Are Insect Brains the Secret to Great AI? | Frances S. Chance | TED

72,438 views ใƒป 2023-01-02

TED


ืื ื ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ืœืžื˜ื” ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ.

ืชืจื’ื•ื: zeeva livshitz ืขืจื™ื›ื”: Ido Dekkers
00:05
Creating intelligence on a computer.
0
5210
2377
ื™ืฆื™ืจืช ืื™ื ื˜ืœื™ื’ื ืฆื™ื” ื‘ืžื—ืฉื‘.
00:08
This has been the Holy Grail for artificial intelligence
1
8129
2837
ื”ื™ืชื” ื”ื’ื‘ื™ืข ื”ืงื“ื•ืฉ ืœื‘ื™ื ื” ืžืœืื›ื•ืชื™ืช
00:11
for quite some time.
2
11007
1377
ืœืžืฉืš ื–ืžืŸ ืžื”.
00:12
But how do we get there?
3
12968
1668
ืื‘ืœ ืื™ืš ื ื’ื™ืข ืœืฉื?
00:15
So we view ourselves as highly intelligent beings.
4
15679
3128
ืื ื—ื ื• ืจื•ืื™ื ืืช ืขืฆืžื ื• ื›ื™ืฆื•ืจื™ื ืชื‘ื•ื ื™ื™ื ืžืื•ื“
00:18
So it's logical to study our own brains,
5
18848
2795
ืื– ื–ื” ื”ื’ื™ื•ื ื™ ืœื—ืงื•ืจ ืืช ื”ืžื•ื— ืฉืœื ื•,
00:21
the substrate of our cognition, for creating artificial intelligence.
6
21685
4212
ืชืฉืชื™ืช ื”ื”ื›ืจื” ืฉืœื ื•, ืœื™ืฆื™ืจืช ื‘ื™ื ื” ืžืœืื›ื•ืชื™ืช.
00:27
Imagine if we could replicate how our own brains work on a computer.
7
27148
4130
ืชืืจื• ืœืขืฆืžื›ื ืฉื”ื™ื™ื ื• ื™ื›ื•ืœื™ื ืœืฉื›ืคืœ ืื™ืš ื”ืžื•ื— ืฉืœื ื• ืขื•ื‘ื“ ื‘ืžื—ืฉื‘.
00:32
But now consider the journey that would be required.
8
32612
2503
ืื‘ืœ ืขื›ืฉื™ื• ืชื—ืฉื‘ื• ืขืœ ื”ืžืกืข ืฉื™ื™ื“ืจืฉ.
00:37
The human brain contains 86 billion neurons.
9
37367
3712
ื”ืžื•ื— ื”ืื ื•ืฉื™ ืžื›ื™ืœ 86 ืžื™ืœื™ืืจื“ ื ื•ื™ืจื•ื ื™ื.
00:42
Each is constantly communicating with thousands of others,
10
42289
3628
ื›ืœ ืื—ื“ ืžืชืงืฉืจ ื›ืœ ื”ื–ืžืŸ ืขื ืืœืคื™ ืื—ืจื™ื,
00:45
and each has individual characteristics of its own.
11
45959
3045
ื•ืœื›ืœ ืื—ื“ ื™ืฉ ืžืืคื™ื™ื ื™ื ืื™ื ื“ื™ื‘ื™ื“ื•ืืœื™ื™ื ืžืฉืœื•.
00:49
Capturing the human brain on a computer
12
49879
2378
ืœื›ื™ื“ืช ื”ืžื•ื— ื”ืื ื•ืฉื™ ื‘ืžื—ืฉื‘
00:52
may simply be too big and too complex a problem
13
52299
3670
ื™ื›ื•ืœื” ืœื”ื™ื•ืช ืคืฉื•ื˜ ื‘ืขื™ื” ื’ื“ื•ืœื” ื•ืžื•ืจื›ื‘ืช ืžื“ื™
00:56
to tackle with the technology and the knowledge that we have today.
14
56011
3795
ื›ื“ื™ ืœื”ืชืžื•ื“ื“ ืขื ื”ื˜ื›ื ื•ืœื•ื’ื™ื” ื•ื”ื™ื“ืข ืฉื™ืฉ ืœื ื• ื”ื™ื•ื.
01:01
I believe that we can capture a brain on a computer,
15
61266
2878
ืื ื™ ืžืืžื™ื ื” ืฉืื ื—ื ื• ื™ื›ื•ืœื™ื ืœืœื›ื•ื“ ืžื•ื— ื‘ืžื—ืฉื‘,
01:04
but we have to start smaller.
16
64144
2461
ืื‘ืœ ืื ื—ื ื• ืฆืจื™ื›ื™ื ืœื”ืชื—ื™ืœ ื‘ืงื˜ืŸ ื™ื•ืชืจ.
01:07
Much smaller.
17
67230
1168
ืงื˜ืŸ ื‘ื”ืจื‘ื”.
01:10
These insects have three of the most fascinating brains in the world to me.
18
70734
4713
ื‘ืขื™ื ื™, ืœื—ืจืงื™ื ื”ืืœื” ื™ืฉ ืฉืœื•ืฉื” ืžื”ืžื•ื—ื•ืช ื”ื›ื™ ืžืจืชืงื™ื ื‘ืขื•ืœื.
01:16
While they do not possess human-level intelligence,
19
76448
2836
ื‘ืขื•ื“ ืฉืื™ืŸ ืœื”ื ืื™ื ื˜ืœื™ื’ื ืฆื™ื” ื‘ืจืžื” ื”ืื ื•ืฉื™ืช,
01:19
each is remarkable at a particular task.
20
79326
3169
ื›ืœ ืื—ื“ ืžื”ื ืžื“ื”ื™ื ื‘ืžืฉื™ืžื” ืžืกื•ื™ืžืช.
01:22
Think of them as highly trained specialists.
21
82954
2544
ื—ื™ืฉื‘ื• ืขืœื™ื”ื ื›ืขืœ ืžื•ืžื—ื™ื ืžื™ื•ืžื ื™ื.
