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

1,274,992 views

2018-07-25 ใƒป TED


New videos

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

1,274,992 views ใƒป 2018-07-25

TED


ไธ‹ใฎ่‹ฑ่ชžๅญ—ๅน•ใ‚’ใƒ€ใƒ–ใƒซใ‚ฏใƒชใƒƒใ‚ฏใ™ใ‚‹ใจๅ‹•็”ปใ‚’ๅ†็”Ÿใงใใพใ™ใ€‚

็ฟป่จณ: Yasushi Aoki ๆ กๆญฃ: Claire Ghyselen
00:12
Look at these images.
0
12876
1151
ใ“ใฎๆ˜ ๅƒใ‚’่ฆ‹ใฆใใ ใ•ใ„
00:14
Now, tell me which Obama here is real.
1
14051
2635
ๆœฌ็‰ฉใฎใ‚ชใƒใƒžๅคง็ตฑ้ ˜ใฏ ใฉใ‚Œใงใ—ใ‚‡ใ†๏ผŸ
00:16
(Video) Barack Obama: To help families refinance their homes,
2
16710
2861
(ใƒใƒฉใ‚ฏใƒปใ‚ชใƒใƒž) ไฝๅฎ…ใƒญใƒผใƒณใ‚’ ๅ€Ÿใ‚Šๆ›ใˆใ‚‹ๅฎถๅบญใ‚’ๅŠฉใ‘ใ‚‹ใ“ใจ
00:19
to invest in things like high-tech manufacturing,
3
19595
2647
ใƒใ‚คใƒ†ใ‚ฏ่ฃฝ้€ ๆฅญ ใ‚ฏใƒชใƒผใƒณใ‚จใƒใƒซใ‚ฎใƒผ
00:22
clean energy
4
22266
1159
ใ‚คใƒณใƒ•ใƒฉใจใ„ใฃใŸใ‚‚ใฎใซ ๆŠ•่ณ‡ใ™ใ‚‹ใ“ใจใง
00:23
and the infrastructure that creates good new jobs.
5
23449
2779
่‰ฏใ„ไป•ไบ‹ใŒๅ‰ตๅ‡บใ•ใ‚Œใพใ™
00:26
Supasorn Suwajanakorn: Anyone?
6
26647
1484
(่ฌ›ๆผ”่€…) ๅˆ†ใ‹ใ‚Šใพใ™ใ‹๏ผŸ
00:28
The answer is none of them.
7
28155
1874
็ญ”ใˆใฏ ๅ…จ้ƒจใƒ‹ใ‚ป่€…ใงใ™
00:30
(Laughter)
8
30053
1114
(็ฌ‘)
00:31
None of these is actually real.
9
31191
1786
ใฉใ‚Œ๏ผ‘ใคๆœฌ็‰ฉใงใฏใ‚ใ‚Šใพใ›ใ‚“
00:33
So let me tell you how we got here.
10
33001
1840
ใ“ใ‚Œใพใงใฎ้“ใฎใ‚Šใซใคใ„ใฆ ใŠ่ฉฑใ—ใ—ใพใ—ใ‚‡ใ†
00:35
My inspiration for this work
11
35940
1578
ใ“ใฎ็ ”็ฉถใ‚’ใ™ใ‚‹ ใƒ’ใƒณใƒˆใซใชใฃใŸใฎใฏ
00:37
was a project meant to preserve our last chance for learning about the Holocaust
12
37542
5411
ใƒ›ใƒญใ‚ณใƒผใ‚นใƒˆใฎ็”Ÿๅญ˜่€…ใŸใกใ‹ใ‚‰ ๅญฆใถๆœ€ๅพŒใฎๆฉŸไผšใ‚’
ไฟๅญ˜ใ—ใ‚ˆใ†ใจใ„ใ† ใƒ—ใƒญใ‚ธใ‚งใ‚ฏใƒˆใงใ—ใŸ
00:42
from the survivors.
13
42977
1768
00:44
It's called New Dimensions in Testimony,
14
44769
2627
ใ€Œ่จผ่จ€ใฎๆ–ฐๅฑ€้ขใ€(New Dimensions in Testimony) ใจใ„ใ†ๅๅ‰ใง
00:47
and it allows you to have interactive conversations
15
47420
3126
ใƒ›ใƒญใ‚ณใƒผใ‚นใƒˆ็”Ÿๅญ˜่€…ใฎ ใƒ›ใƒญใ‚ฐใƒฉใƒ ใจ
00:50
with a hologram of a real Holocaust survivor.
16
50570
2556
ๅฏพ่ฉฑใ™ใ‚‹ใ“ใจใŒใงใใพใ™
00:53
(Video) Man: How did you survive the Holocaust?
17
53793
1966
(็”ท) ใฉใ†ใ‚„ใฃใฆใƒ›ใƒญใ‚ณใƒผใ‚นใƒˆใ‚’ ็”ŸใๅปถใณใŸใ‚“ใงใ™ใ‹๏ผŸ
00:55
(Video) Hologram: How did I survive?
