1
00:00:00,910 --> 00:00:02,183
Some years ago,

2
00:00:02,207 --> 00:00:07,119
I was on an airplane with my son
who was just five years old at the time.

3
00:00:08,245 --> 00:00:13,340
My son was so excited
about being on this airplane with Mommy.

4
00:00:13,364 --> 00:00:16,304
He's looking all around
and he's checking things out

5
00:00:16,328 --> 00:00:18,164
and he's checking people out.

6
00:00:18,188 --> 00:00:19,818
And he sees this man, and he says,

7
00:00:19,842 --> 00:00:22,703
"Hey! That guy looks like Daddy!"

8
00:00:23,882 --> 00:00:25,802
And I look at the man,

9
00:00:25,826 --> 00:00:29,556
and he didn't look anything
at all like my husband,

10
00:00:29,580 --> 00:00:30,949
nothing at all.

11
00:00:31,461 --> 00:00:33,797
And so then I start
looking around on the plane,

12
00:00:33,821 --> 00:00:39,706
and I notice this man was
the only black guy on the plane.

13
00:00:40,874 --> 00:00:42,286
And I thought,

14
00:00:42,310 --> 00:00:43,504
"Alright.

15
00:00:44,369 --> 00:00:46,894
I'm going to have to have
a little talk with my son

16
00:00:46,918 --> 00:00:49,847
about how not all
black people look alike."

17
00:00:49,871 --> 00:00:54,248
My son, he lifts his head up,
and he says to me,

18
00:00:56,246 --> 00:00:58,617
"I hope he doesn't rob the plane."

19
00:00:59,359 --> 00:01:01,874
And I said, "What? What did you say?"

20
00:01:01,898 --> 00:01:05,326
And he says, "Well, I hope that man
doesn't rob the plane."

21
00:01:07,200 --> 00:01:09,952
And I said, "Well, why would you say that?

22
00:01:10,486 --> 00:01:13,149
You know Daddy wouldn't rob a plane."

23
00:01:13,173 --> 00:01:15,485
And he says, "Yeah, yeah,
yeah, well, I know."

24
00:01:16,179 --> 00:01:18,306
And I said, "Well,
why would you say that?"

25
00:01:20,346 --> 00:01:23,303
And he looked at me
with this really sad face,

26
00:01:24,168 --> 00:01:25,422
and he says,

27
00:01:26,890 --> 00:01:28,996
"I don't know why I said that.

28
00:01:30,600 --> 00:01:32,958
I don't know why I was thinking that."

29
00:01:33,724 --> 00:01:37,242
We are living with such severe
racial stratification

30
00:01:37,266 --> 00:01:42,326
that even a five-year-old can tell us
what's supposed to happen next,

31
00:01:43,990 --> 00:01:46,097
even with no evildoer,

32
00:01:46,121 --> 00:01:48,700
even with no explicit hatred.

33
00:01:50,184 --> 00:01:54,097
This association between
blackness and crime

34
00:01:54,121 --> 00:01:58,455
made its way into the mind
of my five-year-old.

35
00:01:59,787 --> 00:02:03,050
It makes its way into all of our children,

36
00:02:04,201 --> 00:02:05,592
into all of us.

37
00:02:06,793 --> 00:02:10,317
Our minds are shaped by
the racial disparities

38
00:02:10,341 --> 00:02:11,933
we see out in the world

39
00:02:12,752 --> 00:02:18,093
and the narratives that help us
to make sense of the disparities we see:

40
00:02:19,637 --> 00:02:22,163
"Those people are criminal."

41
00:02:22,187 --> 00:02:24,112
"Those people are violent."

42
00:02:24,136 --> 00:02:27,101
"Those people are to be feared."

43
00:02:27,814 --> 00:02:31,005
When my research team
brought people into our lab

44
00:02:31,029 --> 00:02:33,312
and exposed them to faces,

45
00:02:33,336 --> 00:02:40,336
we found that exposure to black faces
led them to see blurry images of guns

46
00:02:40,360 --> 00:02:43,616
with greater clarity and speed.

47
00:02:43,640 --> 00:02:46,994
Bias cannot only control what we see,

48
00:02:47,018 --> 00:02:48,666
but where we look.

49
00:02:48,690 --> 00:02:52,134
We found that prompting people
to think of violent crime

50
00:02:52,158 --> 00:02:56,454
can lead them to direct their eyes
onto a black face

51
00:02:56,478 --> 00:02:58,608
and away from a white face.

52
00:02:58,632 --> 00:03:02,474
Prompting police officers
to think of capturing and shooting

53
00:03:02,498 --> 00:03:03,727
and arresting

54
00:03:03,751 --> 00:03:07,610
leads their eyes to settle
on black faces, too.

55
00:03:07,634 --> 00:03:12,700
Bias can infect every aspect
of our criminal justice system.