01:26
African dung beetles are really good at rolling large balls in straight lines.
22
86750
5088
ื—ื™ืคื•ืฉื™ื•ืช ื–ื‘ืœ ืืคืจื™ืงืื™ื•ืช ืžืžืฉ ื˜ื•ื‘ื•ืช ื‘ื’ืœื’ื•ืœ ื›ื“ื•ืจื™ื ื’ื“ื•ืœื™ื ื‘ืงื•ื•ื™ื ื™ืฉืจื™ื.
01:31
(Laughter)
23
91880
1710
(ืฆื—ื•ืง)
01:33
Now, if you've ever made a snowman,
24
93632
1710
ืื ืื™ ืคืขื ื™ืฆืจืชื ืื™ืฉ ืฉืœื’,
01:35
you know that rolling a large ball is not easy.
25
95342
2460
ืืชื ื™ื•ื“ืขื™ื, ืฉืœื’ืœื’ืœ ื›ื“ื•ืจ ื’ื“ื•ืœ ื–ื” ืœื ืงืœ.
01:39
Now picture trying to make that snowman
26
99095
2211
ืขื›ืฉื™ื• ื“ืžื™ื™ื ื• ื ืกื™ื•ืŸ ืœื™ืฆื•ืจ ืืช ืื™ืฉ ื”ืฉืœื’ ื”ื–ื”
01:41
when the ball of snow is as big as you are
27
101348
2377
ื›ืืฉืจ ื›ื“ื•ืจ ื”ืฉืœื’ ื’ื“ื•ืœ ื›ืžื•ื›ื
01:43
and you're standing on your head.
28
103725
1835
ื•ืืชื ืขื•ืžื“ื™ื ืขืœ ื”ืจืืฉ.
01:45
(Laughter)
29
105602
1168
(ืฆื—ื•ืง)
01:47
Sahara desert ants are navigation specialists.
30
107228
3796
ื ืžืœื™ื ืžืžื“ื‘ืจ ืกื”ืจื” ื”ืŸ ืžื•ืžื—ื™ื•ืช ื ื™ื•ื•ื˜.
01:51
They might have to wander a considerable distance to forage for food.
31
111775
3753
ื™ื™ืชื›ืŸ ืฉื”ื ื™ืฆื˜ืจื›ื• ืœืฉื•ื˜ื˜ ืœืžืจื—ืง ื’ื“ื•ืœ ืœื—ื™ืคื•ืฉ ืžื–ื•ืŸ.
01:55
But once they do find sustenance,
32
115862
1752
ืื‘ืœ ื‘ืจื’ืข ืฉื”ื ื›ืŸ ืžื•ืฆืื™ื ืžื–ื•ืŸ,
01:57
they know how to calculate the straightest path home.
33
117656
2585
ื”ืŸ ื™ื•ื“ืขื•ืช ืœื—ืฉื‘ ืืช ื”ืฉื‘ื™ืœ ื”ื›ื™ ื™ืฉืจ ื”ื‘ื™ืชื”.
02:01
And the dragonfly is a hunting specialist.
34
121910
2919
ื•ื”ืฉืคื™ืจื™ืช ื”ื™ื ืžื•ืžื—ื™ืช ืฆื™ื“.
02:05
In the wild, dragonflies capture approximately 95 percent
35
125205
3587
ื‘ื˜ื‘ืข, ืฉืคื™ืจื™ื•ืช ืœื•ื›ื“ื•ืช ื‘ืขืจืš 95 ืื—ื•ื–
02:08
of the prey they choose to go after.
36
128833
1752
ืฉืœ ื”ื˜ืจืฃ ืฉื”ืŸ ื‘ื•ื—ืจื•ืช ืœืœื›ืช ืื—ืจื™ื•.
02:11
These insects are so good at their specialties
37
131336
3003
ื”ื—ืจืงื™ื ื”ืืœื” ื›ืœ ื›ืš ื˜ื•ื‘ื™ื ื‘ื”ืชืžื—ื•ื™ื•ืช ืฉืœื”ื
02:14
that neuroscientists such as myself study them as model systems
38
134381
4337
ืฉืžื“ืขื ื™ ืžื•ื— ื›ืžื•ื ื™ ื—ื•ืงืจื™ื ืื•ืชื ื›ืžืขืจื›ื•ืช ืžื•ื“ืœ
02:18
to understand how animal nervous systems solve particular problems.
39
138760
4046
ืœื”ื‘ื™ืŸ ืื™ืš ืžืขืจื›ื•ืช ื”ืขืฆื‘ื™ื ืฉืœ ื‘ืขืœื™ ื—ื™ื™ื ืคื•ืชืจื•ืช ื‘ืขื™ื•ืช ืžืกื•ื™ืžื•ืช.
02:23
And in my own research, I study brains to bring these solutions,
40
143973
3879
ื•ื‘ืžื—ืงืจ ืฉืœื™, ืื ื™ ื—ื•ืงืจืช ืžื•ื—ื•ืช ื›ื“ื™ ืœื”ื‘ื™ื ืืช ื”ืคืชืจื•ื ื•ืช ื”ืืœื”,
02:27
the best that biology has to offer, to computers.
41
147894
3337
ื”ื›ื™ ื˜ื•ื‘ื™ื ืฉืœื‘ื™ื•ืœื•ื’ื™ื” ื™ืฉ ืœื”ืฆื™ืข, ืœืžื—ืฉื‘ื™ื.
02:31
So consider the dragonfly brain.
42
151272
1877
ืื– ืงื—ื• ื‘ื—ืฉื‘ื•ืŸ ืืช ืžื•ื— ื”ืฉืคื™ืจื™ืช.
02:33
It has only on the order of one million neurons.
43
153900
3003
ื™ืฉ ืœื” ืจืง ืกื“ืจ ื’ื•ื“ืœ ืฉืœ ืžื™ืœื™ื•ืŸ ื ื•ื™ืจื•ื ื™ื.
02:37
Now, it's still not easy to unravel a circuit of even one million neurons.
