AI-Generated Creatures That Stretch the Boundaries of Imagination | Sofia Crespo | TED

45,678 views

2022-11-30 ใƒป TED


New videos

AI-Generated Creatures That Stretch the Boundaries of Imagination | Sofia Crespo | TED

45,678 views ใƒป 2022-11-30

TED


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

ืชืจื’ื•ื: zeeva livshitz ืขืจื™ื›ื”: Ido Dekkers
00:04
I'd like to start by asking you to imagine a color
0
4543
4379
ืื ื™ ืจื•ืฆื” ืœื”ืชื—ื™ืœ ืขื ื‘ืงืฉื” ืฉืชื“ืžื™ื™ื ื• ืฆื‘ืข
00:08
that you've never seen before.
1
8922
2544
ืฉืœื ืจืื™ืชื ืžืขื•ืœื.
00:12
Just for a second give this a try.
2
12217
2252
ืจืง ืœืจื’ืข, ื ืกื• ืืช ื–ื”.
00:14
Can you actually visualize a color that you've never been able to perceive?
3
14928
4463
ื”ืชื•ื›ืœื• ื‘ืืžืช ืœื“ืžื™ื™ืŸ ืฆื‘ืข ืฉืœื ื”ืฆืœื—ืชื ืœืชืคื•ืก?
00:20
I never seem to get tired of trying this
4
20892
2878
ื ืจืื” ืฉืืฃ ืคืขื ืœื ื ืžืืก ืœื™ ืœื ืกื•ืช ื–ืืช
00:23
although I know it's not an easy challenge.
5
23770
2795
ืœืžืจื•ืช ืฉืื ื™ ื™ื•ื“ืขืช ืฉื–ื” ืืชื’ืจ ืœื ืงืœ.
00:26
And the thing is,
6
26606
1210
ื•ื”ืขื ื™ื™ืŸ ื”ื•ื,
00:27
we can't imagine something without drawing upon our experiences.
7
27857
4880
ืื ื—ื ื• ืœื ื™ื›ื•ืœื™ื ืœื“ืžื™ื™ืŸ ืžืฉื”ื• ืžื‘ืœื™ ืœื”ืกืชืžืš ืขืœ ื”ื—ื•ื•ื™ื•ืช ืฉืœื ื•.
00:33
A color we haven't yet seen
8
33655
2002
ืฆื‘ืข ืฉืขื“ื™ื™ืŸ ืœื ืจืื™ื ื•
00:35
outside the spectrum we can perceive
9
35699
2836
ืžื—ื•ืฅ ืœืกืคืงื˜ืจื•ื ืฉืื ื• ื™ื›ื•ืœื™ื ืœืชืคื•ืก
00:38
is outside our ability to conjure up.
10
38535
2502
ื”ื•ื ืžื—ื•ืฅ ืœื™ื›ื•ืœืช ืฉืœื ื• ืœื“ืžื™ื™ืŸ.
00:42
It's almost like there's a boundary to our imagination
11
42372
3212
ื–ื” ื›ืžืขื˜ ื›ืื™ืœื• ื™ืฉ ื’ื‘ื•ืœ ืœื“ืžื™ื•ืŸ ืฉืœื ื•
00:45
where all the colors we can imagine
12
45625
2002
ืฉื‘ื• ื›ืœ ื”ืฆื‘ืขื™ื ืฉืื ื• ื™ื›ื•ืœื™ื ืœื“ืžื™ื™ืŸ
00:47
can only be various shades of other colors we have previously seen.
13
47627
4505
ื™ื›ื•ืœื™ื ืœื”ื™ื•ืช ื’ื•ื•ื ื™ื ืฉื•ื ื™ื ืฉืœ ืฆื‘ืขื™ื ืฉื›ื‘ืจ ืจืื™ื ื• ืงื•ื“ื ืœื›ืŸ.
00:52
Yet we know for a fact
14
52841
2210
ื•ื‘ื›ืœ ื–ืืช ืื ื—ื ื• ื™ื•ื“ืขื™ื ื‘ื•ื•ื“ืื•ืช
00:55
that those color frequencies outside our visible spectrum are there.
15
55051
4463
ื›ื™ ืชื“ืจื™ ื”ืฆื‘ืข ื”ืœืœื• ืžื—ื•ืฅ ืœืกืคืงื˜ืจื•ื ื”ื’ืœื•ื™ ืฉืœื ื• ื ืžืฆืื™ื ืฉื.
01:00
And scientists believe that there are species
16
60223
5339
ื•ืžื“ืขื ื™ื ืžืืžื™ื ื™ื ืฉื™ืฉ ืžื™ื ื™ื
01:05
that have many more photo receptors
17
65562
3503
ืฉื™ืฉ ืœื”ื ื”ืจื‘ื” ื™ื•ืชืจ ืงื•ืœื˜ื ื™ ืชืžื•ื ื”
01:09
than just the three color ones we humans have.
