How we can build AI to help humans, not hurt us | Margaret Mitchell

80,996 views ・ 2018-03-12

TED


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

00:13
I work on helping computers communicate about the world around us.
0
13381
4015
00:17
There are a lot of ways to do this,
1
17754
1793
00:19
and I like to focus on helping computers
2
19571
2592
00:22
to talk about what they see and understand.
3
22187
2874
00:25
Given a scene like this,
4
25514
1571
00:27
a modern computer-vision algorithm
5
27109
1905
00:29
can tell you that there's a woman and there's a dog.
6
29038
3095
00:32
It can tell you that the woman is smiling.
7
32157
2706
00:34
It might even be able to tell you that the dog is incredibly cute.
8
34887
3873
00:38
I work on this problem
9
38784
1349
00:40
thinking about how humans understand and process the world.
10
40157
4212
00:45
The thoughts, memories and stories
11
45577
2952
00:48
that a scene like this might evoke for humans.
12
48553
2818
00:51
All the interconnections of related situations.
13
51395
4285
00:55
Maybe you've seen a dog like this one before,
14
55704
3126
00:58
or you've spent time running on a beach like this one,
15
58854
2969
01:01
and that further evokes thoughts and memories of a past vacation,
16
61847
4778
01:06
past times to the beach,
17
66649
1920
01:08
times spent running around with other dogs.
18
68593
2603
01:11
One of my guiding principles is that by helping computers to understand
19
71688
5207
01:16
what it's like to have these experiences,
20
76919
2896
01:19
to understand what we share and believe and feel,
21
79839
5176
01:26
then we're in a great position to start evolving computer technology
22
86094
4310
01:30
in a way that's complementary with our own experiences.
23
90428
4587
01:35
So, digging more deeply into this,
24
95539
3387
01:38
a few years ago I began working on helping computers to generate human-like stories
25
98950
5905
01:44
from sequences of images.
26
104879
1666
01:47
So, one day,
27
107427
1904
01:49
I was working with my computer to ask it what it thought about a trip to Australia.
28
109355
4622
01:54
It took a look at the pictures, and it saw a koala.
29
114768
2920
01:58
It didn't know what the koala was,
30
118236
1643
01:59
but it said it thought it was an interesting-looking creature.
31
119903
2999
02:04
Then I shared with it a sequence of images about a house burning down.
32
124053
4004
02:09
It took a look at the images and it said,
33
129704
3285
02:13
"This is an amazing view! This is spectacular!"
34
133013
3500
02:17
It sent chills down my spine.
35
137450
2095
02:20
It saw a horrible, life-changing and life-destroying event
36
140983
4572
02:25
and thought it was something positive.
37
145579
2382
02:27
I realized that it recognized the contrast,
38
147985
3441
02:31
the reds, the yellows,
39
151450
2699
02:34
and thought it was something worth remarking on positively.
40
154173
3078
02:37
And part of why it was doing this
41
157928
1615
02:39
was because most of the images I had given it
42
159577
2945
02:42
were positive images.
43
162546
1840
02:44
That's because people tend to share positive images
44
164903
3658
02:48
when they talk about their experiences.
45
168585
2190
02:51
When was the last time you saw a selfie at a funeral?
46
171267
2541
02:55
I realized that, as I worked on improving AI
47
175434
3095
02:58
task by task, dataset by dataset,
48
178553
3714
03:02
that I was creating massive gaps,
49
182291
2897
03:05
holes and blind spots in what it could understand.
50
185212
3999
03:10
And while doing so,
51
190307
1334
03:11
I was encoding all kinds of biases.
52
191665
2483
03:15
Biases that reflect a limited viewpoint,
53
195029
3318
03:18
limited to a single dataset --
54
198371
2261
03:21
biases that can reflect human biases found in the data,
55
201283
3858
03:25
such as prejudice and stereotyping.
56
205165
3104
03:29
I thought back to the evolution of the technology
57
209554
3057
03:32
that brought me to where I was that day --
58
212635
2502
03:35
how the first color images
59
215966
2233
03:38
were calibrated against a white woman's skin,
60
218223
3048
03:41
meaning that color photography was biased against black faces.
61
221665
4145
03:46
And that same bias, that same blind spot
62
226514
2925
03:49
continued well into the '90s.
63
229463
1867
03:51
And the same blind spot continues even today
64
231701
3154
03:54
in how well we can recognize different people's faces
65
234879
3698
03:58
in facial recognition technology.
66
238601
2200
04:01
I though about the state of the art in research today,
67
241323
3143
04:04
where we tend to limit our thinking to one dataset and one problem.
68
244490
4514
04:09
And that in doing so, we were creating more blind spots and biases
69
249688
4881
04:14
that the AI could further amplify.
70
254593
2277
04:17
I realized then that we had to think deeply
71
257712
2079
04:19
about how the technology we work on today looks in five years, in 10 years.
72
259815
5519
04:25
Humans evolve slowly, with time to correct for issues
73
265990
3142
04:29
in the interaction of humans and their environment.
74
269156
3534
04:33
In contrast, artificial intelligence is evolving at an incredibly fast rate.
75
273276
5429
04:39
And that means that it really matters
76
279013
1773
04:40
that we think about this carefully right now --
77
280810
2317
04:44
that we reflect on our own blind spots,
78
284180
3008
04:47
our own biases,
79
287212
2317
04:49
and think about how that's informing the technology we're creating
80
289553
3857
04:53
and discuss what the technology of today will mean for tomorrow.
81
293434
3902
04:58
CEOs and scientists have weighed in on what they think
82
298593
3191
05:01
the artificial intelligence technology of the future will be.
83
301808
3325
05:05
Stephen Hawking warns that