18
55783
1668
(ใƒ›ใƒญใ‚ฐใƒฉใƒ ) ใฉใ†ใ‚„ใฃใฆ็”ŸใๅปถใณใŸใ‹๏ผŸ
00:57
I survived,
19
57912
1807
็งใŒ็”Ÿใๆฎ‹ใ‚ŒใŸใฎใฏโ€”
01:00
I believe,
20
60419
1527
็ฅžใŒ่ฆ‹ๅฎˆใฃใฆใ„ใฆ ใใ‚ŒใŸใ‹ใ‚‰ใ ใจ
01:01
because providence watched over me.
21
61970
3023
็งใฏๆ€ใฃใฆใ„ใพใ™
01:05
SS: Turns out these answers were prerecorded in a studio.
22
65573
3454
(่ฌ›ๆผ”่€…) ็ญ”ใˆใฏใ‚นใ‚ฟใ‚ธใ‚ชใง ใ‚ใ‚‰ใ‹ใ˜ใ‚้Œฒ็”ปใ•ใ‚ŒใŸใ‚‚ใฎใงใ™ใŒ
01:09
Yet the effect is astounding.
23
69051
2452
ใใฎๅŠนๆžœใฏๅŠ‡็š„ใงใ™
01:11
You feel so connected to his story and to him as a person.
24
71527
3619
ใใฎไบบใฎ่ฉฑใ‚„ ใใฎไบบ่‡ช่บซใซๅฏพใ™ใ‚‹ ๅผทใ„็ตใณไป˜ใใ‚’ๆ„Ÿใ˜ใพใ™
01:16
I think there's something special about human interaction
25
76011
3301
ไบบ้–“ๅŒๅฃซใฎใ‚„ใ‚Šๅ–ใ‚Šใซใฏ
็‰นๅˆฅใชๅŠ›ใŒใ‚ใ‚‹ใฎใงใ—ใ‚‡ใ†
01:19
that makes it much more profound
26
79336
2757
ๆœฌใ‚„่ฌ›็พฉใ‚„ๆ˜ ็”ป ใชใฉใ‚ˆใ‚Šใ‚‚ใšใฃใจ
01:22
and personal
27
82117
2198
ๆทฑใๅ€‹ไบบ็š„ใชไฝ“้จ“ใ‚’ ไธŽใˆใฆใใ‚Œใพใ™
01:24
than what books or lectures or movies could ever teach us.
28
84339
3485
01:28
So I saw this and began to wonder,
29
88267
2425
ใใ‚Œใง็งใฏ ๆ€ใ†ใ‚ˆใ†ใซใชใ‚Šใพใ—ใŸ
01:30
can we create a model like this for anyone?
30
90716
2810
ใ“ใ†ใ„ใ†ใƒขใƒ‡ใƒซใ‚’ ่ชฐใซๅฏพใ—ใฆใ‚‚ไฝœใ‚ŒใŸใ‚‰๏ผŸ
01:33
A model that looks, talks and acts just like them?
31
93550
2975
ใใฎไบบ่‡ช่บซใฎใ‚ˆใ†ใซ่ฆ‹ใˆ ่ชžใ‚Š ๆŒฏใ‚‹่ˆžใ†ใƒขใƒ‡ใƒซใงใ™
01:37
So I set out to see if this could be done
32
97573
2007
ใใ‚ŒใŒๅฏ่ƒฝใ‹ๆคœ่จŽใ‚’ๅง‹ใ‚
01:39
and eventually came up with a new solution
33
99604
2310
ๆ—ขๅญ˜ใฎๅ†™็œŸใ‚„ๆ˜ ๅƒใ ใ‘ใ‹ใ‚‰
01:41
that can build a model of a person using nothing but these:
34
101938
3220
ใใฎไบบ็‰ฉใฎใƒขใƒ‡ใƒซใ‚’ไฝœใ‚‹ๆ–นๆณ•ใ‚’ ่€ƒๆกˆใ—ใพใ—ใŸ
01:45
existing photos and videos of a person.
35
105747
2214
ใใฎไบบ็‰ฉใฎใƒขใƒ‡ใƒซใ‚’ไฝœใ‚‹ๆ–นๆณ•ใ‚’ ่€ƒๆกˆใ—ใพใ—ใŸ
01:48
If you can leverage this kind of passive information,
36
108701
2617
ใ“ใ†ใ„ใ† ใใฎ่พบใซใ‚ใ‚‹ ๅ†™็œŸใ‚„ๆ˜ ๅƒใจใ„ใฃใŸ
01:51
just photos and video that are out there,
37
111342
2007
ใ‚ใ‚Šใ‚ใ‚ใ›ใฎ็ด ๆ ใ ใ‘ใงใ‚ˆใ„ใชใ‚‰
01:53
that's the key to scaling to anyone.