18
69107
4880
ืžืืฉืจ ืจืง 3 ื”ืฆื‘ืขื™ื ืฉื™ืฉ ืœื ื• ื‘ื ื™ ื”ืื“ื.
01:13
Which, by the way,
19
73987
1835
ืฉื“ืจืš ืื’ื‘,
01:15
not all humans see the world in the same way.
20
75864
3295
ืœื ื›ืœ ื‘ื ื™ ื”ืื“ื ืจื•ืื™ื ืืช ื”ืขื•ืœื ื‘ืื•ืชื” ื”ื“ืจืš.
01:19
Some of us are colorblind to various degrees,
21
79200
4505
ื—ืœืงื ื• ืขื™ื•ื•ืจื™ ืฆื‘ืขื™ื ื‘ื“ืจื’ื•ืช ืฉื•ื ื•ืช,
01:23
and very often we don't even agree on small things,
22
83705
4713
ื•ืœืขืชื™ื ืงืจื•ื‘ื•ืช ืื ื—ื ื• ืืคื™ืœื• ืœื ืžืกื›ื™ืžื™ื ืขืœ ื“ื‘ืจื™ื ืงื˜ื ื™ื,
01:28
like if a dress on the internet is blue and black or white and gold.
23
88460
5005
ื›ืžื• ืื ืฆื‘ืข ืฉืžืœื” ื‘ืื™ื ื˜ืจื ื˜ ื”ื•ื ื›ื—ื•ืœ ื•ืฉื—ื•ืจ ืื• ืœื‘ืŸ ื•ื–ื”ื‘.
01:34
But my favorite creature, one of my favorite creatures,
24
94215
4380
ืื‘ืœ ื”ื™ืฆื•ืจ ื”ืื”ื•ื‘ ืขืœื™ื™, ืื—ื“ ื”ื™ืฆื•ืจื™ื ื”ืื”ื•ื‘ื™ื ืขืœื™,
01:38
is the peacock mantis shrimp,
25
98595
2335
ื”ื•ื ื—ืกื™ืœื•ืŸ ื’ืžืœ ืฉืœืžื” ื˜ื•ื•ืกื™,
01:40
which is estimated to have 12 to 16 photo receptors.
26
100972
5297
ืืฉืจ ืžื•ืขืจืš ืฉื™ืฉ ืœื• 12 ืขื“-16 ืงื•ืœื˜ื ื™ ืื•ืจ.
01:46
And that indicates the world to them might look so much more colorful.
27
106269
5798
ื•ื–ื” ืžืขื™ื“ ืฉื”ืขื•ืœื ืขื‘ื•ืจื ืื•ืœื™ ื ืจืื” ื”ืจื‘ื” ื™ื•ืชืจ ืฆื‘ืขื•ื ื™.
01:54
So what about artificial intelligence?
28
114194
2836
ืื– ืžื” ืœื’ื‘ื™ ื‘ื™ื ื” ืžืœืื›ื•ืชื™ืช?
01:57
Can AI help us see beyond our human capabilities?
29
117947
3838
ื”ืื ื‘โ€œืž ื™ื›ื•ืœื” ืœืขื–ื•ืจ ืœื ื• ืœืจืื•ืช ืžืขื‘ืจ ืœื™ื›ื•ืœื•ืช ื”ืื ื•ืฉื™ื•ืช ืฉืœื ื•?
02:03
Well, I've been working with AI for the past five years,
30
123203
3670
ื•ื‘ื›ืŸ, ืขื‘ื“ืชื™ ืขื ื‘โ€œืž ื‘ื—ืžืฉ ื”ืฉื ื™ื ื”ืื—ืจื•ื ื•ืช,
02:06
and in my experience, it can see within the data it gets fed.
31
126873
4630
ื•ืžื ื™ืกื™ื•ื ื™, ื”ื™ื ื™ื›ื•ืœื” ืœืจืื•ืช ื‘ืชื•ืš ื”ื ืชื•ื ื™ื ืฉื”ื™ื ืžืงื‘ืœืช.
02:12
But then you might be wondering, OK,
32
132420
2586
ืื‘ืœ ืื– ืื•ืœื™ ืืชื ืชื•ื”ื™ื, ื‘ืกื“ืจ,
02:15
if AI can't help imagine anything new,
33
135006
3337
ืื ื‘โ€œืž ืœื ื™ื›ื•ืœื” ืœืขื–ื•ืจ ืœื“ืžื™ื™ืŸ ืžืฉื”ื• ื—ื“ืฉ,
02:18
why would an artist see any point in using it?
34
138385
2794
ืœืžื” ืฉืืžืŸ ื™ืžืฆื ื˜ืขื ืœื”ืฉืชืžืฉ ื‘ื”?