44
157570
4672
ื–ื” ืขื“ื™ื™ืŸ ืœื ืงืœ ืœืคืขื ื— ืžืขื’ืœ ืฉืœ ืืคื™ืœื• ืžื™ืœื™ื•ืŸ ื ื•ื™ืจื•ื ื™ื.
02:42
But given the choice
45
162701
1167
ืื‘ืœ ื‘ื”ื™ื ืชืŸ ื”ื‘ื—ื™ืจื”
02:43
between trying to tease apart the one-million-neuron brain
46
163910
3003
ื‘ื™ืŸ ื”ื ื™ืกื™ื•ืŸ ืœื”ืคืจื™ื“ ื‘ื™ืŸ ื”ืžื•ื— ื‘ืŸ ืžื™ืœื™ื•ืŸ ื”ื ื•ื™ืจื•ื ื™ื
02:46
versus the 86-billion-neuron brain,
47
166955
2586
ืœืขื•ืžืช ื”ืžื•ื— ืฉืœ 86 ืžื™ืœื™ืืจื“ ื ื•ื™ืจื•ื ื™ื,
02:49
which would you choose to try first?
48
169541
2169
ืžื” ื”ื™ื™ืชื ื‘ื•ื—ืจื™ื ืœื ืกื•ืช ืงื•ื“ื?
02:51
(Laughter)
49
171710
1167
(ืฆื—ื•ืง)
02:53
When studying these smaller insect brains,
50
173837
2586
ื›ืืฉืจ ื—ื•ืงืจื™ื ืืช ืžื•ื—ื•ืช ื”ื—ืจืงื™ื ื”ืงื˜ื ื™ื ื”ืืœื”,
02:56
the immediate goal is not human intelligence.
51
176464
2461
ื”ืžื˜ืจื” ื”ืžื™ื™ื“ื™ืช ืื™ื ื” ืื™ื ื˜ืœื™ื’ื ืฆื™ื” ืื ื•ืฉื™ืช.
02:59
We study these brains for what the insects do well.
52
179551
3461
ืื ื—ื ื• ื—ื•ืงืจื™ื ืืช ื”ืžื•ื—ื•ืช ื”ืืœื” ืขืœ ืžื” ืฉื”ื—ืจืงื™ื ืขื•ืฉื™ื ื˜ื•ื‘.
03:03
And in the case of the dragonfly, that's interception.
53
183805
2878
ื•ื‘ืžืงืจื” ืฉืœ ื”ืฉืคื™ืจื™ืช, ื–ื” ื™ื™ืจื•ื˜.
03:07
So when dragonflies are hunting,
54
187559
1543
ืื– ื›ืฉืฉืคื™ืจื™ื•ืช ืฆื“ื•ืช,
03:09
they do more than just fly straight at the prey.
55
189144
2502
ื”ืŸ ืขื•ืฉื•ืช ื™ื•ืชืจ ืžืกืชื ืœืขื•ืฃ ื™ืฉืจ ืืœ ื”ื˜ืจืฃ.
03:12
They fly in such a way that they will intercept it.
56
192021
2503
ื”ืŸ ืขืคื•ืช ื‘ืฆื•ืจื” ื›ื–ื• ืฉื”ืŸ ื™ื™ืจื˜ื• ืื•ืชื•.
03:14
They aim for where the prey is going to be.
57
194566
2460
ื”ืŸ ืžื›ื•ื•ื ื•ืช ืœืืŸ ืฉื”ื˜ืจืฃ ื”ื•ืœืš ืœื”ื™ื•ืช.
03:17
Much like a soccer player, running to intercept a pass.
58
197402
3003
ืžืžืฉ ื›ืžื• ืฉื—ืงืŸ ื›ื“ื•ืจื’ืœ, ืฉืจืฅ ืœื™ื™ืจื˜ ืžืกื™ืจื”.
03:21
To do this correctly,
59
201865
1459
ื›ื“ื™ ืœืขืฉื•ืช ื–ืืช ื ื›ื•ืŸ,
03:23
dragonflies need to perform what is known as a coordinate transformation,
60
203366
3879
ืฉืคื™ืจื™ื•ืช ืฆืจื™ื›ื•ืช ืœื‘ืฆืข ืืช ืžื” ืฉื™ื“ื•ืข ื›ืฉื™ื ื•ื™ ืงื•ืื•ืจื“ื™ื ื˜ื•ืช,
03:27
going from the eyeโ€™s frame of reference, or what the dragonfly sees,
61
207245
3504
ื™ื•ืฆืื™ื ืžืžืกื’ืจืช ื”ื”ืชื™ื™ื—ืกื•ืช ืฉืœ ื”ืขื™ืŸ, ืื• ืžื” ืฉื”ืฉืคื™ืจื™ืช ืจื•ืื”,
03:30
to the body's frame of reference,
62
210790
1585
ืœืžืกื’ืจืช ื”ื”ืชื™ื™ื—ืกื•ืช ืฉืœ ื”ื’ื•ืฃ,
03:32
or how the dragonfly needs to turn its body to intercept.
63
212375
2836
ืื• ืื™ืš ื”ืฉืคื™ืจื™ืช ืฆืจื™ื›ื” ืœื”ืคื ื•ืช ืืช ื’ื•ืคื” ื›ื“ื™ ืœื™ื™ืจื˜.
03:36
Coordinate transformations are a basic calculation
64
216004
3044
ืชื™ืื•ื ื˜ืจื ืกืคื•ืจืžืฆื™ื•ืช ื”ืŸ ื—ื™ืฉื•ื‘ ื‘ืกื™ืกื™
03:39
that animals need to perform to interact with the world.
65
219048
3713
ืฉื‘ืขืœื™ ื—ื™ื™ื ืฆืจื™ื›ื™ื ืœื‘ืฆืข ื›ื“ื™ ืœืงื™ื™ื ืื™ื ื˜ืจืืงืฆื™ื” ืขื ื”ืขื•ืœื.
03:43
We do them instinctively every time we reach for something.
