules in two dimensions,
112
359833
2739
06:02
but we can also think about robots and rules in three dimensions.
113
362596
3311
06:05
So what if we could create robots that could build together?
114
365931
3603
06:10
And here, we can take inspiration from social insects.
115
370396
3255
06:14
So if you think about mound-building termites
116
374009
2660
06:16
or you think about army ants,
117
376693
2052
06:18
they create incredible, complex nest structures out of mud
118
378769
4253
06:23
and even out of their own bodies.
119
383046
2144
06:26
And like the system I showed you before,
120
386422
2220
06:28
these insects actually also have pattern rules
121
388666
2970
06:31
that help them determine what to build,
122
391660
2038
06:33
but the pattern can be made out of other insects,
123
393722
2302
06:36
or it could be made out of mud.
124
396048
1787
06:38
And we can use that same idea to create rules for robots.
125
398998
4361
06:44
So here, you're going to see some simulated robots.
126
404041
3161
06:47
So the simulated robot has a motion rule,
127
407226
2483
06:49
which is how it traverses through the structure,
128
409733
2333
06:52
looking for a place to fit in,
129
412090
1997
06:54
and it has pattern rules where it looks at groups of blocks
130
414111
3000
06:57
to decide whether to place a block.
131
417135
2205
07:00
And with the right motion rules and the right pattern rules,
132
420464
3063
07:03
we can actually get the robots to build whatever we want.
133
423551
3635
07:08
And of course, everybody wants their own tower.
134
428017
2691
07:11
(Laughter)
135
431170
1982
07:13
So once we have these rules,
136
433820
1684
07:15
we can start to create the robot bodies that go with these rules.
137
435528
3166
07:18
So here, you see a robot that can climb over blocks,
138
438718
3309
07:22
but it can also lift and move these blocks
139
442051
2681
07:24
and it can start to edit the very structure that it's on.
140
444756
2697
07:28
But with these rules,
141
448437
1148
07:29
this is really only one kind of robot body that you could imagine.
142
449609
3479
07:33
You could imagine many different kinds of robot bodies.
143
453112
2579
07:35
So if you think about robots that maybe could move sandbags
144
455715
4610
07:40
and could help build levees,
145
460349
2549
07:42
or we could think of robots that built out of soft materials
146
462922
4301
07:47
and worked together to shore up a collapsed building --
147
467247
3644
07:50
so just the same kind of rules in different kinds of bodies.
148
470915
2998
07:56
Or if, like my group, you are completely obsessed with army ants,
149
476030
4223
08:00
then maybe one day we can make robots that can climb over literally anything
150
480277
4374
08:04
including other members of their tribe,
151
484675
2174
08:06
and self-assemble things out of their own bodies.
152
486873
2349
08:09
Once you understand the rules,
153
489957
1681
08:11
just many different kinds of robot visions become possible.
154
491662
3379
08:18
And coming back to the snorkeling trip,
155
498612
2234
08:20
we actually understand a great deal about the rules that fish schools use.
156
500870
5345
08:26
So if we can invent the bodies to go with that,
157
506589
2836
08:29
then maybe there is a future
158
509449
1428
08:30
where I and my group will get to snorkel with a fish school of our own creation.
159
510901
4522
08:40
Each of these systems that I showed you
160
520670
2129
08:42
brings us closer to having the mathematical and the conceptual tools
161
522823
4277
08:47
to create our own versions of collective power,
162
527124
3381
08:50
and this can enable many different kinds of future applications,
163
530529
3001
08:53
whether you think about robots that build flood barriers
164
533554
3164
08:56
or you think about robotic bee colonies that could pollinate crops
165
536742
4297
09:01
or underwater schools of robots that monitor coral reefs,
166
541063
3524
09:04
or if we reach for the stars and we thinking about programming
167
544611
3103
09:07
constellations of satellites.
168
547738
1619
09:09
In each of these systems,
169
549968
1612
09:11
being able to understand how to design the rules of engagement
170
551604
3547
09:15
and being able to create good collective behavior
171
555175
2514
09:17
becomes a key to realizing these visions.
172
557713
2374
09:22
So, so far I've talked about rules for insects and for fish
173
562562
4107
09:26
and for robots,
174
566693
2369
09:29
but what about the rules that apply to our own human collective?
175
569086
3103
09:32
And the last thought that I'd like to leave you with
176
572686
2430
09:35
is that science is of course itself
177
575140
1681
09:36
an incredible manifestation of collective intelligence,
178
576845
3484
09:40
but unlike the beautiful fish schools that I study,
179
580353
3318
09:43
I feel we still have a much longer evolutionary path to walk.
180
583695
3943
09:48
So in addition to working on improving the science of robot collectives,
181
588566
4604
09:53
I also work on creating robots and thinking about rules
182
593194
3277
09:56
that will improve our own scientific collective.
183
596495
2460
10:00
There's this saying that I love:
184
600018
1668
10:01
who does science determines what science gets done.
185
601710
3404
10:06
Imagine a society
186
606059
2941
10:09
where we had rules of engagement
187
609024
1651
10:10
where every child grew up believing that they could stand here
188
610699
3303
10:14
and be a technologist of the future,
189
614026
2422
10:16
or where every adult
190
616472
1501
10:17
believed that they had the ability not just to understand but to change
191
617997
4119
10:22
how science and technology impacts their everyday lives.
192
622140
3555
10:26
What would that society look like?
193
626640
1899
10:30
I believe that we can do that.
194
630206
1508
10:31
I believe that we can choose our rules,
195
631738
2291
10:34
and we engineer not just robots
196
634053
1757
10:35
but we can engineer our own human collective,
197
635834
2596
10:38
and if we do and when we do, it will be beautiful.
198
638454
3834
10:42
Thank you.
199
642312
1151
10:43
(Applause)
200
643487
6547
About this website

This site will introduce you to YouTube videos that are useful for learning English. You will see English lessons taught by top-notch teachers from around the world. Double-click on the English subtitles displayed on each video page to play the video from there. The subtitles scroll in sync with the video playback. If you have any comments or requests, please contact us using this contact form.

https://forms.gle/WvT1wiN1qDtmnspy7