02:21
And my answer to that is because I think that it can help augment our creativity
35
141930
4254
ื•ื”ืชืฉื•ื‘ื” ืœื–ื” ื”ื™ื ื›ื™ ืื ื™ ื—ื•ืฉื‘ืช ืฉื”ื™ื ื™ื›ื•ืœื” ืœืขื–ื•ืจ ืœื”ื’ื‘ื™ืจ ืืช ื”ื™ืฆื™ืจืชื™ื•ืช ืฉืœื ื•
02:26
as there's value in creating combinations of known elements to form new ones.
36
146226
6173
ืฉื›ืŸ ื™ืฉ ืขืจืš ื‘ื™ืฆื™ืจืช ืฉื™ืœื•ื‘ื™ื ืฉืœ ืืœืžื ื˜ื™ื ื™ื“ื•ืขื™ื ื›ื“ื™ ืœื™ืฆื•ืจ ืืœืžื ื˜ื™ื ื—ื“ืฉื™ื.
02:33
And this boundary of what we can imagine based on what we have experienced
37
153858
5422
ื•ื”ื’ื‘ื•ืœ ื”ื–ื” ืฉืœ ืžื” ืฉืื ื—ื ื• ื™ื›ื•ืœื™ื ืœื“ืžื™ื™ืŸ ืขืœ ืกืžืš ืžื” ืฉื—ื•ื•ื™ื ื•
02:39
is the place that I have been exploring.
38
159322
2419
ื”ื•ื ื”ืžืงื•ื ืื•ืชื• ื—ืงืจืชื™.
02:42
For me, it started with jellyfish on a screen at an aquarium
39
162200
5005
ืืฆืœื™, ื–ื” ื”ืชื—ื™ืœ ืขื ืžื“ื•ื–ื” ืขืœ ืžืกืš ื‘ืืงื•ื•ืจื™ื•ื
02:47
and wearing those old 3D glasses, which I hope you remember,
40
167247
3712
ื•ื”ืจื›ื‘ืช ืžืฉืงืคื™ ื”ืชืœืช ืžื™ืžื“, ืฉืื ื™ ืžืงื•ื•ื” ืฉืืชื ื–ื•ื›ืจื™ื,
02:51
the ones with the blue and red lens.
41
171000
1919
ืืœื” ืขื ื”ืขื“ืฉื” ื”ื›ื—ื•ืœื” ื•ื”ืื“ื•ืžื”.
02:53
And this experience made me want to recreate their textures.
42
173336
4213
ื•ื”ื—ื•ื•ื™ื” ื”ื–ื• ื’ืจืžื” ืœื™ ืœืจืฆื•ืช ืœื™ืฆื•ืจ ืžื—ื“ืฉ ืืช ื”ืžืจืงืžื™ื ืฉืœื”ื.
02:57
But not just that,
43
177882
1168
ืื‘ืœ ืœื ืจืง ื–ื”,
02:59
I also wanted to create new jellyfish
44
179092
1960
ืจืฆื™ืชื™ ื’ื ืœื™ืฆื•ืจ ืžื“ื•ื–ื” ื—ื“ืฉื”
03:01
that I hadn't seen before, like these.
45
181052
2878
ืฉืœื ืจืื™ืชื™ ืงื•ื“ื, ื›ืžื• ืืœื•.
03:04
And what started with jellyfish,
46
184556
1543
ื•ืžื” ืฉื”ืชื—ื™ืœ ืขื ืžื“ื•ื–ื”,
03:06
very quickly escalated to other sea creatures
47
186099
3587
ืžื”ืจ ื”ืชื“ืจื“ืจ ืœื™ืฆื•ืจื™ื ื™ืžื™ื™ื ืื—ืจื™ื
03:09
like sea anemone, coral and fish.
48
189686
4254
ื›ืžื• ืฉื•ืฉื ืช ืžื™ื, ืืœืžื•ื’ื™ื ื•ื“ื’ื™ื.
03:14
And then from there came amphibians, birds and insects.
49
194524
5756
ื•ืื– ืžืฉื ื”ื’ื™ืขื• ื“ื•-ื—ื™ื™ื, ืฆื™ืคื•ืจื™ื ื•ื—ืจืงื™ื.
03:20
And this became a series called โ€œNeural Zooโ€.
50
200739
2919
ื•ื–ื” ื”ืคืš ืœืกื“ืจื” ืฉื ืงืจืืช โ€œื’ืŸ ื”ื—ื™ื•ืช ื”ืขืฆื‘ื™โ€œ.
03:25
But when you look closely, what do you see?
51
205618
3629
ืื‘ืœ ื›ืฉืžืกืชื›ืœื™ื ืžืงืจื•ื‘, ืžื” ืืชื ืจื•ืื™ื?
03:29
There's no single creature in these images.
52
209289
3962
ืื™ืŸ ื™ืฆื•ืจ ื™ื—ื™ื“ ื‘ืชืžื•ื ื•ืช ื”ืœืœื•.