38
113373
2056
่ชฐใซๅฏพใ—ใฆใ‚‚ใƒขใƒ‡ใƒซใ‚’ ไฝœใ‚Œใ‚‹ใ‚ˆใ†ใซใชใ‚Šใพใ™
01:56
By the way, here's Richard Feynman,
39
116119
1777
ใกใชใฟใซ ใ“ใฎไบบ็‰ฉใฏ ใƒชใƒใƒฃใƒผใƒ‰ใƒปใƒ•ใ‚กใ‚คใƒณใƒžใƒณใง
01:57
who in addition to being a Nobel Prize winner in physics
40
117920
3413
ใƒŽใƒผใƒ™ใƒซ็‰ฉ็†ๅญฆ่ณžใฎ ๅ—่ณž่€…ใงใ‚ใ‚‹ใฎใฟใชใ‚‰ใš
02:01
was also known as a legendary teacher.
41
121357
2453
ๅ„ชใ‚ŒใŸๆ•™ๅธซใจใ—ใฆ ใ‚ˆใ็Ÿฅใ‚‰ใ‚Œใฆใ„ใพใ—ใŸ
02:05
Wouldn't it be great if we could bring him back
42
125080
2198
ใ‚‚ใ—ใƒ•ใ‚กใ‚คใƒณใƒžใƒณใ‚’่˜‡ใ‚‰ใ›
02:07
to give his lectures and inspire millions of kids,
43
127302
3265
ไฝ•็™พไธ‡ใจใ„ใ†่‹ฅ่€…ใซ่ฉฑใ‚’ใ—ใฆ ๅˆบๆฟ€ใ‚’ไธŽใˆใฆใ‚‚ใ‚‰ใ„
02:10
perhaps not just in English but in any language?
44
130591
2992
ใ•ใ‚‰ใซใฏไป–ใฎ่จ€่ชžใงใ‚‚่ชžใ‚‰ใ›ใ‚‰ใ‚ŒใŸใชใ‚‰ ็ด ๆ™ดใ‚‰ใ—ใ„ใงใ—ใ‚‡ใ†
02:14
Or if you could ask our grandparents for advice and hear those comforting words
45
134441
4602
ใ‚ใ‚‹ใ„ใฏ ใ‚‚ใ†ใ“ใฎไธ–ใ‚’ๅŽปใฃใฆใ—ใพใฃใŸ ใŠใ˜ใ„ใ•ใ‚“ ใŠใฐใ‚ใ•ใ‚“ใซ
ใ‚ขใƒ‰ใƒใ‚คใ‚นใ‚„ๅฟƒๆธฉใพใ‚‹่จ€่‘‰ใ‚’ ใ‹ใ‘ใฆใ‚‚ใ‚‰ใˆใŸใชใ‚‰
02:19
even if they're no longer with us?
46
139067
1770
02:21
Or maybe using this tool, book authors, alive or not,
47
141683
3396
ใพใŸใ“ใฎใƒ„ใƒผใƒซใ‚’ไฝฟใˆใฐ ๅญ˜ๅ‘ฝใ‹ๅฆใ‹ใซใ‹ใ‹ใ‚ใ‚‰ใš
02:25
could read aloud all of their books for anyone interested.
48
145103
2937
่‘—่€…่‡ช่บซใซๆœฌใฎๆœ—่ชญใ‚’ ใ—ใฆใ‚‚ใ‚‰ใ†ใ“ใจใ‚‚ใงใใ‚‹ใงใ—ใ‚‡ใ†
02:29
The creative possibilities here are endless,
49
149199
2437
ใ“ใ‚ŒใŒๆŒใคๅ‰ต้€ ็š„ๅฏ่ƒฝๆ€งใฏ ้™ใ‚ŠใŒใชใ
02:31
and to me, that's very exciting.
50
151660
1713
ใ™ใ”ใใƒฏใ‚ฏใƒฏใ‚ฏใ•ใ›ใ‚‰ใ‚Œใพใ™
02:34
And here's how it's working so far.
51
154595
2002
ใใฎไป•็ต„ใฟใ‚’ ใŠ่ฉฑใ—ใ—ใพใ—ใ‚‡ใ†
02:36
First, we introduce a new technique
52
156621
1667
ใพใš ้ก”ใฎ็ฒพ็ดฐใช ๏ผ“ๆฌกๅ…ƒใƒขใƒ‡ใƒซใ‚’
02:38
that can reconstruct a high-detailed 3D face model from any image
53
158312
4572
3Dใ‚นใ‚ญใƒฃใƒณใƒ‡ใƒผใ‚ฟใชใ—ใซ
ไปปๆ„ใฎ็”ปๅƒใ‹ใ‚‰ไฝœใ‚Œใ‚‹ ๆ‰‹ๆณ•ใ‚’้–‹็™บใ—ใพใ—ใŸ
02:42
without ever 3D-scanning the person.
54
162908
2119
02:45
And here's the same output model from different views.
55
165890
2642
ใ“ใ‚ŒใฏๅŒใ˜ใƒขใƒ‡ใƒซใ‚’ ๅˆฅใฎ่ฆ–็‚นใ‹ใ‚‰่ฆ‹ใŸใ‚‚ใฎใงใ™
02:49
This also works on videos,
56
169969
1502
ใ“ใฎๆŠ€่ก“ใฏๆ˜ ๅƒใซใ‚‚ไฝฟใˆใพใ™
02:51
by running the same algorithm on each video frame
57
171495
2852
ๆ˜ ๅƒใฎๅ„ใƒ•ใƒฌใƒผใƒ ใซ ๅŒใ˜ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ ใ‚’้ฉ็”จใ—
02:54
and generating a moving 3D model.
