1
00:00:01,077 --> 00:00:02,596
Some years ago,

2
00:00:02,596 --> 00:00:07,166
I was on an airplane with my son
who was just five years old at the time.

3
00:00:08,460 --> 00:00:13,623
My son was so excited
about being on this airplane with Mommy.

4
00:00:13,623 --> 00:00:17,489
He's looking all around
and he's checking things out

5
00:00:17,489 --> 00:00:19,176
and he's checking people out,

6
00:00:19,176 --> 00:00:21,387
and he sees this man, and he says, "Hey,

7
00:00:21,387 --> 00:00:23,843
that guy looks like Daddy."

8
00:00:24,049 --> 00:00:25,992
And I look at the man,

9
00:00:25,992 --> 00:00:29,683
and he didn't look anything
at all like my husband,

10
00:00:29,876 --> 00:00:31,245
nothing at all.

11
00:00:31,668 --> 00:00:34,403
And so then I start
looking around on the plane,

12
00:00:34,403 --> 00:00:36,273
and I notice this man

13
00:00:36,273 --> 00:00:39,715
was the only black guy on the plane.

14
00:00:41,145 --> 00:00:42,531
And I thought,

15
00:00:42,531 --> 00:00:44,321
all right.

16
00:00:44,552 --> 00:00:47,651
I'm going to have to have
a little talk with my son

17
00:00:47,651 --> 00:00:50,070
about how not all black people look alike.

18
00:00:50,293 --> 00:00:54,694
My son, he lifts his head up,
and he says to me,

19
00:00:54,694 --> 00:00:59,203
"I hope he doesn't rob the plane."

20
00:00:59,582 --> 00:01:02,174
And I said, "What? What did you say?"

21
00:01:02,174 --> 00:01:05,979
And he says, "Well, I hope that man
doesn't rob the plane."

22
00:01:05,979 --> 00:01:09,839
And I said, "Well, why would you say that?

23
00:01:10,700 --> 00:01:13,386
You know Daddy wouldn't rob a plane."

24
00:01:13,386 --> 00:01:16,449
And he says, "Yeah, yeah,
yeah, well, I know."

25
00:01:16,449 --> 00:01:18,809
And I said, "Well,
why would you say that?"

26
00:01:18,809 --> 00:01:24,438
And he looked at me
with this really sad face,

27
00:01:24,438 --> 00:01:26,743
and he says,

28
00:01:26,743 --> 00:01:29,599
"I don't know why I said that.

29
00:01:29,599 --> 00:01:33,273
I don't know why I was thinking that."

30
00:01:34,010 --> 00:01:37,551
We are living with such severe
racial stratification

31
00:01:37,551 --> 00:01:39,776
that even a five-year old

32
00:01:39,776 --> 00:01:42,572
can tell us what's supposed
to happen next,

33
00:01:44,292 --> 00:01:46,422
even with no evildoer,

34
00:01:46,422 --> 00:01:50,367
even with no explicit hatred.

35
00:01:50,367 --> 00:01:54,169
This association between
blackness and crime

36
00:01:54,169 --> 00:01:58,357
made its way into the mind
of my five-year old.

37
00:02:00,050 --> 00:02:04,463
It makes its way into all of our children,

38
00:02:04,463 --> 00:02:07,052
into all of us.

39
00:02:07,245 --> 00:02:10,690
Our minds are shaped
by the racial disparities

40
00:02:10,690 --> 00:02:12,864
we see out in the world,

41
00:02:12,864 --> 00:02:18,205
and the narratives that help us
to make sense of the disparities we see.

42
00:02:19,899 --> 00:02:22,448
Those people are criminals.

43
00:02:22,448 --> 00:02:24,335
Those people are violent.

44
00:02:24,335 --> 00:02:27,300
Those people are to be feared.

45
00:02:28,100 --> 00:02:31,657
When my research team
brought people into our lab

46
00:02:31,657 --> 00:02:33,395
and exposed them to faces,

47
00:02:33,395 --> 00:02:36,343
we found that exposure to black faces

48
00:02:37,052 --> 00:02:43,634
led them to see blurry images
of guns with greater clarity and speed.

49
00:02:43,902 --> 00:02:47,114
Bias cannot only control what we see,

50
00:02:47,114 --> 00:02:48,941
but where we look.