66
223261
3086
ืื ื—ื ื• ืขื•ืฉื™ื ืื•ืชื ื‘ืื•ืคืŸ ืื™ื ืกื˜ื™ื ืงื˜ื™ื‘ื™ ื‘ื›ืœ ืคืขื ืฉืื ื—ื ื• ืžื•ืฉื™ื˜ื™ื ืœืžืฉื”ื•.
03:47
When I reach for an object straight in front of me,
67
227098
2878
ื›ืฉืื ื™ ืžื•ืฉื™ื˜ื” ื™ื“ ืœื—ืคืฅ ื™ืฉืจ ืžื•ืœื™,
03:50
my arm takes a very different trajectory than if I turn my head,
68
230018
3545
ื”ื™ื“ ืฉืœื™ ืœื•ืงื—ืช ืžืกืœื•ืœ ืฉื•ื ื” ืžืื•ื“ ืžืืฉืจ ืื ืื ื™ ืžืกื•ื‘ื‘ืช ืืช ืจืืฉื™,
03:53
look at that same object when it is off to one side
69
233605
2460
ืžืกืชื›ืœืช ืขืœ ืื•ืชื• ืื•ื‘ื™ื™ืงื˜ ื›ืืฉืจ ื”ื•ื ืžื•ืคื ื” ืœืฆื“ ืื—ื“
03:56
and reach for it there.
70
236107
1335
ื•ืžื›ื•ื•ื ืช ืœื”ื’ื™ืข ืืœื™ื• ืฉื.
03:58
In both cases, my eyes see the same image of that object,
71
238109
3712
ื‘ืฉื ื™ ื”ืžืงืจื™ื ื”ืขื™ื ื™ื™ื ืฉืœื™ ืจื•ืื•ืช ืืช ืื•ืชื” ืชืžื•ื ื” ืฉืœ ืื•ืชื• ืื•ื‘ื™ื™ืงื˜,
04:01
but my brain is sending my arm on a very different trajectory
72
241821
3712
ืื‘ืœ ื”ืžื•ื— ืฉืœื™ ืฉื•ืœื— ืืช ื–ืจื•ืขื™ ื‘ืžืกืœื•ืœ ืฉื•ื ื” ืžืื•ื“
04:05
based on the position of my neck.
73
245575
1919
ืขืœ ืกืžืš ืชื ื•ื—ืช ื”ืฆื•ื•ืืจ ืฉืœื™.
04:12
And dragonflies are fast.
74
252624
1960
ื•ืฉืคื™ืจื™ื•ืช ื”ืŸ ืžื”ื™ืจื•ืช.
04:15
This means they calculate fast.
75
255293
2085
ื–ื” ืื•ืžืจ ืฉื”ืŸ ืžื—ืฉื‘ื•ืช ืžื”ืจ.
04:18
The latency, or the time it takes for a dragonfly to respond
76
258046
4004
ื”ื”ืฉื”ื™ื”, ืื• ื”ื–ืžืŸ ืฉืœื•ืงื— ืœืฉืคื™ืจื™ืช ืœื”ื’ื™ื‘
04:22
once it sees the prey turn,
77
262091
1752
ื‘ืจื’ืข ืฉื”ื™ื ืจื•ืื” ืืช ื”ื˜ืจืฃ ืžืกืชื•ื‘ื‘,
04:23
is about 50 milliseconds.
78
263885
1960
ื”ื•ื ื‘ืขืจืš 50 ืืœืคื™ื•ืช ื”ืฉื ื™ื™ื”.
04:27
This latency is remarkable.
79
267180
2044
ื”ื”ืฉื”ื™ื” ื”ื–ื• ืžื“ื”ื™ืžื”.
04:30
For one thing, it's only half the time of a human eye blink.
80
270016
3378
ืจืืฉื™ืช, ื–ื” ืจืง ื—ืฆื™ ื”ื–ืžืŸ ืฉืœ ืžืฆืžื•ืฅ ืขื™ืŸ ืื ื•ืฉื™ืช.
04:34
But for another thing,
81
274020
1668
ืื‘ืœ ืœื“ื‘ืจ ืื—ืจ,
04:35
it suggests that dragonflies capture how to intercept
82
275688
3003
ื–ื” ืžืฆื™ืข ืฉืฉืคื™ืจื™ื•ืช ืชื•ืคืกื•ืช ื›ื™ืฆื“ ืœื™ื™ืจื˜
04:38
in only relatively or surprisingly few computational steps.
83
278733
4755
ืจืง ื‘ืฆืขื“ื™ื ื—ื™ืฉื•ื‘ื™ื™ื ืžืขื˜ื™ื ื™ื—ืกื™ืช ืื• ืžืคืชื™ืขื™ื.
04:44
So in the brain,
84
284364
1376
ืื– ื‘ืžื•ื—,
04:45
a computational step is a single neuron
85
285782
2878
ืฉืœื‘ ื—ื™ืฉื•ื‘ื™ ื”ื•ื ื ื•ื™ืจื•ืŸ ื‘ื•ื“ื“
04:48
or a layer of neurons working in parallel.
86
288660
2460
ืื• ืฉื›ื‘ื” ืฉืœ ื ื•ื™ืจื•ื ื™ื ื”ืคื•ืขืœื™ื ื‘ืžืงื‘ื™ืœ.
04:51
It takes a single neuron about 10 milliseconds
87
291996
3087
ืœื•ืงื— ืœื ื•ื™ืจื•ืŸ ื‘ื•ื“ื“ ื‘ืขืจืš 10 ืืœืคื™ื•ืช ื”ืฉื ื™ื™ื”
04:55
to add up all its inputs and respond.
88
295124
2336
ืœื—ื‘ืจ ืืช ื›ืœ ื”ืงืœื˜ื™ื ืฉืœื• ื•ืœื”ื’ื™ื‘.