58
174371
2222
ๅ‹•ใใฎใ‚ใ‚‹๏ผ“ๆฌกๅ…ƒใƒขใƒ‡ใƒซใ‚’ ็”Ÿๆˆใ—ใพใ™
02:57
And here's the same output model from different angles.
59
177538
2772
ใ“ใกใ‚‰ใฏๅŒใ˜ใƒขใƒ‡ใƒซใ‚’ ้•ใ†่ง’ๅบฆใ‹ใ‚‰่ฆ‹ใŸใ‚‚ใฎใงใ™
03:01
It turns out this problem is very challenging,
60
181933
2534
ใ“ใฎๅ•้กŒใฏ ใจใฆใ‚‚้›ฃใ—ใ„ใฎใงใ™ใŒ
03:04
but the key trick is that we are going to analyze
61
184491
2525
ใ‚ใ‚‰ใ‹ใ˜ใ‚ ใใฎไบบ็‰ฉใฎ ๅคง้‡ใฎๅ†™็œŸใ‚’
03:07
a large photo collection of the person beforehand.
62
187040
2966
่งฃๆžใ™ใ‚‹ใ“ใจใŒ ้ตใซใชใ‚Šใพใ™
03:10
For George W. Bush, we can just search on Google,
63
190650
2539
ใ‚ธใƒงใƒผใ‚ธใƒปWใƒปใƒ–ใƒƒใ‚ทใƒฅใชใ‚‰ Googleใง็”ปๅƒๆคœ็ดขใ™ใ‚‹ใ ใ‘ใงใ‚ˆใ
03:14
and from that, we are able to build an average model,
64
194309
2499
ใใ“ใ‹ใ‚‰ๅนณๅ‡ใƒขใƒ‡ใƒซใ‚’ ไฝœใ‚‹ใ“ใจใŒใงใ
03:16
an iterative, refined model to recover the expression
65
196832
3111
ๆฎต้šŽ็š„ใซใƒขใƒ‡ใƒซใ‚’ ๆ”นๅ–„ใ—ใฆใ„ใฃใฆ
03:19
in fine details, like creases and wrinkles.
66
199967
2336
็šบใฎใ‚ˆใ†ใช ่กจๆƒ…ใฎ ็ดฐ้ƒจใ‚’ๅ†็พใ—ใพใ™
03:23
What's fascinating about this
67
203326
1403
ใ“ใ‚Œใฎใ„ใ„ใจใ“ใ‚ใฏ
03:24
is that the photo collection can come from your typical photos.
68
204753
3423
ๅ†™็œŸใฏ ใ”ใๆ™ฎ้€šใฎใ‚‚ใฎใง ใ‚ˆใ„ใจใ„ใ†ใ“ใจใงใ™
03:28
It doesn't really matter what expression you're making
69
208200
2603
ใฉใ†ใ„ใ†่กจๆƒ…ใ‹ใจใ‹ ใฉใ“ใงๆ’ฎใ‚‰ใ‚ŒใŸใ‹ใจใ‹ใฏ
03:30
or where you took those photos.
70
210827
1885
ใ‚ใพใ‚Šๅ•้กŒใงใฏใ‚ใ‚Šใพใ›ใ‚“
03:32
What matters is that there are a lot of them.
71
212736
2400
ๅคงไบ‹ใชใฎใฏๅ†™็œŸใŒ ใŸใใ•ใ‚“ใ‚ใ‚‹ใจใ„ใ†ใ“ใจใงใ™
03:35
And we are still missing color here,
72
215160
1736
ใพใ ่‰ฒใŒไป˜ใ„ใฆใ„ใชใ„ใฎใง
03:36
so next, we develop a new blending technique
73
216920
2348
ๆฌกใซๆ–ฐใ—ใ„ใƒ–ใƒฌใƒณใƒ‡ใ‚ฃใƒณใ‚ฐๆŠ€ๆณ•ใ‚’้–‹็™บใ—
03:39
that improves upon a single averaging method
74
219292
2836
ๅนณๅ‡ใƒขใƒ‡ใƒซใ‚’ๆ”น่‰ฏใ—ใฆ
ใใฃใใ‚Šใจใ—ใŸ้ก”ใฎ ่ณชๆ„Ÿใ‚„่‰ฒใ‚’ไป˜ใ‘ใพใ™
03:42
and produces sharp facial textures and colors.
75
222152
2818
03:45
And this can be done for any expression.
76
225779
2771
ใ“ใ‚Œใฏใฉใ‚“ใช่กจๆƒ…ใซ ๅฏพใ—ใฆใ‚‚่กŒใˆใพใ™
03:49
Now we have a control of a model of a person,
77
229485
2499
ใ“ใ‚Œใงไบบ็‰ฉใฎๅ‹•ใ‹ใ›ใ‚‹ ใƒขใƒ‡ใƒซใŒใงใใพใ—ใŸ
03:52
and the way it's controlled now is by a sequence of static photos.
