Fake videos of real people -- and how to spot them | Supasorn Suwajanakorn

1,289,800 views ใƒป 2018-07-25

TED


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

ืชืจื’ื•ื: ืขืจื™ื›ื”: zeeva livshitz
00:12
Look at these images.
0
12876
1151
ื”ืกืชื›ืœื• ืขืœ ื”ืชืžื•ื ื•ืช ื”ืืœื•.
00:14
Now, tell me which Obama here is real.
1
14051
2635
ื•ืืžืจื• ืœื™: ืื™ื–ื” ืื•ื‘ืžื” ื›ืืŸ ื”ื•ื ื”ืืžื™ืชื™?
00:16
(Video) Barack Obama: To help families refinance their homes,
2
16710
2861
(ืกืจื˜ื•ืŸ ื‘ืจืง ืื•ื‘ืžื”: ืœืขื–ื•ืจ ืœืžืฉืคื—ื•ืช ืœืžืžืŸ ืžื—ื“ืฉ ืืช ื‘ืชื™ื”ืŸ,
00:19
to invest in things like high-tech manufacturing,
3
19595
2647
ืœื”ืฉืงื™ืข ื‘ื“ื‘ืจื™ื ื›ืžื• ืชืขืฉื™ื•ืช ื”ื™ื™-ื˜ืง,
00:22
clean energy
4
22266
1159
ืื ืจื’ื™ื” ื ืงื™ื™ื”
00:23
and the infrastructure that creates good new jobs.
5
23449
2779
ื•ืชืฉืชื™ื•ืช ืฉืžื™ื™ืฆืจื•ืช ืžืงื•ืžื•ืช ืขื‘ื•ื“ื” ื—ื“ืฉื™ื ื•ืื™ื›ื•ืชื™ื™ื.
00:26
Supasorn Suwajanakorn: Anyone?
6
26647
1484
ืกื•ืคืืกื•ืจืŸ ืกื•ื•ืื™ืื ืืงื•ืจืŸ: ืžื™ืฉื”ื•?
00:28
The answer is none of them.
7
28155
1874
ื”ืชืฉื•ื‘ื” ื”ื™ื: ืืฃ ืกืจื˜ื•ืŸ.
00:30
(Laughter)
8
30053
1114
(ืฆื—ื•ืง)
00:31
None of these is actually real.
9
31191
1786
ืืฃ ืื—ื“ ืžื”ืกื™ืจื˜ื•ื ื™ื ืื™ื ื• ืืžื™ืชื™.
ืื– ื”ืจืฉื• ืœื™ ืœืกืคืจ ืœื›ื ืื™ืš ื”ื’ืขื ื• ืœื›ืืŸ.
00:33
So let me tell you how we got here.
10
33001
1840
00:35
My inspiration for this work
11
35940
1578
ื”ื”ืฉืจืื” ืฉืœื™ ืœืžื—ืงืจ ื”ื–ื”
00:37
was a project meant to preserve our last chance for learning about the Holocaust
12
37542
5411
ื”ื™ืชื” ืคืจื•ื™ืงื˜ ืฉื ื•ืขื“ ืœืฉืžืจ ืืช ื”ืืคืฉืจื•ืช ื”ืื—ืจื•ื ื” ืฉืœื ื• ืœืœืžื•ื“ ืขืœ ื”ืฉื•ืื”
00:42
from the survivors.
13
42977
1768
ืžื”ื ื™ืฆื•ืœื™ื.
00:44
It's called New Dimensions in Testimony,
14
44769
2627
ื”ื•ื ื ืงืจื โ€œืžื™ืžื“ื™ื ื—ื“ืฉื™ื ื‘ืขื“ื•ืชโ€œ,
00:47
and it allows you to have interactive conversations
15
47420
3126
ื•ื”ื•ื ืžืืคืฉืจ ืœื›ื ืœื ื”ืœ ืฉื™ื—ื•ืช ืื™ื ื˜ืจืงื˜ื™ื‘ื™ื•ืช
00:50
with a hologram of a real Holocaust survivor.
16
50570
2556
ืขื ื”ื•ืœื•ื’ืจืžื” ืฉืœ ื ื™ืฆื•ืœ ืฉื•ืื” ืืžื™ืชื™.
00:53
(Video) Man: How did you survive the Holocaust?
17
53793
1966
(ืกืจื˜ื•ืŸ) ืื™ืฉ: ืื™ืš ืฉืจื“ืช ืืช ื”ืฉื•ืื”?
00:55
(Video) Hologram: How did I survive?
18
55783
1668
(ืกืจื˜ื•ืŸ) ื”ื•ืœื•ื’ืจืžื”: ืื™ืš ืฉืจื“ืชื™?