03:33
And AI augments my creative process
53
213710
3378
ื‘โ€œืž ืžื’ื“ื™ืœื” ืืช ื”ืชื”ืœื™ืš ื”ื™ืฆื™ืจืชื™ ืฉืœื™
03:37
by allowing me to distill and recombine textures.
54
217130
4671
ื‘ื›ืš ืฉื”ื™ื ืžืืคืฉืจืช ืœื™ ืœื–ืงืง ื•ืœืฉืœื‘ ืžื—ื“ืฉ ืžืจืงืžื™ื.
03:41
And that's something that would otherwise take me months to draw by hand.
55
221801
4588
ื•ื–ื” ืžืฉื”ื• ืฉืื ืœื ื›ืš ื™ืงื— ืœื™ ื—ื•ื“ืฉื™ื ืœืฆื™ื™ืจ ื‘ื™ื“.
03:47
Plus I'm actually terrible at drawing.
56
227015
1877
ื‘ื ื•ืกืฃ, ืื ื™ ืžืžืฉ ื’ืจื•ืขื” ื‘ืฆื™ื•ืจ.
03:49
So you could say, in a way, what I'm doing
57
229476
3169
ืื– ืชื•ื›ืœื• ืœื•ืžืจ, ื‘ืžื•ื‘ืŸ ืžืกื•ื™ื ืฉืžื” ืฉืื ื™ ืขื•ืฉื”
03:52
is a contemporary version of something
58
232687
1960
ื”ื•ื ื’ื™ืจืกื” ืขื›ืฉื•ื•ื™ืช ืฉืœ ืžืฉื”ื•
03:54
that humans have already been doing for a long time,
59
234689
2753
ืฉื‘ื ื™ ืื“ื ื›ื‘ืจ ืขืฉื• ื”ืจื‘ื” ื–ืžืŸ,
03:57
even before cameras existed.
60
237484
2669
ืืคื™ืœื• ืœืคื ื™ ืฉื”ื™ื• ืžืฆืœืžื•ืช.
04:01
In medieval times,
61
241738
1793
ื‘ื™ืžื™ ื”ื‘ื™ื ื™ื™ื,
04:03
people went on expeditions,
62
243573
2211
ืื ืฉื™ื ื™ืฆืื• ืœืžืกืขื•ืช,
04:05
and when they came back they would share about what they saw
63
245825
3504
ื•ื›ืฉื”ื ื—ื–ืจื• ื”ื ืฉื™ืชืคื• ืืช ืžื” ืฉื”ื ืจืื•
04:09
to an illustrator.
64
249370
1377
ืขื ืžืื™ื™ืจ.
04:10
And the illustrator, having never seen what was being described,
65
250789
4129
ื•ื”ืžืื™ื™ืจ ืฉืžืขื•ืœื ืœื ืจืื” ืืช ืžื” ืฉืชื•ืืจ,
04:14
would end up drawing
66
254959
1627
ืฆื™ื™ืจ ื‘ืกื•ืคื• ืฉืœ ื“ื‘ืจ
04:16
based on the creatures that they had previously seen
67
256628
2460
ื‘ื”ืชื‘ืกืก ืขืœ ื”ื™ืฆื•ืจื™ื ืฉื”ื•ื ืจืื” ืงื•ื“ื ืœื›ืŸ,
04:19
and in the process creating hybrid animals of some sort.
68
259088
3170
ื•ื‘ืชื”ืœื™ืš ื™ืฆืจ ื—ื™ื•ืช ื”ื™ื‘ืจื™ื“ื™ื•ืช ืžืกื•ื’ ื›ืœืฉื”ื•.
04:22
So an explorer might describe a beaver, but having never seen one,
69
262675
4338
ืื– ื—ื•ืงืจ ืขืฉื•ื™ ืœืชืืจ ื‘ื•ื ื”, ืื‘ืœ ืžื›ื™ื•ื•ืŸ ืฉืœื ืจืื” ื›ื–ื”
04:27
the illustrator might give it the head of a rodent,
70
267055
2419
ื”ืžืื™ื™ืจ ืขืฉื•ื™ ืœืฆื™ื™ืจ ืœื• ืจืืฉ ืฉืœ ื—ื•ืœื“,
04:29
the body of a dog and a fish-like tail.
71
269474
2419
ื’ื•ืฃ ืฉืœ ื›ืœื‘ ื•ื–ื ื‘ ื“ืžื•ื™ ื“ื’.
04:32
In the series โ€œArtificial Natural Historyโ€,
72
272560
2670
ื‘ืกื“ืจื” โ€œื”ื™ืกื˜ื•ืจื™ื” ื˜ื‘ืขื™ืช ืžืœืื›ื•ืชื™ืชโ€œ,
04:35
I took thousands of illustrations from a natural history archives,
73
275271
4588
ืฆื™ืœืžืชื™, ืืœืคื™ ืื™ื•ืจื™ื ืžืืจื›ื™ื•ื ื™ ื”ื™ืกื˜ื•ืจื™ื” ื˜ื‘ืขื™ืช,
04:39
and I fed them to a neural network to generate new versions of them.