04:58
The 50-millisecond response time
89
298169
2336
ื–ืžืŸ ืชื’ื•ื‘ื” ืฉืœ 50 ืืœืคื™ื•ืช ื”ืฉื ื™ื™ื”
05:00
means that once the dragonfly sees its prey turn,
90
300547
3503
ืื•ืžืจ ืฉื›ืฉืฉืคื™ืจื™ืช ืจื•ืื” ืืช ื”ื˜ืจืฃ ืฉืœื” ืžืกืชื•ื‘ื‘,
05:04
there's only time for maybe four of these computational steps
91
304092
3336
ื™ืฉ ืจืง ื–ืžืŸ ืื•ืœื™ ืœืืจื‘ืขื” ืฉืœื‘ื™ื ื—ื™ืฉื•ื‘ื™ื™ื ืืœื”
05:07
or four layers of neurons, working in sequence, one after the other,
92
307470
3837
ืื• ืืจื‘ืข ืฉื›ื‘ื•ืช ืฉืœ ื ื•ื™ืจื•ื ื™ื, ืขื•ื‘ื“ื™ื ื‘ืจืฆืฃ, ืื—ื“ ืื—ืจื™ ื”ืฉื ื™,
05:11
to calculate how the dragonfly needs to turn.
93
311349
2377
ืœื—ืฉื‘ ืื™ืš ื”ืฉืคื™ืจื™ืช ืฆืจื™ื›ื” ืœื”ืกืชื•ื‘ื‘.
05:14
In other words, if I want to study
94
314811
2085
ื‘ืžื™ืœื™ื ืื—ืจื•ืช, ืื ืื ื™ ืจื•ืฆื” ืœืœืžื•ื“
05:16
how the dragonfly does coordinate transformations,
95
316896
4213
ืื™ืš ื”ืฉืคื™ืจื™ืช ืžืชืืžืช ื˜ืจื ืกืคื•ืจืžืฆื™ื•ืช,
05:21
the neural circuit that I need to understand,
96
321109
2753
ื”ืžืขื’ืœ ื”ืขืฆื‘ื™ ืฉืื ื™ ืฆืจื™ื›ื” ืœื”ื‘ื™ืŸ,
05:23
the neural circuit that I need to study,
97
323903
2252
ื”ืžืขื’ืœ ื”ืขืฆื‘ื™ ืฉืื ื™ ืฆืจื™ื›ื” ืœื—ืงื•ืจ,
05:26
can have at most four layers of neurons.
98
326197
2670
ื™ื›ื•ืœื•ืช ืœื”ื™ื•ืช ืœื• ืœื›ืœ ื”ื™ื•ืชืจ ืืจื‘ืข ืฉื›ื‘ื•ืช ืฉืœ ื ื•ื™ืจื•ื ื™ื.
05:29
Each layer may have many neurons,
99
329784
2377
ื‘ื›ืœ ืฉื›ื‘ื” ืขืฉื•ื™ื™ื ืœื”ื™ื•ืช ื ื•ื™ืจื•ื ื™ื ืจื‘ื™ื,
05:32
but this is a small neural circuit.
100
332203
2127
ืื‘ืœ ื–ื” ืžืขื’ืœ ืขืฆื‘ื™ ืงื˜ืŸ.
05:34
Small enough that we can identify it
101
334789
2002
ืงื˜ืŸ ืžืกืคื™ืง ื›ื“ื™ ืฉื ื•ื›ืœ ืœื–ื”ื•ืช ืื•ืชื•
05:36
and study it with the tools that are available today.
102
336833
2669
ื•ืœื—ืงื•ืจ ืื•ืชื• ืขื ื”ื›ืœื™ื ืฉื–ืžื™ื ื™ื ื”ื™ื•ื.
05:40
And this is what I'm trying to do.
103
340670
1835
ื•ื–ื” ืžื” ืฉืื ื™ ืžื ืกื” ืœืขืฉื•ืช.
05:43
I have built a model of what I believe is the neural circuit
104
343298
3044
ื‘ื ื™ืชื™ ืžื•ื“ืœ ืฉืœ ืžื” ืฉืื ื™ ืžืืžื™ื ื” ืฉื”ื•ื ื”ืžืขื’ืœ ื”ืขืฆื‘ื™
05:46
that calculates how the dragonfly should turn.
105
346384
2544
ืฉืžื—ืฉื‘ ืื™ืš ื”ืฉืคื™ืจื™ืช ืฆืจื™ื›ื” ืœื”ืกืชื•ื‘ื‘.
05:49
And here is the cool result.
106
349596
1584
ื•ื”ื ื” ื”ืชื•ืฆืื” ื”ืžื’ื ื™ื‘ื”.
05:51
In the model,
107
351222
1627
ื‘ื“ื’ื,
05:52
dragonflies do coordinate transformations in only one computational step,
108
352891
4713
ืฉืคื™ืจื™ื•ืช ืื›ืŸ ืžืชืืžื•ืช ื˜ืจื ืกืคื•ืจืžืฆื™ื•ืช ืจืง ื‘ืฉืœื‘ ื—ื™ืฉื•ื‘ื™ ืื—ื“,
05:57
one layer of neurons.
109
357645
1627
ืฉื›ื‘ื” ืื—ืช ืฉืœ ื ื•ื™ืจื•ื ื™ื.
05:59
This is something we can test and understand.
110
359898
2877
ื–ื” ืžืฉื”ื• ืฉืื ื—ื ื• ื™ื›ื•ืœื™ื ืœื‘ื“ื•ืง ื•ืœื”ื‘ื™ืŸ.
06:03
In a computer simulation,
111
363651
1502
ื‘ื”ื“ืžื™ื™ืช ืžื—ืฉื‘,
06:05
I can predict the activities of individual neurons
112
365194
3170
ืื ื™ ื™ื›ื•ืœื” ืœื—ื–ื•ืช ืืช ื”ืคืขื™ืœื•ื™ื•ืช ืฉืœ ื ื•ื™ืจื•ื ื™ื ื‘ื•ื“ื“ื™ื
06:08
while the dragonfly is hunting.
113
368364
1669
ื‘ื–ืžืŸ ืฉื”ืฉืคื™ืจื™ืช ืฆื“ื”.