84
305157
1618
05:06
"Artificial intelligence could end mankind."
85
306799
3007
05:10
Elon Musk warns that it's an existential risk
86
310307
2683
05:13
and one of the greatest risks that we face as a civilization.
87
313014
3574
05:17
Bill Gates has made the point,
88
317665
1452
05:19
"I don't understand why people aren't more concerned."
89
319141
3185
05:23
But these views --
90
323412
1318
05:25
they're part of the story.
91
325618
1734
05:28
The math, the models,
92
328079
2420
05:30
the basic building blocks of artificial intelligence
93
330523
3070
05:33
are something that we call access and all work with.
94
333617
3135
05:36
We have open-source tools for machine learning and intelligence
95
336776
3785
05:40
that we can contribute to.
96
340585
1734
05:42
And beyond that, we can share our experience.
97
342919
3340
05:46
We can share our experiences with technology and how it concerns us
98
346760
3468
05:50
and how it excites us.
99
350252
1467
05:52
We can discuss what we love.
100
352251
1867
05:55
We can communicate with foresight
101
355244
2031
05:57
about the aspects of technology that could be more beneficial
102
357299
4857
06:02
or could be more problematic over time.
103
362180
2600
06:05
If we all focus on opening up the discussion on AI
104
365799
4143
06:09
with foresight towards the future,
105
369966
1809
06:13
this will help create a general conversation and awareness
106
373093
4270
06:17
about what AI is now,
107
377387
2513
06:21
what it can become
108
381212
2001
06:23
and all the things that we need to do
109
383237
1785
06:25
in order to enable that outcome that best suits us.
110
385046
3753
06:29
We already see and know this in the technology that we use today.
111
389490
3674
06:33
We use smart phones and digital assistants and Roombas.
112
393767
3880
06:38
Are they evil?
113
398457
1150
06:40
Maybe sometimes.
114
400268
1547
06:42
Are they beneficial?
115
402664
1333
06:45
Yes, they're that, too.
116
405005
1533
06:48
And they're not all the same.
117
408236
1761
06:50
And there you already see a light shining on what the future holds.
118
410489
3540
06:54
The future continues on from what we build and create right now.
119
414942
3619
06:59
We set into motion that domino effect
120
419165
2642
07:01
that carves out AI's evolutionary path.
121
421831
2600
07:05
In our time right now, we shape the AI of tomorrow.
122
425173
2871
07:08
Technology that immerses us in augmented realities
123
428566
3699
07:12
bringing to life past worlds.
124
432289
2566
07:15
Technology that helps people to share their experiences
125
435844
4312
07:20
when they have difficulty communicating.
126
440180
2262
07:23
Technology built on understanding the streaming visual worlds
127
443323
4532
07:27
used as technology for self-driving cars.
128
447879
3079
07:32
Technology built on understanding images and generating language,
129
452490
3413
07:35
evolving into technology that helps people who are visually impaired
130
455927
4063
07:40
be better able to access the visual world.
131
460014
2800
07:42
And we also see how technology can lead to problems.
132
462838
3261
07:46
We have technology today
133
466885
1428
07:48
that analyzes physical characteristics we're born with --
134
468337
3835
07:52
such as the color of our skin or the look of our face --
135
472196
3272
07:55
in order to determine whether or not we might be criminals or terrorists.
136
475492
3804
07:59
We have technology that crunches through our data,
137
479688
2905
08:02
even data relating to our gender or our race,
138
482617
2896
08:05
in order to determine whether or not we might get a loan.
139
485537
2865
08:09
All that we see now
140
489494
1579
08:11
is a snapshot in the evolution of artificial intelligence.
141
491097
3617
08:15
Because where we are right now,
142
495763
1778
08:17
is within a moment of that evolution.
143
497565
2238
08:20
That means that what we do now will affect what happens down the line
144
500690
3802
08:24
and in the future.
145
504516
1200
08:26
If we want AI to evolve in a way that helps humans,
146
506063
3951
08:30
then we need to define the goals and strategies
147
510038
2801
08:32
that enable that path now.
148
512863
1733
08:35
What I'd like to see is something that fits well with humans,
149
515680
3738
08:39
with our culture and with the environment.
150
519442
2800
08:43
Technology that aids and assists those of us with neurological conditions
151
523435
4484
08:47
or other disabilities
152
527943
1721
08:49
in order to make life equally challenging for everyone.
153
529688
3216
08:54
Technology that works
154
534097
1421
08:55
regardless of your demographics or the color of your skin.
155
535542
3933
09:00
And so today, what I focus on is the technology for tomorrow
156
540383
4742
09:05
and for 10 years from now.
157
545149
1733
09:08
AI can turn out in many different ways.
158
548530
2634
09:11
But in this case,
159
551688
1225
09:12
it isn't a self-driving car without any destination.
160
552937
3328
09:16
This is the car that we are driving.
161
556884
2400
09:19
We choose when to speed up and when to slow down.
162
559953
3595
09:23
We choose if we need to make a turn.
163
563572
2400
09:26
We choose what the AI of the future will be.
164
566868
3000
09:31
There's a vast playing field
165
571186
1337
09:32
of all the things that artificial intelligence can become.
166
572547
2965
09:36
It will become many things.
167
576064
1800
09:39
And it's up to us now,
168
579694
1732
09:41
in order to figure out what we need to put in place
169
581450
3061
09:44
to make sure the outcomes of artificial intelligence
170
584535
3807
09:48
are the ones that will be better for all of us.
171
588366
3066
09:51
Thank you.
172
591456
1150
09:52
(Applause)
173
592630
2187
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