74
279901
5297
ื•ื”ื–ื ืชื™ ืื•ืชื ืœืจืฉืช ืขืฆื‘ื™ืช ื›ื“ื™ ืœื™ืฆื•ืจ ื’ืจืกืื•ืช ื—ื“ืฉื•ืช ืฉืœื”ื.
04:45
But up until now, all my work was done in 2D.
75
285990
4171
ืื‘ืœ ืขื“ ืขื›ืฉื™ื•, ื›ืœ ื”ืขื‘ื•ื“ื” ืฉืœื™ ื ืขืฉืชื” ื‘ื“ื• ืžื™ืžื“.
04:51
And with the help of my studio partner, Feileacan McCormick,
76
291246
3545
ื•ื‘ืขื–ืจืช ื”ืฉื•ืชืฃ ืฉืœื™ ืœืกื˜ื•ื“ื™ื•, ืคื™ื™ืœืืงืŸ ืžืงื•ืจืžื™ืง,
04:54
we decided to train a neural network on a data set of 3D scanned beetles.
77
294833
5422
ื”ื—ืœื˜ื ื• ืœืืžืŸ ืจืฉืช ืขืฆื‘ื™ืช ืขืœ ืกื˜ ื ืชื•ื ื™ื ืฉืœ ื—ื™ืคื•ืฉื™ื•ืช ืกืจื•ืงื•ืช ื‘ืชืœืช ืžื™ืžื“.
05:00
But I must warn you that our first results were extremely blurry,
78
300839
4546
ืื‘ืœ ืขืœื™ ืœื”ื–ื”ื™ืจ ืืชื›ื ืฉื”ืชื•ืฆืื•ืช ื”ืจืืฉื•ื ื™ื•ืช ืฉืœื ื• ื”ื™ื• ืžื˜ื•ืฉื˜ืฉื•ืช ื‘ื™ื•ืชืจ,
05:05
and they looked like the blobs you see here.
79
305426
2253
ื•ื ืจืื• ื›ืžื• ื”ื›ืชืžื™ื ืฉืืชื ืจื•ืื™ื ื›ืืŸ.
05:08
And this could be due to many reasons,
80
308346
1835
ื•ื–ื” ื™ื›ื•ืœ ืœื”ื™ื•ืช ื‘ื’ืœืœ ื”ืจื‘ื” ืกื™ื‘ื•ืช,
05:10
but one of them being that there aren't really a lot
81
310181
2461
ืื‘ืœ ืื—ืช ืžื”ืŸ ื”ื™ื
ืฉืื™ืŸ ื”ืจื‘ื” ืžื™ื“ืข ื–ืžื™ืŸ ืขืœ ื—ืจืงื™ื ืชืœืช ืžื™ืžื“ื™ื™ื.
05:12
of openly available data sets of 3D insects.
82
312642
4504
05:17
And also we were repurposing
83
317188
2127
ื•ื’ื ืขืฉื™ื ื• ืฉื™ืžื•ืฉ ืžื—ื“ืฉ
05:19
a neural network that normally gets used to generate images to generate 3D.
84
319357
5214
ื‘ืจืฉืช ืขืฆื‘ื™ืช ืฉื‘ื“ืจืš ื›ืœืœ ืžืชืจื’ืœืช ืœื™ืฆื•ืจ ืชืžื•ื ื•ืช ื›ื“ื™ ืœื™ืฆื•ืจ ืชืœืช ืžื™ืžื“.
05:24
So believe it or not, these are very exciting blobs to us.
85
324612
4380
ืื– ืชืืžื™ื ื• ืื• ืœื, ืืœื” ื›ืชืžื™ื ืžืื•ื“ ืžืจื’ืฉื™ื ืขื‘ื•ืจื ื•.
05:29
But with time and some very hacky solutions
86
329784
3879
ืื‘ืœ ืขื ื”ื–ืžืŸ ื•ื›ืžื” ืคืชืจื•ื ื•ืช ื”ืืงื™ื™ื ืžืื•ื“
05:33
like data augmentation,
87
333705
2294
ื›ืžื• ื”ื’ื“ืœืช ื ืชื•ื ื™ื,
05:36
where we threw in ants and other beetle-like insects
88
336040
3754
ืฉื‘ื• ื–ืจืงื ื• ืคื ื™ืžื” ื ืžืœื™ื ื•ื—ืจืงื™ื ื“ืžื•ื™ื™ ื—ื™ืคื•ืฉื™ืช ืื—ืจื™ื
05:39
to enhance the data set,
89
339794
2503
ื›ื“ื™ ืœืฉืคืจ ืืช ืžืขืจืš ื”ื ืชื•ื ื™ื,
05:42
we ended up getting this,
90
342338
2628
ื‘ืกื•ืคื• ืฉืœ ื“ื‘ืจ ืงื™ื‘ืœื ื• ืืช ื–ื”,
05:44
which we've been told they look like grilled chicken.