00:57
I survived,
19
57912
1807
ืฉืจื“ืชื™,
01:00
I believe,
20
60419
1527
ืื ื™ ืžืืžื™ืŸ,
01:01
because providence watched over me.
21
61970
3023
ืฉื”ื”ืฉื’ื—ื” ื”ืขืœื™ื•ื ื” ืฉืžืจื” ืขืœื™.
01:05
SS: Turns out these answers were prerecorded in a studio.
22
65573
3454
ืก.ืก: ืžืกืชื‘ืจ ืฉืชืฉื•ื‘ื•ืช ืืœื• ื”ื•ืงืœื˜ื• ืžืจืืฉ ื‘ืกื˜ื•ื“ื™ื•.
01:09
Yet the effect is astounding.
23
69051
2452
ื‘ื›ืœ ื–ืืช ื”ืืคืงื˜ ืžื“ื”ื™ื.
01:11
You feel so connected to his story and to him as a person.
24
71527
3619
ืืชื ืžืจื’ื™ืฉื™ื ื›ืœ ื›ืš ืžื—ื•ื‘ืจื™ื ืœืกื™ืคื•ืจ ืฉืœื• ื•ืืœื™ื• ื›ืžืกืคืจ.
ืื ื™ ื—ื•ืฉื‘ ืฉื™ืฉ ืžืฉื”ื• ืžื™ื•ื—ื“ ื‘ืชืงืฉื•ืจืช ืื ื•ืฉื™ืช
01:16
I think there's something special about human interaction
25
76011
3301
01:19
that makes it much more profound
26
79336
2757
ืฉื”ื•ืคืš ืื•ืชื” ืœื”ืจื‘ื” ื™ื•ืชืจ ืขืžื•ืงื”
01:22
and personal
27
82117
2198
ื•ืื™ืฉื™ืช
01:24
than what books or lectures or movies could ever teach us.
28
84339
3485
ื™ื•ืชืจ ืžืžื” ืฉืกืคืจื™ื ืื• ื”ืจืฆืื•ืช ืื• ืกืจื˜ื™ื ื™ื•ื›ืœื• ืื™ ืคืขื ืœืœืžื“ ืื•ืชื ื•.
01:28
So I saw this and began to wonder,
29
88267
2425
ืื– ืจืื™ืชื™ ืืช ื–ื” ื•ื”ืชื—ืœืชื™ ืœืชื”ื•ืช,
01:30
can we create a model like this for anyone?
30
90716
2810
ื”ืื ื ื•ื›ืœ ืœื™ืฆื•ืจ ืžื•ื“ืœ ื›ื–ื” ืœื›ืœ ืื—ื“?
01:33
A model that looks, talks and acts just like them?
31
93550
2975
ืžื•ื“ืœ ืฉื ืจืื”, ืžื“ื‘ืจ, ื•ืžืชื ื”ื’ ื‘ื“ื™ื•ืง ื›ืžื•ื”ื?
01:37
So I set out to see if this could be done
32
97573
2007
ืื– ื™ืฆืืชื™ ืœื‘ื“ื•ืง ืื ื ื™ืชืŸ ืœืขืฉื•ืช ืืช ื–ื”
01:39
and eventually came up with a new solution
33
99604
2310
ื•ืœื‘ืกื•ืฃ ืžืฆืืชื™ ืคืชืจื•ืŸ ื—ื“ืฉ
01:41
that can build a model of a person using nothing but these:
34
101938
3220
ืฉืžืืคืฉืจ ื‘ื ื™ื™ืช ืžื•ื“ืœ ืขโ€œื™ ืฉื™ืžื•ืฉ ืจืง ื‘ืืœื”:
01:45
existing photos and videos of a person.
35
105747
2214
ืชืžื•ื ื•ืช ืงื™ื™ืžื•ืช ื•ืกืจื˜ื•ื ื™ื ืฉืœ ืื“ื.
01:48
If you can leverage this kind of passive information,
36
108701
2617
ืื ืชื•ื›ืœื• ืœืžื ืฃ ืกื•ื’ ื›ื–ื” ืฉืœ ืžื™ื“ืข ืคืกื™ื‘ื™,
01:51
just photos and video that are out there,
37
111342
2007
ืคืฉื•ื˜ ืชืžื•ื ื•ืช ื•ืกืจื˜ื•ื ื™ื ืฉืืคืฉืจ ืœืžืฆื•ื ืฉื,
01:53
that's the key to scaling to anyone.
38
113373
2056
ื–ื” ื”ืžืคืชื— ืœื”ืชืื™ื ืืช ื–ื” ืœื›ืœ ืื—ื“.