78
232008
3795
ๅ‹•ใใฏไธ€้€ฃใฎๅ†™็œŸใซ ๅˆใ‚ใ›ใŸใ‚‚ใฎใซใชใ‚Šใพใ™
03:55
Notice how the wrinkles come and go, depending on the expression.
79
235827
3126
่กจๆƒ…ใซๅฟœใ˜ใฆ็šบใŒ็พใ‚ŒใŸใ‚Š ๆถˆใˆใŸใ‚Šใ™ใ‚‹ใฎใซๆณจๆ„ใ—ใฆใใ ใ•ใ„
04:00
We can also use a video to drive the model.
80
240109
2746
ใƒขใƒ‡ใƒซใ‚’ๅ‹•ใ‹ใ™ใฎใซ ๆ˜ ๅƒใ‚’ไฝฟใ†ใ“ใจใ‚‚ใงใใพใ™
04:02
(Video) Daniel Craig: Right, but somehow,
81
242879
2593
(ใƒ€ใƒ‹ใ‚จใƒซใƒปใ‚ฏใƒฌใ‚คใ‚ฐ) ใˆใˆ ใ—ใ‹ใ—็ง้”ใฏใฉใ†ใซใ‹
04:05
we've managed to attract some more amazing people.
82
245496
3771
ใ•ใ‚‰ใซ็ด ๆ™ดใ‚‰ใ—ใ„ไบบใŸใกใ‚’ ๅผ•ใๅ…ฅใ‚Œใพใ—ใŸ
(่ฌ›ๆผ”่€…) ใ“ใ‚Œใง้ข็™ฝใ„ใ“ใจใŒใงใใพใ™
04:10
SS: And here's another fun demo.
83
250021
1642
04:11
So what you see here are controllable models
84
251687
2246
ใ“ใ“ใซๅ‡บใฆใ„ใ‚‹ใฎใฏ
04:13
of people I built from their internet photos.
85
253957
2444
ใƒใƒƒใƒˆไธŠใฎๅ†™็œŸใ‹ใ‚‰ไฝœใฃใŸ ๆœ‰ๅไบบใฎใƒขใƒ‡ใƒซใงใ™
04:16
Now, if you transfer the motion from the input video,
86
256425
2904
ๅ…ฅๅŠ›ๆบใจใชใ‚‹ๆ˜ ๅƒใฎ ๅ‹•ใใซๅˆใ‚ใ›ใฆ
04:19
we can actually drive the entire party.
87
259353
2152
ๅ…จ้ƒจใฎ้ก”ใ‚’ๅ‹•ใ‹ใ™ ใ“ใจใŒใงใใพใ™
04:21
George W. Bush: It's a difficult bill to pass,
88
261529
2172
(ใƒ–ใƒƒใ‚ทใƒฅ) ใ“ใ‚Œใฏ ้€šใ™ใฎใŒ้›ฃใ—ใ„ๆณ•ๆกˆใง
04:23
because there's a lot of moving parts,
89
263725
2303
ๆง‹ๆˆ่ฆ็ด ใŒใŸใใ•ใ‚“ใ‚ใ‚Š
็ซ‹ๆณ•ใฎ้Ž็จ‹ใฏ่ฆ‹่‹ฆใ—ใ„ใ‚‚ใฎใซ ใชใ‚‹ใ‹ใ‚‚ใ—ใ‚Œใพใ›ใ‚“
04:26
and the legislative processes can be ugly.
90
266052
5231
04:31
(Applause)
91
271307
1630
(ๆ‹ๆ‰‹)
04:32
SS: So coming back a little bit,
92
272961
1837
(่ฌ›ๆผ”่€…) ๅฐ‘ใ—่ฉฑใ‚’ๆˆปใ™ใจ
04:34
our ultimate goal, rather, is to capture their mannerisms
93
274822
3191
็ง้”ใฎ็ฉถๆฅตใฎ็›ฎๆจ™ใฏ
ใใ‚Œใžใ‚ŒใฎไบบใŒ่ฉฑใ—ใŸใ‚Š็ฌ‘ใฃใŸใ‚Šใ™ใ‚‹ๆ™‚ใฎ ็‹ฌ็‰นใชใ‚„ใ‚Šๆ–นใ‚„็™–ใ‚’ๆ‰ใˆใ‚‹ใจใ„ใ†ใ“ใจใงใ™
04:38
or the unique way each of these people talks and smiles.
94
278037
3045
04:41
So to do that, can we actually teach the computer
95
281106
2313
ใใฎไบบ็‰ฉใŒ่ฉฑใ—ใฆใ„ใ‚‹ ๆ˜ ๅƒใ‚’่ฆ‹ใ›ใ‚‹ใ ใ‘ใง
04:43
to imitate the way someone talks
96
283443
2222
ใ‚ณใƒณใƒ”ใƒฅใƒผใ‚ฟใƒผใŒ ใใฎไบบใฎ่ฉฑใ—ๆ–นใ‚’
04:45
by only showing it video footage of the person?