06:11
For example, here I am predicting the action potentials, or the spikes,
114
371367
4338
ืœืžืฉืœ, ื›ืืŸ ืื ื™ ื—ื•ื–ื” ืืช ืคื•ื˜ื ืฆื™ืืœ ื”ืคืขื•ืœื”, ืื• ื”ื”ื‘ื–ืงื™ื,
06:15
that are fired by one of these neurons
115
375747
1835
ืฉื ื•ืจื™ื ืขืœ ื™ื“ื™ ืื—ื“ ืžื”ื ื•ื™ืจื•ื ื™ื ื”ืืœื”
06:17
when the dragonfly sees the prey move.
116
377582
2627
ื›ืฉื”ืฉืคื™ืจื™ืช ืจื•ืื” ืืช ื”ื˜ืจืฃ ื–ื–.
06:22
To test the model, my collaborators and I
117
382545
2169
ื›ื“ื™ ืœื‘ื“ื•ืง ืืช ื”ืžื•ื“ืœ, ื”ืฉื•ืชืคื™ื ืฉืœื™ ื•ืื ื™
06:24
are now comparing these predicted neural responses
118
384756
2836
ืžืฉื•ื•ื™ื ืขื›ืฉื™ื• ืืช ืืœื” ืฉื—ื–ื• ืชื’ื•ื‘ื•ืช ืขืฆื‘ื™ื•ืช
06:27
with responses of neurons recorded in living dragonfly brains.
119
387592
3837
ืขื ืชื’ื•ื‘ื•ืช ืฉืœ ื ื•ื™ืจื•ื ื™ื ืฉืžืชื•ืขื“ื•ืช ื‘ืžื•ื—ื•ืช ืฉืคื™ืจื™ื•ืช ื—ื™ื•ืช.
06:33
These are ongoing experiments
120
393431
1668
ืืœื” ื”ื ื ื™ืกื•ื™ื™ื ืžืชืžืฉื›ื™ื
06:35
in which we put living dragonflies in virtual reality.
121
395099
3796
ืฉื‘ื”ื ืื ื• ืฉืžื™ื ืฉืคื™ืจื™ื•ืช ื—ื™ื•ืช ื‘ืžืฆื™ืื•ืช ืžื“ื•ืžื”.
06:40
(Laughter)
122
400188
1918
(ืฆื—ื•ืง)
06:42
Now, it's not practical to put VR goggles on a dragonfly.
123
402148
3754
ืขื›ืฉื™ื•, ื–ื” ืœื ืžืขืฉื™ ืœืฉื™ื ืžืฉืงืคื™ VR ืขืœ ืฉืคื™ืจื™ืช.
06:47
So instead, we show movies of moving targets to the dragonfly,
124
407070
4588
ืื– ื‘ืžืงื•ื ื–ื”, ืื ื—ื ื• ืžืจืื™ื ืœืฉืคื™ืจื™ืช ืกืจื˜ื™ื ืฉืœ ืžื˜ืจื•ืช ื ืขื•ืช,
06:51
while an electrode records activity patterns of individual neurons
125
411658
3670
ื‘ื–ืžืŸ ืฉืืœืงื˜ืจื•ื“ื” ืžืชืขื“ืช ืคืขื™ืœื•ืช ื“ืคื•ืกื™ื ืฉืœ ื ื•ื™ืจื•ื ื™ื ื‘ื•ื“ื“ื™ื
06:55
in the brain.
126
415328
1251
ื‘ืžื•ื—.
06:58
Yeah, he likes the movies.
127
418331
1335
ื›ืŸ, ื”ื™ื ืื•ื”ื‘ืช ืืช ื”ืกืจื˜ื™ื.
07:01
If the responses that we record in the brain
128
421167
2628
ืื ื”ืชื’ื•ื‘ื•ืช ืฉืื ื• ืจื•ืฉืžื™ื ื‘ืžื•ื—
07:03
match those predicted by the model,
129
423836
2128
ืžืชืื™ืžื•ืช ืœืืœื” ืฉื—ื–ื” ื”ืžื•ื“ืœ,
07:06
we will have identified which neurons are responsible
130
426005
2711
ื ื–ื”ื” ืื™ืœื• ืชืื™ ืขืฆื‘ ืื—ืจืื™ื
07:08
for coordinate transformations.
131
428758
1543
ืœื”ืชืืžื•ืช ื˜ืจื ืกืคื•ืจืžืฆื™ื•ืช.
07:11
The next step will be to understand the specifics
132
431010
2294
ื”ืฉืœื‘ ื”ื‘ื ื™ื”ื™ื” ืœื”ื‘ื™ืŸ ืืช ื”ืคืจื˜ื™ื
07:13
of how these neurons work together to do the calculation.
133
433346
3337
ื›ื™ืฆื“ ื ื•ื™ืจื•ื ื™ื ืืœื” ืคื•ืขืœื™ื ื™ื—ื“ ืœืขืฉื•ืช ืืช ื”ื—ื™ืฉื•ื‘.
07:16
But this is how we begin to understand how brains do basic
134
436724
3462
ืื‘ืœ ื›ื›ื” ืื ื—ื ื• ืžืชื—ื™ืœื™ื ืœื”ื‘ื™ืŸ ืื™ืš ืžื•ื—ื•ืช ืขื•ืฉื™ื
07:20
or primitive calculations.
135
440186
1627
ื—ื™ืฉื•ื‘ื™ื ื‘ืกื™ืกื™ื™ื ืื• ืคืจื™ืžื™ื˜ื™ื‘ื™ื.
07:22
Calculations that I regard as building blocks for more complex functions,
136
442188
4880
ื—ื™ืฉื•ื‘ื™ื ืฉืื ื™ ืจื•ืื” ื‘ื”ื ืื‘ื ื™ ื‘ื ื™ื™ืŸ ืœืคื•ื ืงืฆื™ื•ืช ืžื•ืจื›ื‘ื•ืช ื™ื•ืชืจ,
07:27
not only for interception but also for cognition.
137
447110
3086
ืœื ืจืง ืœื™ื™ืจื•ื˜ ืืœื ื’ื ืœื”ื›ืจื”.