01:56
By the way, here's Richard Feynman,
39
116119
1777
ื“ืจืš ืื’ื‘, ื–ื” ืจื™ืฆ'ืจื“ ืคื™ื™ื ืžืŸ,
01:57
who in addition to being a Nobel Prize winner in physics
40
117920
3413
ืฉื‘ื ื•ืกืฃ ืœื”ื™ื•ืชื• ื–ื•ื›ื” ืคืจืก ื ื•ื‘ืœ ื‘ืคื™ื–ื™ืงื”
02:01
was also known as a legendary teacher.
41
121357
2453
ื”ื™ื” ื™ื“ื•ืข ื’ื ื›ืžื•ืจื” ืžื“ื”ื™ื,
02:05
Wouldn't it be great if we could bring him back
42
125080
2198
ื”ืื ืœื ื™ื”ื™ื” ื–ื” ื ืคืœื, ืœื• ื™ื›ื•ืœื ื• ืœื”ื—ื–ื™ืจ ืื•ืชื•
02:07
to give his lectures and inspire millions of kids,
43
127302
3265
ื›ื“ื™ ืฉื™ืœืžื“ ืืช ื”ื”ืจืฆืื•ืช ืฉืœื• ื•ื™ืขื ื™ืง ื”ืฉืจืื” ืœืžื™ืœื™ื•ื ื™ ื™ืœื“ื™ื,
02:10
perhaps not just in English but in any language?
44
130591
2992
ืื•ืœื™ ืœื ืจืง ื‘ืื ื’ืœื™ืช ืืœื ื‘ื›ืœ ืฉืคื”? ืื• ืœื• ื™ื›ื•ืœื ื•
02:14
Or if you could ask our grandparents for advice and hear those comforting words
45
134441
4602
ืœื”ืชื™ื™ืขืฅ ืขื ื”ืกื‘ื™ื ื•ื”ืกื‘ืชื•ืช ืฉืœื ื•, ื•ืœืฉืžื•ืข ืืช ืื•ืชืŸ ืžื™ืœื™ื ืžื ื—ืžื•ืช
02:19
even if they're no longer with us?
46
139067
1770
ื’ื ืื ื”ื ื›ื‘ืจ ืœื ืื™ืชื ื•?
02:21
Or maybe using this tool, book authors, alive or not,
47
141683
3396
ืื• ืฉืื•ืœื™ ื‘ืขื–ืจืช ื”ื›ืœื™ ื”ื–ื” ืกื•ืคืจื™ื ื—ื™ื™ื ืื• ืžืชื™ื,
02:25
could read aloud all of their books for anyone interested.
48
145103
2937
ื™ืงืจื™ืื• ื‘ืงื•ืœ ืืช ื›ืœ ื”ืกืคืจื™ื ืฉืœื”ื, ืœื›ืœ ืžื™ ืฉืจื•ืฆื”.
02:29
The creative possibilities here are endless,
49
149199
2437
ื”ืืคืฉืจื•ื™ื•ืช ื”ื™ืฆื™ืจืชื™ื•ืช ื›ืืŸ ื”ืŸ ืื™ื ืกื•ืคื™ื•ืช,
02:31
and to me, that's very exciting.
50
151660
1713
ื•ื‘ืฉื‘ื™ืœื™, ื–ื” ืžืื•ื“ ืžืจื’ืฉ.
02:34
And here's how it's working so far.
51
154595
2002
ื•ื”ื ื” ืื™ืš ื–ื” ืขื•ื‘ื“ ืขื“ ื›ื”.
02:36
First, we introduce a new technique
52
156621
1667
ืงื•ื“ื ื›ืœ, ื”ืคืขืœื ื• ื˜ื›ื ื™ืงื” ื—ื“ืฉื”
02:38
that can reconstruct a high-detailed 3D face model from any image
53
158312
4572
ืฉื™ื›ื•ืœื” ืœืฉื—ื–ืจ ืžื›ืœ ืชืžื•ื ื”, ืžื•ื“ืœ ืชืœืช ืžื™ืžื“ื™ ืžืคื•ืจื˜ ืฉืœ ืคื ื™ื,
02:42
without ever 3D-scanning the person.
54
162908
2119
ืžื‘ืœื™ ืฉืกืจืงื ื• ื‘ืชืœืช ืžื™ืžื“ ืืช ื”ืื“ื ืขืฆืžื•.
02:45
And here's the same output model from different views.
55
165890
2642
ื•ื”ื ื” ืื•ืชื” ื”ื”ื•ืœื•ื’ืจืžื” ืžื–ื•ื•ื™ื•ืช ืื—ืจื•ืช.
02:49
This also works on videos,
56
169969
1502
ื–ื” ืขื•ื‘ื“ ื’ื ื‘ืกืจื˜ื•ื ื™ื,
02:51
by running the same algorithm on each video frame
57
171495
2852
ืขโ€œื™ ื”ืคืขืœืช ืื•ืชื• ืืœื’ื•ืจื™ืชื ื‘ื›ืœ ื—ึทืœึผื•ึนื ึดื™ืช ื•ื™ื“ืื•
02:54
and generating a moving 3D model.