91
344966
2544
ืžื” ืฉืืžืจื• ืœื ื• ืฉื”ื ื ืจืื™ื ื›ืžื• ืขื•ืฃ ื‘ื’ืจื™ืœ.
05:47
(Laughter)
92
347552
1752
(ืฆื—ื•ืง)
05:49
But hungry for more, we pushed our technique,
93
349345
4797
ืื‘ืœ ืจืขื‘ื™ื ืœืขื•ื“, ื“ื—ืคื ื• ืืช ื”ื˜ื›ื ื™ืงื” ืฉืœื ื•,
05:54
and eventually they ended up looking like this.
94
354183
3713
ื•ื‘ืกื•ืคื• ืฉืœ ื“ื‘ืจ ื”ื ืกื™ื™ืžื• ื›ืฉื”ื ื ืจืื™ื ื›ืš.
05:58
We use something called 3D style transfer to map textures onto them,
95
358479
5047
ืื ื—ื ื• ืžืฉืชืžืฉื™ื ื‘ืžืฉื”ื• ืฉื ืงืจื ื”ืขื‘ืจื” ื‘ืกื’ื ื•ืŸ ืชืœืช ืžื™ืžื“
ืœืžืคื•ืช ืขืœื™ื”ื ื˜ืงืกื˜ื•ืจื•ืช, ื•ื’ื ืื™ืžื ื• ืžื•ื“ืœ ืฉืคื” ื˜ื‘ืขื™ืช
06:03
and we also trained a natural language model
96
363568
3670
06:07
to generate scientific-like names
97
367280
2544
ืœื™ืฆื•ืจ ืฉืžื•ืช ืžื“ืขื™ื™ื
06:09
and anatomical descriptions.
98
369866
1918
ื•ืชื™ืื•ืจื™ื ืื ื˜ื•ืžื™ื™ื.
06:12
And eventually we even found a network architecture that could handle 3D meshes.
99
372869
4963
ื•ื‘ืกื•ืคื• ืฉืœ ื“ื‘ืจ ืืคื™ืœื• ืžืฆืื ื• ืจืฉืช ืืจื›ื™ื˜ืงื˜ื•ืจื” ืฉื™ื›ื•ืœื” ืœื”ืชืžื•ื“ื“
ืขื ืจืฉืชื•ืช ืชืœืช ืžื™ืžื“ื™ื•ืช. ืื– ื‘ืกื•ืคื• ืฉืœ ื“ื‘ืจ ื”ื ื ืจืื• ื›ื›ื”.
06:17
So they ended up looking like this.
100
377874
2377
06:21
And for us, this became a way of creating kind of a speculative study --
101
381836
4838
ื•ื‘ืฉื‘ื™ืœื ื• ื–ื” ื”ืคืš ืœื”ื™ื•ืช ื“ืจืš ืœื™ืฆื™ืจืช ืกื•ื’ ืฉืœ ืžื—ืงืจ ืกืคืงื•ืœื˜ื™ื‘ื™ --
06:26
(Applause)
102
386716
2669
(ืžื—ื™ืื•ืช ื›ืคื™ื™ื)
06:29
A speculative study of creatures that never existed,
103
389427
3629
ืžื—ืงืจ ืกืคืงื•ืœื˜ื™ื‘ื™ ืฉืœ ื™ืฆื•ืจื™ื ืฉืžืขื•ืœื ืœื ื”ื™ื• ืงื™ื™ืžื™ื,
06:33
kind of like a speculative biology.
104
393097
2294
ืกื•ื’ ืฉืœ ื‘ื™ื•ืœื•ื’ื™ื” ืกืคืงื•ืœื˜ื™ื‘ื™ืช.
06:37
But I didn't want to talk about AI and its potential
105
397685
4421
ืื‘ืœ ืœื ืจืฆื™ืชื™ ืœื“ื‘ืจ ืขืœ ื‘โ€œืž ื•ื”ืคื•ื˜ื ืฆื™ืืœ ืฉืœื”
06:42
unless it brought me closer to a real species.
106
402148
3629
ืืœื ืื ื›ืŸ ื–ื” ืงื™ืจื‘ ืื•ืชื™ ืœืžื™ื ื™ื ืืžื™ืชื™ื™ื.
06:46
Which of these do you think is easier to find data about online?
107
406486
4880
ืขืœ ืื™ื–ื” ืžื‘ื™ืŸ ืืœื” ืœื“ืขืชื›ื ืงืœ ื™ื•ืชืจ ืœืžืฆื•ื ื ืชื•ื ื™ื ื‘ืื™ื ื˜ืจื ื˜?
06:51
(Laughter)
108
411366
1918
(ืฆื—ื•ืง)
06:53
Yeah, well, as you guessed correctly, the red panda.