07:32
The way that these neurons compute may be different from anything
138
452156
3170
ื”ื“ืจืš ืฉื‘ื” ื ื•ื™ืจื•ื ื™ื ืืœื” ืžื—ืฉื‘ื™ื ืขืฉื•ื™ื” ืœื”ื™ื•ืช ืฉื•ื ื” ืžื›ืœ ื“ื‘ืจ
07:35
that exists on a computer today.
139
455368
1877
ืฉืงื™ื™ื ื”ื™ื•ื ื‘ืžื—ืฉื‘.
07:37
And the goal of this work is to do more than just write code
140
457870
3587
ื•ืžื˜ืจืช ื”ืขื‘ื•ื“ื” ื”ื–ื• ื”ื™ื ืœืขืฉื•ืช ื™ื•ืชืจ ืžืกืชื ืœื›ืชื•ื‘ ืงื•ื“
07:41
that replicates the activity patterns of neurons.
141
461457
2419
ืฉืžืฉื›ืคืœ ืืช ื”ืคืขื™ืœื•ืช ืฉืœ ื“ืคื•ืกื™ื ืฉืœ ื ื•ื™ืจื•ื ื™ื.
07:44
We aim to build a computer chip
142
464293
1794
ืื ื• ืฉื•ืืคื™ื ืœื‘ื ื•ืช ืฉื‘ื‘ ืžื—ืฉื‘
07:46
that not only does the same things as biological brains
143
466129
2794
ืฉืœื ืจืง ืขื•ืฉื” ืืช ืื•ืชื ื”ื“ื‘ืจื™ื ื›ืžื•ื—ื•ืช ื‘ื™ื•ืœื•ื’ื™ื™ื
07:48
but does them in the same way as biological brains.
144
468965
2711
ืื‘ืœ ืขื•ืฉื” ืื•ืชื ื‘ืื•ืชื• ืื•ืคืŸ ื›ืžื•ื—ื•ืช ื‘ื™ื•ืœื•ื’ื™ื™ื.
07:52
This could lead to drones driven by computers
145
472885
3254
ื–ื” ืขืฉื•ื™ ืœื”ื•ื‘ื™ืœ ืœืจื—ืคื ื™ื ืžื•ื ืขื™ื ืขืœ ื™ื“ื™ ืžื—ืฉื‘ื™ื
07:56
the same size of the dragonfly's brain
146
476139
2002
ื‘ืื•ืชื• ื’ื•ื“ืœ ืฉืœ ืžื•ื—ื” ืฉืœ ื”ืฉืคื™ืจื™ืช
07:58
that captures some targets and avoid others.
147
478182
2837
ืฉืœื•ื›ื“ื™ื ื›ืžื” ืžื˜ืจื•ืช ื•ื ืžื ืขื™ื ืžืื—ืจื•ืช.
08:01
Personally, I'm hoping for a small army of these
148
481853
2586
ืื™ืฉื™ืช, ืื ื™ ืžืงื•ื•ื” ืœืฆื‘ื ืงื˜ืŸ ืฉืœ ืืœื”
08:04
to defend my backyard from mosquitoes in the summer.
149
484480
2461
ืœื”ื’ืŸ ืขืœ ื”ื—ืฆืจ ื”ืื—ื•ืจื™ืช ืฉืœื™ ืžืคื ื™ ื™ืชื•ืฉื™ื ื‘ืงื™ืฅ.
08:06
(Laughter)
150
486983
1460
(ืฆื—ื•ืง)
08:09
The GPS on your phone could be replaced by a new navigation device
151
489027
4170
ื ื™ืชืŸ ืœื”ื—ืœื™ืฃ ืืช ื”-GPS ื‘ื˜ืœืคื•ืŸ ืฉืœื›ื ื‘ืžื›ืฉื™ืจ ื ื™ื•ื•ื˜ ื—ื“ืฉ
08:13
based on dung beetles or ants
152
493239
1710
ืฉืžื‘ื•ืกืก ืขืœ ื—ื™ืคื•ืฉื™ื•ืช ื–ื‘ืœ ืื• ื ืžืœื™ื
08:14
that could guide you to the straight or the easy path home.
153
494991
3045
ืฉื™ื›ื•ืœ ืœื”ื“ืจื™ืš ืืชื›ื ืœื“ืจืš ื”ื™ืฉืจื” ืื• ื”ืงืœื” ื”ื‘ื™ืชื”.
08:18
And what would the power requirements of these devices be like?
154
498953
3378
ื•ืžื” ื™ื”ื™ื• ื“ืจื™ืฉื•ืช ื”ื—ืฉืžืœ ืฉืœ ื”ืžื›ืฉื™ืจื™ื ื”ืืœื”?
08:23
As small as it is --
155
503458
1501
ืงื˜ืŸ ื›ื›ืœ ืฉื™ื”ื™ื” --
08:25
Or, sorry -- as large as it is,
156
505001
1877
ืื•, ืกืœื™ื—ื” - ื’ื“ื•ืœ ื›ื›ืœ ืฉื™ื”ื™ื”,
08:26
the human brain is estimated to have the same power requirements
157
506878
3628
ื”ื”ืขืจื›ื” ื”ื™ื ืฉืœืžื•ื— ื”ืื ื•ืฉื™ ื™ืฉ ืืช ืื•ืชืŸ ื“ืจื™ืฉื•ืช ื”ืกืคืง
08:30
as a 20-watt light bulb.
158
510548
2044
ื›ืžื• ื ื•ืจืช 20 ื•ื•ืื˜.
08:32
Imagine if all brain-inspired computers
159
512633
2211
ืชืืจื• ืœืขืฆืžื›ื ืื ืœื›ืœ ื”ืžื—ืฉื‘ื™ื ื‘ื”ืฉืจืืช ื”ืžื•ื—
08:34
had the same extremely low-power requirements.
160
514844
2669
ื”ื™ื• ืื•ืชืŸ ื“ืจื™ืฉื•ืช ื”ืกืคืง ื ืžื•ื›ื•ืช ืžืื•ื“.
08:38
Your smartphone or your smartwatch probably needs charging every day.