58
174371
2222
ื•ื™ืฆื™ืจืช ืžื•ื“ืœ ืชืœืช ืžื™ืžื“ื™ ื ืข.
02:57
And here's the same output model from different angles.
59
177538
2772
ื•ื”ื ื” ืื•ืชื• ืคืœื˜ ืžื•ื“ืœ ืžื–ื•ื•ื™ื•ืช ืื—ืจื•ืช.
03:01
It turns out this problem is very challenging,
60
181933
2534
ืžืกืชื‘ืจ ืฉื”ื‘ืขื™ื” ื”ื–ื• ืžืื•ื“ ืžืืชื’ืจืช,
03:04
but the key trick is that we are going to analyze
61
184491
2525
ืื‘ืœ ืžื” ืฉื—ืฉื•ื‘ ืคื” ื”ื•ื, ืฉืื ื—ื ื• ื ื ืชื—
03:07
a large photo collection of the person beforehand.
62
187040
2966
ืœืคื ื™ ื›ืŸ.ื”ืจื‘ื” ืžืื•ื“ ืชืžื•ื ื•ืช ืฉืœ ื”ืื“ื
03:10
For George W. Bush, we can just search on Google,
63
190650
2539
ืชืžื•ื ื•ืช ืฉืœ ื’โ€˜ื•ืจื’โ€™ ื•ื•โ€ฒ ื‘ื•ืฉ, ืื ื—ื ื• ืคืฉื•ื˜ ื ื—ืคืฉ ื‘ื’ื•ื’ืœ,
03:14
and from that, we are able to build an average model,
64
194309
2499
ื•ืžื”ืŸ ืื ื—ื ื• ื™ื›ื•ืœื™ื ืœื‘ื ื•ืช ืžื•ื“ืœ ืกื‘ื™ืจ.
03:16
an iterative, refined model to recover the expression
65
196832
3111
ืžื•ื“ืœ ืžืขื•ื“ืŸ ื—ื•ื–ืจ ื•ื ืฉื ื”, ื›ื“ื™ ืœืงื‘ืœ ืืช ื”ื”ื‘ืขื”
03:19
in fine details, like creases and wrinkles.
66
199967
2336
ื‘ืคืจื˜ื™ ืคืจื˜ื™ื, ื›ืžื• ืงืžื˜ื™ ื”ื‘ืขื” ื•ื’ื™ืœ.
03:23
What's fascinating about this
67
203326
1403
ืžื” ืฉืžืจืชืง ื‘ื–ื”
03:24
is that the photo collection can come from your typical photos.
68
204753
3423
ื”ื•ื ืฉื”ืชืžื•ื ื•ืช ื™ื›ื•ืœื•ืช ืœื”ื’ื™ืข ืžื›ืœ ืื•ืกืฃ ืชืžื•ื ื•ืช ืจื’ื™ืœ.
03:28
It doesn't really matter what expression you're making
69
208200
2603
ื–ื” ืœื ื‘ืืžืช ืžืฉื ื” ืžื”ื™ ื”ื”ื‘ืขื” ืฉืชืฆื™ื’ื•
03:30
or where you took those photos.
70
210827
1885
ืื• ืžืื™ืคื” ืœืงื—ืชื ืืช ื”ืชืžื•ื ื•ืช.
03:32
What matters is that there are a lot of them.
71
212736
2400
ืžื” ืฉื—ืฉื•ื‘ ื–ื” ืฉื™ื”ื™ื• ื”ืจื‘ื” ืžื”ืŸ.
03:35
And we are still missing color here,
72
215160
1736
ื•ืขื“ื™ื™ืŸ ื—ืกืจ ืœื ื• ืฆื‘ืข ื›ืืŸ,
03:36
so next, we develop a new blending technique
73
216920
2348
ืื– ืœืื—ืจ ืžื›ืŸ ืคื™ืชื—ื ื• ื˜ื›ื ื™ืงื” ืœืขืจื‘ื•ื‘ ืฆื‘ืขื™ื
03:39
that improves upon a single averaging method
74
219292
2836
ืฉืžืฉืชืคืจืช ืชื•ืš ื›ื“ื™ ืฉื™ืžื•ืฉ ืจื’ื™ืœ
03:42
and produces sharp facial textures and colors.
75
222152
2818
ื•ื™ื•ืฆืจืช ื‘ืžื“ื•ื™ืง ืชื•ื•ื™ ืคื ื™ื, ืžืจืงื ื•ืฆื‘ืขื™ื.
03:45
And this can be done for any expression.
76
225779
2771
ื•ื–ื” ื™ื›ื•ืœ ืœื”ื™ืขืฉื•ืช ืœื›ืœ ื”ื‘ืขื”.