51
00:02:48,941 --> 00:02:52,408
We found that prompting people
to think of violent crime

52
00:02:52,408 --> 00:02:56,597
can lead them to direct their eyes
onto a black face

53
00:02:56,858 --> 00:02:59,011
and away from a white face.

54
00:02:59,011 --> 00:03:03,041
Prompting police officers
to think of capturing and shooting

55
00:03:03,041 --> 00:03:04,131
and arresting

56
00:03:04,131 --> 00:03:07,936
leads their eyes to settle
on black faces too.

57
00:03:07,936 --> 00:03:12,772
Bias can infect every aspect
of our criminal justice system.

58
00:03:13,378 --> 00:03:16,536
In a large dataset
of death-eligible defendants,

59
00:03:16,536 --> 00:03:18,478
we found that looking more black

60
00:03:18,478 --> 00:03:22,986
more than doubled their chances
of receiving a death sentence,

61
00:03:23,828 --> 00:03:26,345
at least when their victims were white.

62
00:03:26,345 --> 00:03:29,232
This effect is significant
even though we controlled

63
00:03:29,232 --> 00:03:33,234
for the severity of the crime

64
00:03:33,234 --> 00:03:34,637
and the defendant's attractiveness,

65
00:03:34,637 --> 00:03:36,089
and no matter what we controlled for,

66
00:03:36,089 --> 00:03:38,654
we found that black people

67
00:03:38,654 --> 00:03:40,785
were punished in proportion

68
00:03:40,785 --> 00:03:44,041
to the blackness
of their physical features:

69
00:03:44,041 --> 00:03:45,674
the more black,

70
00:03:45,674 --> 00:03:47,600
the more death-worthy.

71
00:03:47,600 --> 00:03:52,107
Bias can also influence
how teachers discipline students.

72
00:03:52,107 --> 00:03:54,113
My colleagues and I

73
00:03:54,113 --> 00:03:56,606
have found that teachers express a desire

74
00:03:56,606 --> 00:03:59,116
to discipline a black
middle school student

75
00:03:59,116 --> 00:04:03,535
more harshly than a white student
for the same repeated infractions.

76
00:04:04,200 --> 00:04:05,440
In a recent study,

77
00:04:05,440 --> 00:04:09,453
we're finding that teachers
treat black students as a group

78
00:04:09,453 --> 00:04:12,388
but white students as individuals.

79
00:04:12,388 --> 00:04:15,947
If, for example,
one black student misbehaves

80
00:04:15,947 --> 00:04:20,799
and then a different black student
misbehaves a few days later,

81
00:04:20,799 --> 00:04:24,050
the teacher responds
to that second black student

82
00:04:24,050 --> 00:04:26,675
as if he had misbehaved twice.

83
00:04:27,214 --> 00:04:30,048
It's as though the sins of one child

84
00:04:30,048 --> 00:04:32,158
get piled on to the other.

85
00:04:32,718 --> 00:04:36,035
We create categories
to make sense of the world,

86
00:04:36,035 --> 00:04:39,725
to assert some control and coherence

87
00:04:39,725 --> 00:04:44,270
to the stimuli that we're
constantly being bombarded with.

88
00:04:44,270 --> 00:04:48,360
Categorization and the bias that it seeds

89
00:04:48,360 --> 00:04:51,198
allow our brains to make judgments

90
00:04:51,198 --> 00:04:53,370
more quickly and efficiently,

91
00:04:53,370 --> 00:04:56,684
and we do this by instinctively
relying on patterns

92
00:04:56,684 --> 00:04:58,376
that seem predictable.

93
00:04:58,376 --> 00:05:01,758
Yet just as the categories we create

94
00:05:01,758 --> 00:05:04,463
allow us to make quick decisions,

95
00:05:04,463 --> 00:05:06,878
they also reinforce bias.

96
00:05:06,878 --> 00:05:11,295
So the very things that help us
to see the world

97
00:05:11,295 --> 00:05:13,851
also can blind us to it.

98
00:05:13,851 --> 00:05:16,549
They render our choices effortless,

99
00:05:16,549 --> 00:05:19,441
friction-free.

100
00:05:19,441 --> 00:05:22,397
Yet the exact a heavy toll.

101
00:05:22,397 --> 00:05:24,809
So what can we do?

102
00:05:24,809 --> 00:05:27,284
We are all vulnerable to bias,

103
00:05:27,284 --> 00:05:30,212
but we don't act on bias all the time.

104
00:05:30,212 --> 00:05:33,681
There are certain conditions
that can bring bias alive

105
00:05:33,681 --> 00:05:36,281
and other conditions that can muffle it.