109
413326
4004
ื›ืŸ, ื˜ื•ื‘, ื›ืžื• ืฉื ื™ื—ืฉืชื ื ื›ื•ืŸ, ื”ืคื ื“ื” ื”ืื“ื•ืžื”.
06:57
And this maybe could be due to many reasons,
110
417830
3420
ื•ื–ื” ืื•ืœื™ ื™ื›ื•ืœ ืœื”ื™ื•ืช ืžืกื™ื‘ื•ืช ืจื‘ื•ืช,
07:01
but one of them being how cute they are,
111
421292
4046
ืื‘ืœ ืื—ืช ืžื”ืŸ ื”ื™ื ื›ืžื” ืฉื”ื ื—ืžื•ื“ื™ื,
07:05
which means we photograph and talk about them a lot,
112
425338
4588
ืžื” ืฉืื•ืžืจ ืฉืื ื—ื ื• ืžืฆืœืžื™ื ื•ืžื“ื‘ืจื™ื ืขืœื™ื”ื ื”ืจื‘ื”,
07:09
unlike the boreal felt lichen.
113
429926
2002
ื‘ื ื™ื’ื•ื“ ืœื—ื–ื–ื™ืช ื”ืœื‘ื“ ื”ืฆืคื•ื ื™ืช.
07:12
But both of them are classified as endangered.
114
432679
2877
ืื‘ืœ ืฉืชื™ื”ืŸ ืžืกื•ื•ื’ื•ืช ื›ื‘ืกื›ื ืช ื”ื›ื—ื“ื”.
07:16
So I wanted to bring visibility to other endangered species
115
436349
4880
ืื– ืจืฆื™ืชื™ ืœื”ื‘ื™ื ื ื™ืจืื•ืช ืœืžื™ื ื™ื ืื—ืจื™ื ื‘ืกื›ื ืช ื”ื›ื—ื“ื”
07:21
that don't get the same amount of digital representation
116
441270
4755
ืฉืœื ืžืงื‘ืœื™ื ืืช ืื•ืชื” ื›ืžื•ืช ืฉืœ ื™ื™ืฆื•ื’ ื“ื™ื’ื™ื˜ืœื™
07:26
as a cute, fluffy red panda.
117
446025
2002
ื›ืžื• ืคื ื“ื” ืื“ื•ืžื” ื—ืžื•ื“ื” ื•ืจื›ื”.
07:28
And to do this,
118
448861
1710
ื•ื›ื“ื™ ืœืขืฉื•ืช ื–ืืช,
07:30
we trained an AI on millions of images of the natural world,
119
450613
5005
ืื™ืžื ื• ื‘โ€œืž ืขืœ ืžื™ืœื™ื•ื ื™ ืชืžื•ื ื•ืช ืฉืœ ืขื•ืœื ื”ื˜ื‘ืข,
07:35
and then we prompted with text
120
455660
2085
ื•ืื– ื‘ื™ืงืฉื ื• ืขื ื˜ืงืกื˜
07:37
to generate some of these creatures.
121
457787
2294
ืœื™ืฆื•ืจ ื›ืžื” ืžื”ื™ืฆื•ืจื™ื ื”ืืœื”.
07:40
So when prompted with a text,
122
460123
3712
ืื– ื›ืฉืžื‘ืงืฉื™ื ืขื ื˜ืงืกื˜,
07:43
"an image of a critically endangered spider, the peacock tarantula"
123
463876
4713
โ€œืชืžื•ื ื” ืฉืœ ืขื›ื‘ื™ืฉ ื‘ืกื›ื ืช ื”ื›ื—ื“ื” ื—ืžื•ืจื”, ื˜ืจื ื˜ื•ืœืช ื”ื˜ื•ื•ืกโ€
07:48
and its scientific name,
124
468589
1669
ื•ื”ืฉื ื”ืžื“ืขื™ ืฉืœื•,
07:50
our model generated this.
125
470299
2378
ื”ืžื•ื“ืœ ืฉืœื ื• ื™ืฆืจ ืืช ื–ื”.
07:55
And here's an image of the real peacock tarantula,
126
475972
3211
ื•ื”ื ื” ืชืžื•ื ื” ืฉืœ ื˜ืจื ื˜ื•ืœืช ื”ื˜ื•ื•ืก ื”ืืžื™ืชื™,
07:59
which is a wonderful spider endemic to India.
127
479225
3128
ืฉื”ื•ื ืขื›ื‘ื™ืฉ ื ืคืœื ืื ื“ืžื™ ืœื”ื•ื“ื•.
08:02
But when prompted with a text
128
482812
2878
ืื‘ืœ ื›ืืฉืจ ืžื‘ืงืฉื™ื ืขื ื˜ืงืกื˜
08:05
"an image of a critically endangered bird, the mangrove finch,"
129
485732
4170
โ€œืชืžื•ื ื” ืฉืœ ืฆื™ืคื•ืจ ื‘ืกื›ื ืช ื”ื›ื—ื“ื” ื—ืžื•ืจื”, ื—ื•ื—ื™ืช ื”ืžื ื’ืจื•ื‘ื™ืโ€œ,
08:09
our model generated this.