03:49
Now we have a control of a model of a person,
77
229485
2499
ืขื›ืฉื™ื•, ื›ืฉื™ืฉ ืœื ื• ืฉืœื™ื˜ื” ืขืœ ืžื•ื“ืœ ืฉืœ ืื“ื
ื•ื”ืื•ืคืŸ ื‘ื• ื”ื•ื ื ืฉืœื˜ ื›ืขืช ื”ื•ื ืขโ€œื™ ืฉื™ืžื•ืฉ ื‘ืจืฆืฃ ืชืžื•ื ื•ืช.
03:52
and the way it's controlled now is by a sequence of static photos.
78
232008
3795
03:55
Notice how the wrinkles come and go, depending on the expression.
79
235827
3126
ืฉื™ืžื• ืœื‘ ืื™ืš ื”ืงืžื˜ื™ื ืžื•ืคื™ืขื™ื ื•ื ืขืœืžื™ื ื‘ื”ืชืื ืœื”ื‘ืขื”.
04:00
We can also use a video to drive the model.
80
240109
2746
ืื ื—ื ื• ื™ื›ื•ืœื™ื ื’ื ืœื”ืฉืชืžืฉ ื‘ื•ื™ื“ืื• ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ื“ื’ื.
04:02
(Video) Daniel Craig: Right, but somehow,
81
242879
2593
(ื‘ื•ื™ื“ืื•) ื“ื ื™ืืœ ืงืจื™ื™ื’: ื ื›ื•ืŸ, ืื‘ืœ ืื™ื›ืฉื”ื•,
04:05
we've managed to attract some more amazing people.
82
245496
3771
ื”ืฆืœื—ื ื• ืœืžืฉื•ืš ืขื•ื“ ื›ืžื” ืื ืฉื™ื ืžื“ื”ื™ืžื™ื.
04:10
SS: And here's another fun demo.
83
250021
1642
ืก.ืก: ื•ื”ื ื” ืขื•ื“ ื”ื“ื’ืžื” ื›ื™ืคื™ืช.
04:11
So what you see here are controllable models
84
251687
2246
ืื– ืžื” ืฉืืชื ืจื•ืื™ื ื›ืืŸ ื”ื ื“ื’ืžื™ื ื ื™ืชื ื™ื ืœืฉืœื™ื˜ื”
04:13
of people I built from their internet photos.
85
253957
2444
ืฉืœ ืื ืฉื™ื ืฉื™ืฆืจืชื™ ืžืชืžื•ื ื•ืช ืฉืœื”ื ื‘ืื™ื ื˜ืจื ื˜.
04:16
Now, if you transfer the motion from the input video,
86
256425
2904
ืื ืชืขื‘ื™ืจื• ืืช ื”ืชื ื•ืขื” ืžืงืœื˜ ื”ื•ื™ื“ืื•,
04:19
we can actually drive the entire party.
87
259353
2152
ืื ื—ื ื• ื™ื›ื•ืœื™ื ืœื”ืคืขื™ืœ ืืช ื›ืœ ื”ื—ื‘ื•ืจื”.
04:21
George W. Bush: It's a difficult bill to pass,
88
261529
2172
ื’'ื•ืจื’' ื•ื•. ื‘ื•ืฉ: ื–ื”ื• ื—ื•ืง ืงืฉื” ืœื”ืขื‘ืจื”,
04:23
because there's a lot of moving parts,
89
263725
2303
ื›ื™ ื™ืฉื ื ื”ืจื‘ื” ื—ืœืงื™ื ื ืขื™ื,
04:26
and the legislative processes can be ugly.
90
266052
5231
ื•ื”ืชื”ืœื™ื›ื™ื ื”ื—ื•ืงืชื™ื™ื ื™ื›ื•ืœื™ื ืœื”ื™ื•ืช ืžื›ื•ืขืจื™ื.
04:31
(Applause)
91
271307
1630
(ืžื—ื™ืื•ืช ื›ืคื™ื™ื)
04:32
SS: So coming back a little bit,
92
272961
1837
ืก.ืก: ืื ื ื—ื–ื•ืจ ืžืขื˜ ืื—ื•ืจื”,
04:34
our ultimate goal, rather, is to capture their mannerisms
93
274822
3191
ื”ืžื˜ืจื” ื”ืกื•ืคื™ืช ืฉืœื ื•, ื”ื™ื ื‘ืขืฆื ืœืชืคื•ืก ืืช ื”ื”ื‘ืขื•ืช ืฉืœื”ื
04:38
or the unique way each of these people talks and smiles.