106
00:05:36,281 --> 00:05:38,933
Let me give you an example.

107
00:05:38,933 --> 00:05:43,061
Many people are familiar
with the tech company Nextdoor.

108
00:05:43,460 --> 00:05:51,738
So their whole purpose is to create
stronger, healthier, safer neighborhoods.

109
00:05:51,738 --> 00:05:55,393
And so they offer this online space

110
00:05:55,393 --> 00:05:57,983
where neighbors can gather
and share information.

111
00:05:57,983 --> 00:06:02,015
Yet Nextdoor soon found
that they had a problem

112
00:06:02,015 --> 00:06:04,187
with racial profiling.

113
00:06:04,187 --> 00:06:06,130
In the typical case,

114
00:06:06,130 --> 00:06:08,713
people would look outside their window

115
00:06:08,713 --> 00:06:12,785
and see a black man
in their otherwise white neighborhood

116
00:06:12,785 --> 00:06:14,976
and make the snap judgment

117
00:06:14,976 --> 00:06:17,299
that he was up to no good,

118
00:06:17,299 --> 00:06:20,865
even when there was no evidence
of criminal wrongdoing.

119
00:06:21,127 --> 00:06:23,831
In many ways, how we behave online

120
00:06:23,831 --> 00:06:26,945
is a reflection of how
we behave in the world.

121
00:06:27,816 --> 00:06:31,372
But what we don't want to do
is create an easy-to-use system

122
00:06:31,372 --> 00:06:34,195
that can amplify bias

123
00:06:34,195 --> 00:06:36,439
and deepen racial disparities,

124
00:06:36,439 --> 00:06:39,117
rather than dismantling them.

125
00:06:39,117 --> 00:06:42,491
So the co-founder of Nextdoor
reached out to me and to others

126
00:06:42,491 --> 00:06:44,755
to try to figure out what to do,

127
00:06:44,755 --> 00:06:48,052
and they realized that
to curb racial profiling on the platform,

128
00:06:48,710 --> 00:06:50,601
they were going to have to add friction.

129
00:06:50,601 --> 00:06:53,244
That is, they were going
to have to slow people down.

130
00:06:53,244 --> 00:06:55,423
So Nextdoor had a choice to make,

131
00:06:55,423 --> 00:06:58,066
and against every impulse

132
00:06:58,066 --> 00:07:00,524
they decided to add friction.

133
00:07:00,524 --> 00:07:03,948
And they did this by adding
a simple checklist.

134
00:07:03,948 --> 00:07:06,310
There were three items on it.

135
00:07:06,310 --> 00:07:08,766
First, they asked users to pause

136
00:07:08,766 --> 00:07:12,608
and think, what was this person doing

137
00:07:12,608 --> 00:07:14,399
that made him suspicious?

138
00:07:14,399 --> 00:07:19,519
The category "black man"
is not ground for suspicion.

139
00:07:19,713 --> 00:07:24,833
Second, they asked users to describe
the person's physical features,

140
00:07:24,833 --> 00:07:27,551
not simply their race and gender.

141
00:07:27,551 --> 00:07:31,107
Third, they realized that a lot of people

142
00:07:31,107 --> 00:07:34,345
didn't seem to know
what racial profiling was,

143
00:07:34,345 --> 00:07:36,645
nor that they were engaging in it.

144
00:07:36,875 --> 00:07:40,069
So Nextdoor provided them
with a definition

145
00:07:40,069 --> 00:07:43,177
and told them that it
was strictly prohibited.

146
00:07:43,484 --> 00:07:47,060
Most of you have seen
those signs in airports

147
00:07:47,060 --> 00:07:50,262
and Metro stations,
"If you see something, say something."

148
00:07:50,262 --> 00:07:53,919
Nextdoor tried modifying this.

149
00:07:53,919 --> 00:07:59,329
"If you see something suspicious,
say something specific."

150
00:07:59,587 --> 00:08:01,700
And using this strategy

151
00:08:01,700 --> 00:08:04,290
by simply slowing people down,

152
00:08:04,290 --> 00:08:09,834
Nextdoor was able to curb
racial profiling by 75 percent.