97
285689
2420
็œŸไผผใ‚‰ใ‚Œใ‚‹ใ‚ˆใ†ใซ ใงใใ‚‹ใฎใงใ—ใ‚‡ใ†ใ‹๏ผŸ
04:48
And what I did exactly was, I let a computer watch
98
288898
2577
ใใ‚Œใง ใ‚ชใƒใƒžใŒๆผ”่ชฌใ—ใฆใ„ใ‚‹ 14ๆ™‚้–“ใฎๆ˜ ๅƒใ‚’
04:51
14 hours of pure Barack Obama giving addresses.
99
291499
3277
ใ‚ณใƒณใƒ”ใƒฅใƒผใ‚ฟใƒผใซ ่ฆ‹ใ›ใ‚‹ใ“ใจใซใ—ใพใ—ใŸ
04:55
And here's what we can produce given only his audio.
100
295443
3516
ใ“ใ‚Œใฏใ‚ชใƒใƒžใŒ่ฉฑใ™ๅฃฐใ ใ‘ใ‹ใ‚‰ ็”Ÿๆˆใ—ใŸๆ˜ ๅƒใงใ™
04:58
(Video) BO: The results are clear.
101
298983
1777
(ใ‚ชใƒใƒž) ็ตๆžœใฏๆ˜Žใ‚‰ใ‹ใงใ™
05:00
America's businesses have created 14.5 million new jobs
102
300784
4349
ใ‚ขใƒกใƒชใ‚ซใฎ็”ฃๆฅญ็•Œใฏ 75ใƒถๆœˆใซใ‚ใŸใ‚Š
1450ไธ‡ใฎๆ–ฐใŸใชไป•ไบ‹ใ‚’ ็”Ÿใฟๅ‡บใ—ใŸใฎใงใ™
05:05
over 75 straight months.
103
305157
2774
05:07
SS: So what's being synthesized here is only the mouth region,
104
307955
2905
(่ฌ›ๆผ”่€…) ใ“ใ“ใงๅˆๆˆใ•ใ‚Œใฆใ„ใ‚‹ใฎใฏ ๅฃใฎ้ƒจๅˆ†ใ ใ‘ใง
05:10
and here's how we do it.
105
310884
1540
ใ“ใ‚“ใช้ขจใซใ—ใฆใ„ใพใ™
05:12
Our pipeline uses a neural network
106
312764
1826
็ง้”ใฎใ‚ทใ‚นใƒ†ใƒ ใฏ ใƒ‹ใƒฅใƒผใƒฉใƒซใƒใƒƒใƒˆใƒฏใƒผใ‚ฏใ‚’ไฝฟใฃใฆ
05:14
to convert and input audio into these mouth points.
107
314614
2936
ๅ…ฅๅŠ›ใ•ใ‚ŒใŸ้Ÿณๅฃฐใ‚’ ๅฃใฎไฝ็ฝฎใ‚’่กจใ™็‚นใซๅค‰ๆ›ใ—ใพใ™
05:18
(Video) BO: We get it through our job or through Medicare or Medicaid.
108
318547
4225
(ใ‚ชใƒใƒž) ไป•ไบ‹ใ‚„ ใƒกใƒ‡ใ‚ฃใ‚ฑใ‚ข ใƒกใƒ‡ใ‚ฃใ‚ฑใ‚คใƒ‰ใ‚’้€šใ˜ใฆๅพ—ใฆใ„ใพใ™
05:22
SS: Then we synthesize the texture, enhance details and teeth,
109
322796
3420
(่ฌ›ๆผ”่€…) ใใ‚Œใ‹ใ‚‰่ณชๆ„Ÿใ‚’ๅˆๆˆใ— ็ดฐ้ƒจใ‚„ๆญฏใ‚’่ฃœใ„
05:26
and blend it into the head and background from a source video.
110
326240
3074
ๅ…ƒใฎๆ˜ ๅƒใฎ้ ญ้ƒจใจ่ƒŒๆ™ฏใซ ๅŸ‹ใ‚่พผใฟใพใ™
05:29
(Video) BO: Women can get free checkups,
111
329338
1905
(ใ‚ชใƒใƒž) ๅฅณๆ€งใฏ็„กๆ–™ใฎ ๅฅๅบท่จบๆ–ญใ‚’ๅ—ใ‘ใ‚‰ใ‚Œ
05:31
and you can't get charged more just for being a woman.
112
331267
2968
ๅฅณๆ€งใจใ„ใ†ใ ใ‘ใง ไฝ™ๅˆ†ใซๆ”ฏๆ‰•ใ†ใ“ใจใฏใ‚ใ‚Šใพใ›ใ‚“
05:34
Young people can stay on a parent's plan until they turn 26.
113
334973
3306
ๅญไพ›ใฏ26ๆญณใซใชใ‚‹ใพใง ่ฆชใฎไฟ้™บใŒไฝฟใˆใพใ™
05:39
SS: I think these results seem very realistic and intriguing,
114
339267
2952
(่ฌ›ๆผ”่€…) ็ตๆžœใจใ—ใฆใงใใŸใ‚‚ใฎใฏ ใจใฆใ‚‚ใƒชใ‚ขใƒซใง่ˆˆๅ‘ณๆทฑใ„ใ‚‚ใฎใงใ™ใŒ
05:42
but at the same time frightening, even to me.