94
278037
3045
ืื• ืืช ื”ื“ืจืš ื”ื™ื™ื—ื•ื“ื™ืช ื‘ื” ื›ืœ ืื—ื“ ืžื”ืื ืฉื™ื ื”ืืœื” ืžื“ื‘ืจ ื•ืžื—ื™ื™ืš.
04:41
So to do that, can we actually teach the computer
95
281106
2313
ื›ื“ื™ ืœืขืฉื•ืช ื–ืืช, ื”ืื ืื ื—ื ื• ื™ื›ื•ืœื™ื ืœืœืžื“ ืืช ื”ืžื—ืฉื‘
04:43
to imitate the way someone talks
96
283443
2222
ืœื—ืงื•ืช ืืช ื”ืื•ืคืŸ ื‘ื• ืžื™ืฉื”ื• ืžื“ื‘ืจ
04:45
by only showing it video footage of the person?
97
285689
2420
ืจืง ื‘ืืžืฆืขื•ืช ื”ืฆื’ื” ืฉืœ ืฆื™ืœื•ืžื™ ื•ื™ื“ืื• ืฉืœ ืื•ืชื• ืื“ื?
04:48
And what I did exactly was, I let a computer watch
98
288898
2577
ื•ืžื” ืฉืขืฉื™ืชื™ ื‘ื“ื™ื•ืง, ื ืชืชื™ ืœืžื—ืฉื‘ ืœืฆืคื•ืช
04:51
14 hours of pure Barack Obama giving addresses.
99
291499
3277
ื‘ืžืฉืš 14 ืฉืขื•ืช ืฉืœ ื ืื•ืžื™ื ืฉืœ ื‘ืจืง ืื•ื‘ืžื”.
04:55
And here's what we can produce given only his audio.
100
295443
3516
ื•ื”ื ื” ืžื” ืฉืื ื—ื ื• ื™ื›ื•ืœื™ื ืœื”ืคื™ืง ืจืง ื‘ืืžืฆืขื•ืช ื”ืื•ื“ื™ื•.
04:58
(Video) BO: The results are clear.
101
298983
1777
(ื•ื™ื“ืื•) ื‘.ื: ื”ืชื•ืฆืื•ืช ื‘ืจื•ืจื•ืช.
05:00
America's businesses have created 14.5 million new jobs
102
300784
4349
ื”ืขืกืงื™ื ื‘ืืžืจื™ืงื” ื™ืฆืจื• 14.5 ืžื™ืœื™ื•ืŸ ืžืงื•ืžื•ืช ืขื‘ื•ื“ื” ื—ื“ืฉื™ื
05:05
over 75 straight months.
103
305157
2774
ื‘ืžืฉืš 75 ื—ื•ื“ืฉื™ื ืจืฆื•ืคื™ื.
05:07
SS: So what's being synthesized here is only the mouth region,
104
307955
2905
ืก.ืก: ืื– ื”ื“ื‘ืจ ื”ื™ื—ื™ื“ ืฉืžืกื•ื ืชื– ื›ืืŸ ื–ื” ืื–ื•ืจ ื”ืคื”,
05:10
and here's how we do it.
105
310884
1540
ื•ื›ืš ืื ื—ื ื• ืขื•ืฉื™ื ืืช ื–ื”.
05:12
Our pipeline uses a neural network
106
312764
1826
ื”ืžืกืœื•ืœ ืฉืœื ื• ืžืฉืชืžืฉ ื‘ืจืฉืช ื ื•ื™ืจื•ื ื™ืช
05:14
to convert and input audio into these mouth points.
107
314614
2936
ื›ื“ื™ ืœื”ืžื™ืจ ื•ืœืงืœื•ื˜ ื™ื—ื™ื“ื•ืช ืื•ื“ื™ื• ืœื ืงื•ื“ื•ืช ื”ืคื” ื”ืืœื”.
05:18
(Video) BO: We get it through our job or through Medicare or Medicaid.
108
318547
4225
(ื•ื™ื“ืื•) ื‘.ื: ืื ื—ื ื• ืžืฉื™ื’ื™ื ืืช ื–ื” ื‘ืืžืฆืขื•ืช ื”ืขื‘ื•ื“ื” ืฉืœื ื• ืื• ื‘ืืžืฆืขื•ืช ืžื“ื™ืงื™ื™ืจ ืื• ืžื“ื™ืงื™ื™ื“.
05:22
SS: Then we synthesize the texture, enhance details and teeth,
109
322796
3420
ืก.ืก: ืื– ืื ื—ื ื• ืžืกื ืชื–ื™ื ืืช ื”ื˜ืงืกื˜ื•ืจื”, ืžืขืฆื™ืžื™ื ืคืจื˜ื™ื ื•ืฉื™ื ื™ื™ื,
05:26
and blend it into the head and background from a source video.