153
00:08:10,766 --> 00:08:12,879
Now, people often will say to me,

154
00:08:12,879 --> 00:08:17,768
you can't add friction
in every situation, in every context,

155
00:08:17,768 --> 00:08:21,914
and especially for people who make
split second decisions all the time,

156
00:08:22,809 --> 00:08:25,586
but it turns out we can add friction

157
00:08:25,586 --> 00:08:28,246
to more situations than we think.

158
00:08:28,246 --> 00:08:30,691
Working with the Oakland Police Department

159
00:08:30,691 --> 00:08:31,919
in California,

160
00:08:31,919 --> 00:08:35,859
I and a number of my colleagues
were able to help the Department

161
00:08:35,859 --> 00:08:38,360
to reduce the number of stops they made

162
00:08:38,360 --> 00:08:41,805
of people who were not
committing any serious crimes,

163
00:08:42,238 --> 00:08:44,500
and we did this by pushing officers

164
00:08:44,500 --> 00:08:46,364
to ask themselves a question

165
00:08:46,364 --> 00:08:48,917
before each and every stop they made:

166
00:08:49,451 --> 00:08:52,797
is this stop intelligence-led,

167
00:08:52,797 --> 00:08:54,207
yes or no?

168
00:08:55,581 --> 00:08:57,215
In other words,

169
00:08:57,215 --> 00:09:02,317
do I have prior information
to tie this particular person

170
00:09:02,317 --> 00:09:04,295
to a specific crime?

171
00:09:04,882 --> 00:09:06,672
By adding that question

172
00:09:06,672 --> 00:09:09,435
to the form officer's
complete doing the stop,

173
00:09:09,435 --> 00:09:11,376
they slow down, they pause,

174
00:09:11,376 --> 00:09:15,237
they think, why am I considering
pulling this person over?

175
00:09:16,912 --> 00:09:22,203
In 2017, before we added that
intelligence-led question to the form,

176
00:09:23,655 --> 00:09:27,479
officers made about 32,000 stops
across the city.

177
00:09:27,934 --> 00:09:29,593
In that next year,

178
00:09:29,593 --> 00:09:31,887
with the addition of this question,

179
00:09:31,887 --> 00:09:34,401
that fell to 19,000 stops.

180
00:09:34,579 --> 00:09:40,612
African-American stops alone
fell by 43 percent.

181
00:09:40,612 --> 00:09:44,669
And stopping fewer black people
did not make the city any more dangerous.

182
00:09:44,669 --> 00:09:47,292
In fact, the crime rate continued to fall

183
00:09:47,292 --> 00:09:50,202
and the city became safer for everybody.

184
00:09:50,486 --> 00:09:56,334
So one solution can come from reducing
the number of unnecessary stops.

185
00:09:56,334 --> 00:10:00,690
Another can come from improving
the quality of the stops

186
00:10:00,690 --> 00:10:02,743
officers do make.

187
00:10:02,743 --> 00:10:05,349
And technology can help us here.

188
00:10:05,349 --> 00:10:07,878
We all know about George Floyd's death

189
00:10:07,878 --> 00:10:11,196
because those who tried to come to his aid

190
00:10:11,517 --> 00:10:17,131
held cell phone cameras
to record that horrific, fatal encounter

191
00:10:17,453 --> 00:10:18,918
with the police.

192
00:10:18,918 --> 00:10:21,703
But we have all sorts of technology

193
00:10:21,703 --> 00:10:24,133
that we're not putting to good use.

194
00:10:24,133 --> 00:10:26,396
Police departments across the country

195
00:10:26,396 --> 00:10:30,094
are now required to wear
body-worn cameras,

196
00:10:30,094 --> 00:10:35,570
so we have recordings of not only
the most extreme and horrific encounters

197
00:10:35,570 --> 00:10:38,354
but of everyday interactions.

198
00:10:38,753 --> 00:10:41,799
With an interdisciplinary
team at Stanford,

199
00:10:41,799 --> 00:10:44,806
we've begun to use
machine learning techniques

200
00:10:44,806 --> 00:10:47,898
to analyze large numbers of encounters.

201
00:10:47,898 --> 00:10:52,528
This is to better understand what happens
in routine traffic stops.

202
00:10:52,528 --> 00:10:56,094
What we found was that
even when police officers

203
00:10:56,094 --> 00:10:58,356
are behaving professionally,

204
00:10:58,356 --> 00:11:02,906
they speak to black drivers
less respectfully than white drivers.

205
00:11:03,529 --> 00:11:07,334
In fact, from the words officers use alone

206
00:11:07,334 --> 00:11:12,973
we could predict whether they were talking
to a black driver or a white driver.