115
342243
3173
ๅŒๆ™‚ใซ็ง่‡ช่บซใงใ‚‚ ๆ€–ใ„ใจๆ„Ÿใ˜ใพใ™
05:45
Our goal was to build an accurate model of a person, not to misrepresent them.
116
345440
4015
็ง้”ใฎ็›ฎๆจ™ใฏไบบ็‰ฉใฎๆญฃ็ขบใชใƒขใƒ‡ใƒซใ‚’ไฝœใ‚‹ใ“ใจใง ่ชฐใ‹ใ‚’้จ™ใ‚‹ใ“ใจใงใฏใ‚ใ‚Šใพใ›ใ‚“
05:49
But one thing that concerns me is its potential for misuse.
117
349956
3111
ใ—ใ‹ใ—ใ“ใ‚ŒใŒๆ‚ช็”จใ•ใ‚Œใ‚‹ๅฏ่ƒฝๆ€งใ‚’ ๅฑๆƒงใ—ใฆใ„ใพใ™
05:53
People have been thinking about this problem for a long time,
118
353958
2971
ใ“ใฎๅ•้กŒใซใคใ„ใฆใฏ ใƒ•ใ‚ฉใƒˆใ‚ทใƒงใƒƒใƒ—ใŒ็พใ‚Œใฆไปฅๆฅ
05:56
since the days when Photoshop first hit the market.
119
356953
2381
ใฟใ‚“ใชใšใฃใจ่€ƒใˆใฆใใพใ—ใŸ
05:59
As a researcher, I'm also working on countermeasure technology,
120
359862
3801
็ ”็ฉถ่€…ใจใ—ใฆ ็งใฏ ๅฏพ็ญ–ๆŠ€่ก“ใฎ้–‹็™บใ‚‚ใ—ใฆใ„ใฆ
06:03
and I'm part of an ongoing effort at AI Foundation,
121
363687
2942
AI Foundationใงใฎ ๅ–ใ‚Š็ต„ใฟใซๅ‚ๅŠ ใ—ใฆใ„ใพใ™
06:06
which uses a combination of machine learning and human moderators
122
366653
3397
ใใ‚ŒใฏๆฉŸๆขฐๅญฆ็ฟ’ใจ ไบบ้–“ใฎใƒขใƒ‡ใƒฌใƒผใ‚ฟใƒผใฎ็ต„ใฟๅˆใ‚ใ›ใซใ‚ˆใ‚Š
06:10
to detect fake images and videos,
123
370074
2144
ๅฝ็‰ฉใฎ็”ปๅƒใ‚„ๆ˜ ๅƒใ‚’ ๆคœๅ‡บใ—ใ‚ˆใ†ใจใ„ใ†
06:12
fighting against my own work.
124
372242
1514
็ง่‡ช่บซใฎ็ ”็ฉถใซ ๅฏพๆŠ—ใ™ใ‚‹ใ‚‚ใฎใงใ™
06:14
And one of the tools we plan to release is called Reality Defender,
125
374675
3190
ๅ…ฌ้–‹ใ‚’ไบˆๅฎšใ—ใฆใ„ใ‚‹ใƒ„ใƒผใƒซใซ Reality Defender ใŒใ‚ใ‚Š
06:17
which is a web-browser plug-in that can flag potentially fake content
126
377889
4039
ใ“ใ‚Œใฏใƒ–ใƒฉใ‚ฆใ‚ถใƒผใฎใƒ—ใƒฉใ‚ฐใ‚คใƒณใง
ๅฝ็‰ฉใฎๅฏ่ƒฝๆ€งใฎใ‚ใ‚‹ใ‚ณใƒณใƒ†ใƒณใƒ„ใซๅฏพใ—ใฆ ่ญฆๅ‘Šใ‚’ๅ‡บใ™ใ‚ˆใ†ใซใชใฃใฆใ„ใพใ™
06:21
automatically, right in the browser.
127
381952
2533
06:24
(Applause)
128
384509
4228
(ๆ‹ๆ‰‹)
06:28
Despite all this, though,
129
388761
1453
ใใ‚Œใงใ‚‚็œŸๅฝใฎ็ขบ่ชใŒ ใชใ•ใ‚Œใ‚‹ๅ‰ใซ
06:30
fake videos could do a lot of damage,
130
390238
1840
ๅฝใฎๆ˜ ๅƒใŒๅคงใใช่ขซๅฎณใ‚’ ใ‚‚ใŸใ‚‰ใ™ใ“ใจใ‚‚ใ‚ใ‚Šใ†ใ‚‹ใฎใง
06:32
even before anyone has a chance to verify,
131
392102
3294
ๅฝใฎๆ˜ ๅƒใŒๅคงใใช่ขซๅฎณใ‚’ ใ‚‚ใŸใ‚‰ใ™ใ“ใจใ‚‚ใ‚ใ‚Šใ†ใ‚‹ใฎใง
06:35
so it's very important that we make everyone aware
132
395420
2722
็พๅœจใฉใฎใ‚ˆใ†ใชใ“ใจใŒ ๅฏ่ƒฝใซใชใฃใฆใ„ใ‚‹ใฎใ‹ใ‚’
ใฟใ‚“ใช็†่งฃใ—ใฆใ„ใ‚‹ใ“ใจใŒ้‡่ฆใง
06:38
of what's currently possible
133
398166
2007
06:40
so we can have the right assumption and be critical about what we see.