110
326240
3074
ื•ืžืฉืœื‘ื™ื ื”ื›ืœ ืœืชื•ืš ื”ืจืืฉ ื•ืœืจืงืข ืžื•ื™ื“ืื• ื”ืžืงื•ืจ.
05:29
(Video) BO: Women can get free checkups,
111
329338
1905
(ื•ื™ื“ืื•) ื‘.ื: ื ืฉื™ื ื™ื›ื•ืœื•ืช ืœืงื‘ืœ ื‘ื“ื™ืงื•ืช ื—ื™ื ื,
05:31
and you can't get charged more just for being a woman.
112
331267
2968
ื•ืื™ ืืคืฉืจ ืœื’ื‘ื•ืช ืžืžืš ื™ื•ืชืจ, ืคืฉื•ื˜ ื›ื™ ืืช ืื™ืฉื”.
05:34
Young people can stay on a parent's plan until they turn 26.
113
334973
3306
ืื ืฉื™ื ืฆืขื™ืจื™ื ื™ื›ื•ืœื™ื ืœื”ื™ืฉืืจ ืขืœ ื”ืชื›ื ื™ืช ืฉืœ ื”ื”ื•ืจื” ืขื“ ืœื’ื™ืœ 26.
05:39
SS: I think these results seem very realistic and intriguing,
114
339267
2952
ืก.ืก: ืื ื™ ื—ื•ืฉื‘ ืฉื”ืชื•ืฆืื•ืช ื”ืืœื” ื ืจืื•ืช ืžืื•ื“ ืืžื™ืชื™ื•ืช ื•ืžืืชื’ืจื•ืช,
05:42
but at the same time frightening, even to me.
115
342243
3173
ืื‘ืœ ื‘ืื•ืชื• ื–ืžืŸ ื”ืŸ ืžืคื—ื™ื“ื•ืช, ืืคื™ืœื• ืื•ืชื™.
05:45
Our goal was to build an accurate model of a person, not to misrepresent them.
116
345440
4015
ื”ืžื˜ืจื” ืฉืœื ื• ื”ื™ืชื” ืœื‘ื ื•ืช ืžื•ื“ืœ ืžื“ื•ื™ืง ืฉืœ ืื“ื ืœื ืœื”ืฆื™ื’ ืื•ืชื• ื‘ืื•ืคืŸ ืฉื’ื•ื™.
05:49
But one thing that concerns me is its potential for misuse.
117
349956
3111
ืื‘ืœ ืื—ื“ ื”ื“ื‘ืจื™ื ืฉืžื“ืื™ื’ื™ื ืื•ืชื™ ื–ื” ื”ืคื•ื˜ื ืฆื™ืืœ ืœืฉื™ืžื•ืฉ ืœืจืขื”.
05:53
People have been thinking about this problem for a long time,
118
353958
2971
ืื ืฉื™ื ื—ื•ืฉื‘ื™ื ืขืœ ื”ื‘ืขื™ื” ื”ื–ื• ื›ื‘ืจ ื–ืžืŸ ืจื‘.
05:56
since the days when Photoshop first hit the market.
119
356953
2381
ืžืื– ื”ื™ืžื™ื ื‘ื”ื ืคื•ื˜ื•ืฉื•ืค ืจืง ื”ื’ื™ืข ืœืฉื•ืง.
05:59
As a researcher, I'm also working on countermeasure technology,
120
359862
3801
ื›ื—ื•ืงืจ, ืื ื™ ืขื•ื‘ื“ ื’ื ืขืœ ื˜ื›ื ื•ืœื•ื’ื™ื” ืžืคืงื—ืช,
06:03
and I'm part of an ongoing effort at AI Foundation,
121
363687
2942
ื•ืื ื™ ื—ืœืง ืžืคืจื•ื™ืงื˜ ืžืชืžืฉืš ื‘-ืื™.ืื™ื™ ืคืื•ื ื“ื™ื™ืฉืŸ,
06:06
which uses a combination of machine learning and human moderators
122
366653
3397
ืฉืžืฉืชืžืฉืช ื‘ืฉื™ืœื•ื‘ ืฉืœ ืœืžื™ื“ืช ืžื›ื•ื ื” ื•ืคื™ืงื•ื— ืื ื•ืฉื™
06:10
to detect fake images and videos,
123
370074
2144
ื›ื“ื™ ืœื’ืœื•ืช ืกืจื˜ื•ื ื™ ื•ื™ื“ืื• ื•ืชืžื•ื ื•ืช ืžื–ื•ื™ืคื•ืช,
06:12
fighting against my own work.
124
372242
1514
ื‘ืœื—ื™ืžื” ื ื’ื“ ื”ืขื‘ื•ื“ื” ืฉืœื™ ืžืžืฉ.