207
00:11:13,655 --> 00:11:18,797
The problem is that the vast majority
of the footage from these cameras

208
00:11:18,797 --> 00:11:21,424
is not used by police departments

209
00:11:21,424 --> 00:11:23,858
to understand what's
going on on the street

210
00:11:23,858 --> 00:11:26,428
or to train officers.

211
00:11:26,659 --> 00:11:28,844
And that's a shame.

212
00:11:28,844 --> 00:11:31,532
How does a routine stop

213
00:11:31,532 --> 00:11:33,812
turn into a deadly encounter?

214
00:11:34,077 --> 00:11:36,287
How did this happen
in George Floyd's case?

215
00:11:36,287 --> 00:11:40,511
How did it happen in others?

216
00:11:40,511 --> 00:11:43,252
When my eldest son was 16 years old,

217
00:11:43,252 --> 00:11:46,277
he discovered that
when white people look at him,

218
00:11:46,277 --> 00:11:47,838
they feel fear.

219
00:11:49,338 --> 00:11:52,409
Elevators are the worst, he said.

220
00:11:52,409 --> 00:11:54,592
When those doors close,

221
00:11:54,592 --> 00:11:57,775
people are trapped in this tiny space

222
00:11:57,775 --> 00:12:00,262
with someone they have been taught

223
00:12:00,262 --> 00:12:02,242
to associate with danger.

224
00:12:03,046 --> 00:12:06,163
My son senses their discomfort,

225
00:12:06,163 --> 00:12:09,400
and he smiles to put them at ease,

226
00:12:09,400 --> 00:12:11,566
to calm their fears.

227
00:12:11,566 --> 00:12:13,534
When he speaks,

228
00:12:13,534 --> 00:12:15,720
their bodies relax.

229
00:12:15,720 --> 00:12:17,697
They breathe easier.

230
00:12:17,697 --> 00:12:20,251
They take pleasure in his cadence,

231
00:12:20,251 --> 00:12:22,418
his diction, his word choice.

232
00:12:23,241 --> 00:12:25,157
He sounds like one of them.

233
00:12:25,157 --> 00:12:29,867
I used to think that my son
was a natural extrovert like his father,

234
00:12:29,867 --> 00:12:34,413
but I realized at that moment
in that conversation

235
00:12:34,413 --> 00:12:38,006
that his smile was not a sign

236
00:12:38,006 --> 00:12:41,049
that he wanted to connect
with would-be strangers.

237
00:12:42,175 --> 00:12:45,850
It was a talisman he'd used
to protect himself,

238
00:12:45,850 --> 00:12:49,638
a survival skill he had honed
over thousands of elevator rides.

239
00:12:49,638 --> 00:12:53,571
He was learning to accommodate the tension

240
00:12:53,571 --> 00:12:56,887
that his skin color generated

241
00:12:58,164 --> 00:13:01,396
and that put his own life at risk.

242
00:13:02,619 --> 00:13:06,187
We know that the brain is wired for bias,

243
00:13:06,187 --> 00:13:09,201
and one way to interrupt that bias

244
00:13:09,559 --> 00:13:13,445
is to pause and to reflect
on the evidence of our assumptions.

245
00:13:13,445 --> 00:13:15,156
So we need to ask ourselves,

246
00:13:15,156 --> 00:13:19,702
what assumptions do we bring

247
00:13:19,702 --> 00:13:21,530
when we step onto an elevator?

248
00:13:21,530 --> 00:13:23,393
Or an airplane?

249
00:13:23,739 --> 00:13:28,422
How do we make ourselves aware
of our own unconscious bias?

250
00:13:28,422 --> 00:13:32,337
Who do those assumptions keep safe?

251
00:13:32,798 --> 00:13:35,557
Who do they put at risk?

252
00:13:35,824 --> 00:13:38,832
Until we ask these questions,

253
00:13:38,832 --> 00:13:43,702
and insist that our schools
and our courts and our police departments

254
00:13:43,702 --> 00:13:47,074
and every institution do the same,

255
00:13:47,074 --> 00:13:51,031
we will continue to allow bias

256
00:13:51,883 --> 00:13:53,179
to blind us,

257
00:13:53,579 --> 00:13:54,988
and if we do,

258
00:13:55,717 --> 00:13:59,274
none of us are truly safe.

259
00:14:00,358 --> 00:14:02,619
Thank you.