134
400197
3369
ใใ‚Œใซใ‚ˆใ‚Š้ฉๅˆ‡ใชไปฎๅฎšใ‚’ใ— ๆ‰นๅˆค็š„ใช็›ฎใง็‰ฉไบ‹ใ‚’่ฆ‹ใ‚‰ใ‚Œใ‚‹ใ‚ˆใ†ใซใชใ‚‹ใงใ—ใ‚‡ใ†
06:44
There's still a long way to go before we can fully model individual people
135
404423
5007
ไบบ็‰ฉใฎๅฎŒๅ…จใชใƒขใƒ‡ใƒซใ‚’ไฝœใ‚Œ ๅฎ‰ๅ…จๆ€งใ‚‚็ขบไฟใงใใ‚‹ใพใงใซใฏ
06:49
and before we can ensure the safety of this technology.
136
409454
2786
ใพใ ใพใ ๆ™‚้–“ใŒใ‹ใ‹ใ‚‹ใงใ—ใ‚‡ใ†
ใงใ‚‚็งใฏๅธŒๆœ›ใจ็†ฑๆ„ใ‚’ๆŒใฃใฆใ„ใพใ™
06:53
But I'm excited and hopeful,
137
413097
1587
06:54
because if we use it right and carefully,
138
414708
3539
ใ“ใฎๆŠ€่ก“ใ‚’ๆญฃใ—ใ ๆณจๆ„ใ—ใฆไฝฟใ†ใชใ‚‰
06:58
this tool can allow any individual's positive impact on the world
139
418271
4309
่ชฐใ‚‚ใŒๅบƒใไธ–็•Œใซ
่‰ฏใ„ๅฝฑ้Ÿฟใ‚’ไธŽใˆใ‚‰ใ‚Œใ‚‹ใ‚ˆใ†ใซใชใ‚Š
07:02
to be massively scaled
140
422604
2190
07:04
and really help shape our future the way we want it to be.
141
424818
2742
ใฟใ‚“ใชใŒๆœ›ใ‚€ๆœชๆฅใ‚’็ฏ‰ใ ๅŠฉใ‘ใซใชใ‚‹ใฏใšใ ใ‹ใ‚‰ใงใ™
07:07
Thank you.
142
427584
1151
ใ‚ใ‚ŠใŒใจใ†ใ”ใ–ใ„ใพใ—ใŸ
07:08
(Applause)
143
428759
5090
(ๆ‹ๆ‰‹)
ใ“ใฎใ‚ฆใ‚งใƒ–ใ‚ตใ‚คใƒˆใซใคใ„ใฆ

ใ“ใฎใ‚ตใ‚คใƒˆใงใฏ่‹ฑ่ชžๅญฆ็ฟ’ใซๅฝน็ซ‹ใคYouTubeๅ‹•็”ปใ‚’็ดนไป‹ใ—ใพใ™ใ€‚ไธ–็•Œไธญใฎไธ€ๆต่ฌ›ๅธซใซใ‚ˆใ‚‹่‹ฑ่ชžใƒฌใƒƒใ‚นใƒณใ‚’่ฆ‹ใ‚‹ใ“ใจใŒใงใใพใ™ใ€‚ๅ„ใƒ“ใƒ‡ใ‚ชใฎใƒšใƒผใ‚ธใซ่กจ็คบใ•ใ‚Œใ‚‹่‹ฑ่ชžๅญ—ๅน•ใ‚’ใƒ€ใƒ–ใƒซใ‚ฏใƒชใƒƒใ‚ฏใ™ใ‚‹ใจใ€ใใ“ใ‹ใ‚‰ใƒ“ใƒ‡ใ‚ชใ‚’ๅ†็”Ÿใ™ใ‚‹ใ“ใจใŒใงใใพใ™ใ€‚ๅญ—ๅน•ใฏใƒ“ใƒ‡ใ‚ชใฎๅ†็”ŸใจๅŒๆœŸใ—ใฆใ‚นใ‚ฏใƒญใƒผใƒซใ—ใพใ™ใ€‚ใ”ๆ„่ฆ‹ใƒปใ”่ฆๆœ›ใŒใ”ใ–ใ„ใพใ—ใŸใ‚‰ใ€ใ“ใกใ‚‰ใฎใŠๅ•ใ„ๅˆใ‚ใ›ใƒ•ใ‚ฉใƒผใƒ ใ‚ˆใ‚Šใ”้€ฃ็ตกใใ ใ•ใ„ใ€‚

https://forms.gle/WvT1wiN1qDtmnspy7