130
489944
2669
ื”ืžื•ื“ืœ ืฉืœื ื• ื™ืฆืจ ืืช ื–ื”.
08:14
And here's a photo of the real mangrove finch.
131
494532
2711
ื•ื”ื ื” ืชืžื•ื ื” ืฉืœ ื—ื•ื—ื™ืช ื”ืžื ื’ืจื•ื‘ ื”ืืžื™ืชื™ืช.
08:17
Both these creatures exist in the wild,
132
497702
2878
ืฉื ื™ ื”ื™ืฆื•ืจื™ื ื”ืืœื” ืงื™ื™ืžื™ื ื‘ื˜ื‘ืข,
08:20
but the accuracy of each generated image is fully dependent on the data available.
133
500621
5965
ืื‘ืœ ื”ื“ื™ื•ืง ืฉืœ ื›ืœ ืชืžื•ื ื” ืฉื ื•ืฆืจื” ืชืœื•ื™ ืœื—ืœื•ื˜ื™ืŸ ื‘ื ืชื•ื ื™ื ื”ื–ืžื™ื ื™ื.
08:27
These chimeras of our everyday data
134
507462
2961
ื”ื›ื™ืžืจื•ืช ื”ืืœื” ืฉืœ ื”ื ืชื•ื ื™ื ื”ื™ื•ืžื™ื•ืžื™ื™ื ืฉืœื ื•
08:30
to me are a different way of how the future could be.
135
510423
3670
ื‘ืฉื‘ื™ืœื™ ื–ื• ื“ืจืš ืื—ืจืช ืœืื™ืš ื”ืขืชื™ื“ ื™ื›ื•ืœ ืœื”ื™ื•ืช.
08:34
Not in a literal sense, perhaps,
136
514802
2836
ืื•ืœื™ ืœื ื‘ืžื•ื‘ืŸ ื”ืžื™ืœื•ืœื™,
08:37
but in the sense that through practicing the expanding of our own imagination
137
517680
6423
ืืœื ื‘ืžื•ื‘ืŸ ืฉื“ืจืš ืชืจื’ื•ืœ ื”ืจื—ื‘ืช ื”ื“ืžื™ื•ืŸ ืฉืœื ื•
08:44
about the ecosystems we are a part of,
138
524145
3045
ื‘ื ื•ืฉื ื”ืžืขืจื›ื•ืช ื”ืืงื•ืœื•ื’ื™ื•ืช ืฉืื ื• ื—ืœืง ืžื”ืŸ,
08:47
we might just be better equipped to recognize new opportunities
139
527231
3170
ืื•ืœื™ ื ื”ื™ื” ืžืฆื•ื™ื“ื™ื ื™ื•ืชืจ ื˜ื•ื‘ ืœื–ื”ื•ืช ื”ื–ื“ืžื ื•ื™ื•ืช ื—ื“ืฉื•ืช
08:50
and potential.
140
530443
1335
ื•ืคื•ื˜ื ืฆื™ืืœ.
08:52
Knowing that there's a boundary to our imagination
141
532236
3128
ื”ื™ื“ื™ืขื” ืฉื™ืฉ ื’ื‘ื•ืœ ืœื“ืžื™ื•ืŸ ืฉืœื ื•
08:55
doesn't have to feel limiting.
142
535406
2211
ืœื ืžื—ื™ื™ื‘ืช ืœื—ื•ืฉ ืžื•ื’ื‘ืœื•ืช
08:58
On the contrary,
143
538159
1293
ืœื”ื™ืคืš,
08:59
it can help motivate us to expand that boundary further
144
539452
3420
ื–ื” ื™ื›ื•ืœ ืœื”ื ื™ืข ืื•ืชื ื• ืœื”ืจื—ื™ื‘ ืืช ื”ื’ื‘ื•ืœ ื”ื–ื” ืขื•ื“ ื™ื•ืชืจ
09:02
and to seek out colors and things we haven't yet seen
145
542914
3962
ื•ืœื—ืคืฉ ืฆื‘ืขื™ื ื•ื“ื‘ืจื™ื ืฉืขื“ื™ื™ืŸ ืœื ืจืื™ื ื•
09:06
and perhaps enrich our imagination as a result.
146
546918
3295
ื•ืื•ืœื™ ืœื”ืขืฉื™ืจ ืืช ื”ื“ืžื™ื•ืŸ ืฉืœื ื• ื›ืชื•ืฆืื” ืžื›ืš.
09:10
So thank you.
147
550546
1168
ืื– ืชื•ื“ื” ืœื›ื,
09:11
(Applause)
148
551714
4046
(ื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