06:14
And one of the tools we plan to release is called Reality Defender,
125
374675
3190
ื•ืื—ื“ ื”ื›ืœื™ื ืฉืื ื—ื ื• ืžืชื›ื ื ื™ื ืœื”ื•ืฆื™ื ื ืงืจื: โ€œืžื’ืŸ ืžืฆื™ืื•ืชโ€œ,
06:17
which is a web-browser plug-in that can flag potentially fake content
126
377889
4039
ืฉื”ื•ื ื—ื™ื‘ื•ืจ ืจื›ื™ื‘ ืœื“ืคื“ืคืŸ ืฉื™ื›ื•ืœ ืœื”ืชืจื™ืข ืขืœ ืืคืฉืจื•ืช ืฉืœ ืชื•ื›ืŸ ืžื–ื•ื™ืฃ
06:21
automatically, right in the browser.
127
381952
2533
ื‘ืื•ืคืŸ ืื•ื˜ื•ืžื˜ื™, ืžืžืฉ ื‘ื“ืคื“ืคืŸ.
06:24
(Applause)
128
384509
4228
(ืžื—ื™ืื•ืช ื›ืคื™ื™ื)
06:28
Despite all this, though,
129
388761
1453
ืœืžืจื•ืช ื›ืœ ื–ื”,
06:30
fake videos could do a lot of damage,
130
390238
1840
ืงื˜ืขื™ ื•ื™ื“ืื• ืžื–ื•ื™ืคื™ื ื™ื›ื•ืœื™ื ืœื’ืจื•ื ื ื–ืง ืจื‘,
06:32
even before anyone has a chance to verify,
131
392102
3294
ืืคื™ืœื• ืœืคื ื™ ืฉื™ืฉื ื” ืœืžื™ืฉื”ื• ืืคืฉืจื•ืช ืœืืžืช ืื•ืชื,
06:35
so it's very important that we make everyone aware
132
395420
2722
ืื– ืžืžืฉ ื—ืฉื•ื‘ ืฉื›ื•ืœื ื™ื”ื™ื• ืžื•ื“ืขื™ื
06:38
of what's currently possible
133
398166
2007
ืœืžื” ืฉืืคืฉืจื™ ืขื›ืฉื™ื•
06:40
so we can have the right assumption and be critical about what we see.
134
400197
3369
ื›ืš ืฉื™ื”ื™ื• ืœื ื• ืืช ื”ื”ื ื—ื•ืช ื”ื ื›ื•ื ื•ืช ื•ื ื”ื™ื” ื‘ื™ืงื•ืจืชื™ื™ื ื›ืœืคื™ ืžื” ืฉืื ื—ื ื• ืจื•ืื™ื.
06:44
There's still a long way to go before we can fully model individual people
135
404423
5007
ื™ืฉื ื” ืขื“ื™ื™ืŸ ื“ืจืš ืืจื•ื›ื” ืœืคื ื™ ืฉื ื•ื›ืœ ืœื“ื’ื•ื ืื ืฉื™ื ื‘ืื•ืคืŸ ืžืœื
06:49
and before we can ensure the safety of this technology.
136
409454
2786
ื•ืœืคื ื™ ืฉื ื•ื›ืœ ืœื•ื•ื“ื ืฉื”ื˜ื›ื ื•ืœื•ื’ื™ื” ื”ื–ื• ื‘ื˜ื•ื—ื”.
06:53
But I'm excited and hopeful,
137
413097
1587
ืื‘ืœ ืื ื™ ื ืจื’ืฉ ื•ืžืœื ืชืงื•ื•ื”,
06:54
because if we use it right and carefully,
138
414708
3539
ื›ื™ ืื ื ืฉืชืžืฉ ื‘ื” ื‘ืื•ืคืŸ ื ื›ื•ืŸ ื•ื‘ื–ื”ื™ืจื•ืช,
06:58
this tool can allow any individual's positive impact on the world
139
418271
4309
ื”ื›ืœื™ ื”ื–ื” ื™ื•ื›ืœ ืœืืคืฉืจ ืœื”ื’ื‘ื™ืจ ืืช ื”ื”ืฉืคืขื” ื”ื—ื™ื•ื‘ื™ืช ืฉื™ืฉ ืœื›ืœ ืื—ื“ ื‘ืขื•ืœื.
07:02
to be massively scaled
140
422604
2190
ื‘ืงื ื” ืžื™ื“ื” ื’ื“ื•ืœ
07:04
and really help shape our future the way we want it to be.
141
424818
2742
ื•ื™ื•ื›ืœ ืœืขื–ื•ืจ ืœื ื• ืœืขืฆื‘ ืืช ืขืชื™ื“ื ื• ื‘ื“ืจืš ื‘ื” ื ืจืฆื”.
07:07
Thank you.
142
427584
1151
ืชื•ื“ื” ืจื‘ื”.
07:08
(Applause)
143
428759
5090
(ื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