161
518431
3837
ื”ื˜ืœืคื•ืŸ ื”ื—ื›ื, ืื• ื”ืฉืขื•ืŸ ื”ื—ื›ื ืฉืœื›ื ื›ื ืจืื” ืฆืจื™ื›ื™ื ื˜ืขื™ื ื” ื›ืœ ื™ื•ื.
08:42
Your new brain-inspired device might only need charging every few months,
162
522268
3587
ื”ืžื›ืฉื™ืจ ื”ื—ื“ืฉ ืฉืœื›ื ื‘ื”ืฉืจืืช ื”ืžื•ื— ืขืฉื•ื™ ืœื”ื–ื“ืงืง ืœื˜ืขื™ื ื” ืจืง ื›ืœ ื›ืžื” ื—ื•ื“ืฉื™ื,
08:45
or maybe even every few years.
163
525897
1668
ืื• ืื•ืœื™ ืืคื™ืœื• ื›ืœ ื›ืžื” ืฉื ื™ื.
08:49
The famous physicist, Richard Feynman, once said,
164
529275
3670
ื”ืคื™ื–ื™ืงืื™ ื”ืžืคื•ืจืกื, ืจื™ืฆโ€™ืจื“ ืคื™ื™ื ืžืŸ, ืืžืจ ืคืขื,
08:52
"What I cannot create, I do not understand."
165
532945
2878
โ€œืžื” ืฉืื ื™ ืœื ื™ื›ื•ืœ ืœื™ืฆื•ืจ, ืื ื™ ืœื ืžื‘ื™ืŸ.โ€
08:56
What I see in insect nervous systems
166
536783
2085
ืžื” ืฉืื ื™ ืจื•ืื” ื‘ืžืขืจื›ื•ืช ื”ืขืฆื‘ื™ื ืฉืœ ื—ืจืงื™ื
08:58
is an opportunity to understand brains
167
538868
2169
ื–ื• ื”ื–ื“ืžื ื•ืช ืœื”ื‘ื™ืŸ ืžื•ื—ื•ืช
09:01
through the creation of computers that work as brains do.
168
541079
3003
ื‘ืืžืฆืขื•ืช ื™ืฆื™ืจืช ืžื—ืฉื‘ื™ื ืฉืขื•ื‘ื“ื™ื ื›ืžื• ืฉื”ืžื•ื— ืขื•ื‘ื“.
09:05
And creation of these computers will not just be for knowledge.
169
545083
3712
ื•ื™ืฆื™ืจืช ื”ืžื—ืฉื‘ื™ื ื”ืืœื” ืœื ื™ื”ื™ื” ืจืง ืœื™ื“ืข.
09:08
There's potential for real impact on your devices, your vehicles,
170
548795
4254
ื™ืฉ ืคื•ื˜ื ืฆื™ืืœ ืœื”ืฉืคืขื” ืžืžืฉื™ืช ืขืœ ื”ืžื›ืฉื™ืจื™ื ืฉืœื›ื, ื›ืœื™ ื”ืจื›ื‘ ืฉืœื›ื,
09:13
maybe even artificial intelligences.
171
553091
2210
ืื•ืœื™ ืืคื™ืœื• ืื™ื ื˜ืœื™ื’ื ืฆื™ื•ืช ืžืœืื›ื•ืชื™ื•ืช.
09:16
So next time you see an insect,
172
556177
2419
ืื– ื‘ืคืขื ื”ื‘ืื” ืฉืชืจืื• ื—ืจืง,
09:18
consider that these tiny brains can lead to remarkable computers.
173
558638
4045
ืงื—ื• ื‘ื—ืฉื‘ื•ืŸ ืฉื”ืžื•ื—ื•ืช ื”ื–ืขื™ืจื™ื ื”ืืœื” ื™ื›ื•ืœื™ื ืœื”ื•ื‘ื™ืœ ืœืžื—ืฉื‘ื™ื ื™ื•ืฆืื™ ื“ื•ืคืŸ.
09:23
And think of the potential that they offer us for the future.
174
563518
3211
ื•ืชื—ืฉื‘ื• ืขืœ ื”ืคื•ื˜ื ืฆื™ืืœ ืฉื”ื ืžืฆื™ืขื™ื ืœื ื• ืœืขืชื™ื“.
09:27
Thank you.
175
567271
1126
ืชื•ื“ื”.
09:28
(Applause)
176
568439
3379
(ืžื—ื™ืื•ืช ื›ืคื™ื™ื)
ืขืœ ืืชืจ ื–ื”

ืืชืจ ื–ื” ื™ืฆื™ื’ ื‘ืคื ื™ื›ื ืกืจื˜ื•ื ื™ YouTube ื”ืžื•ืขื™ืœื™ื ืœืœื™ืžื•ื“ ืื ื’ืœื™ืช. ืชื•ื›ืœื• ืœืจืื•ืช ืฉื™ืขื•ืจื™ ืื ื’ืœื™ืช ื”ืžื•ืขื‘ืจื™ื ืขืœ ื™ื“ื™ ืžื•ืจื™ื ืžื”ืฉื•ืจื” ื”ืจืืฉื•ื ื” ืžืจื—ื‘ื™ ื”ืขื•ืœื. ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ื”ืžื•ืฆื’ื•ืช ื‘ื›ืœ ื“ืฃ ื•ื™ื“ืื• ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ ืžืฉื. ื”ื›ืชื•ื‘ื™ื•ืช ื’ื•ืœืœื•ืช ื‘ืกื ื›ืจื•ืŸ ืขื ื”ืคืขืœืช ื”ื•ื•ื™ื“ืื•. ืื ื™ืฉ ืœืš ื”ืขืจื•ืช ืื• ื‘ืงืฉื•ืช, ืื ื ืฆื•ืจ ืื™ืชื ื• ืงืฉืจ ื‘ืืžืฆืขื•ืช ื˜ื•ืคืก ื™ืฆื™ืจืช ืงืฉืจ ื–ื”.

https://forms.gle/WvT1wiN1qDtmnspy7