56
00:03:13,100 --> 00:03:16,031
In a large data set
of death-eligible defendants,

57
00:03:16,055 --> 00:03:20,412
we found that looking more black
more than doubled their chances

58
00:03:20,436 --> 00:03:22,493
of receiving a death sentence --

59
00:03:23,494 --> 00:03:25,921
at least when their victims were white.

60
00:03:25,945 --> 00:03:27,383
This effect is significant,

61
00:03:27,407 --> 00:03:30,708
even though we controlled
for the severity of the crime

62
00:03:30,732 --> 00:03:33,013
and the defendant's attractiveness.

63
00:03:33,037 --> 00:03:35,686
And no matter what we controlled for,

64
00:03:35,710 --> 00:03:39,055
we found that black
people were punished

65
00:03:39,079 --> 00:03:43,404
in proportion to the blackness
of their physical features:

66
00:03:43,428 --> 00:03:45,309
the more black,

67
00:03:45,333 --> 00:03:47,086
the more death-worthy.

68
00:03:47,110 --> 00:03:51,319
Bias can also influence
how teachers discipline students.

69
00:03:51,781 --> 00:03:56,188
My colleagues and I have found
that teachers express a desire

70
00:03:56,212 --> 00:03:59,778
to discipline a black
middle school student more harshly

71
00:03:59,802 --> 00:04:00,970
than a white student

72
00:04:00,994 --> 00:04:03,570
for the same repeated infractions.

73
00:04:03,594 --> 00:04:04,888
In a recent study,

74
00:04:04,912 --> 00:04:09,270
we're finding that teachers
treat black students as a group

75
00:04:09,294 --> 00:04:11,725
but white students as individuals.

76
00:04:12,126 --> 00:04:15,725
If, for example,
one black student misbehaves

77
00:04:15,749 --> 00:04:20,534
and then a different black student
misbehaves a few days later,

78
00:04:20,558 --> 00:04:23,786
the teacher responds
to that second black student

79
00:04:23,810 --> 00:04:26,435
as if he had misbehaved twice.

80
00:04:26,952 --> 00:04:29,763
It's as though the sins of one child

81
00:04:29,787 --> 00:04:31,963
get piled onto the other.

82
00:04:31,987 --> 00:04:35,281
We create categories
to make sense of the world,

83
00:04:35,305 --> 00:04:39,788
to assert some control and coherence

84
00:04:39,812 --> 00:04:43,902
to the stimuli that we're constantly
being bombarded with.

85
00:04:43,926 --> 00:04:47,894
Categorization and the bias that it seeds

86
00:04:47,918 --> 00:04:52,940
allow our brains to make judgments
more quickly and efficiently,

87
00:04:52,964 --> 00:04:56,366
and we do this by instinctively
relying on patterns

88
00:04:56,390 --> 00:04:58,059
that seem predictable.

89
00:04:58,083 --> 00:05:04,026
Yet, just as the categories we create
allow us to make quick decisions,

90
00:05:04,050 --> 00:05:06,552
they also reinforce bias.

91
00:05:06,576 --> 00:05:09,968
So the very things that help us
to see the world

92
00:05:11,104 --> 00:05:13,084
also can blind us to it.

93
00:05:13,509 --> 00:05:16,287
They render our choices effortless,

94
00:05:16,311 --> 00:05:17,680
friction-free.

95
00:05:18,956 --> 00:05:21,401
Yet they exact a heavy toll.

96
00:05:22,158 --> 00:05:23,569
So what can we do?

97
00:05:24,507 --> 00:05:26,998
We are all vulnerable to bias,

98
00:05:27,022 --> 00:05:29,702
but we don't act on bias all the time.

99
00:05:29,726 --> 00:05:33,370
There are certain conditions
that can bring bias alive

100
00:05:33,394 --> 00:05:35,927
and other conditions that can muffle it.

101
00:05:35,951 --> 00:05:37,798
Let me give you an example.

102
00:05:38,663 --> 00:05:43,223
Many people are familiar
with the tech company Nextdoor.

103
00:05:44,073 --> 00:05:50,526
So, their whole purpose is to create
stronger, healthier, safer neighborhoods.

104
00:05:51,468 --> 00:05:54,389
And so they offer this online space

105
00:05:54,413 --> 00:05:57,562
where neighbors can gather
and share information.

106
00:05:57,586 --> 00:06:01,712
Yet, Nextdoor soon found
that they had a problem

107
00:06:01,736 --> 00:06:03,404
with racial profiling.

108
00:06:04,012 --> 00:06:05,979
In the typical case,

109
00:06:06,003 --> 00:06:08,399
people would look outside their window

110
00:06:08,423 --> 00:06:12,472
and see a black man
in their otherwise white neighborhood

111
00:06:12,496 --> 00:06:17,211
and make the snap judgment
that he was up to no good,

112
00:06:17,235 --> 00:06:20,586
even when there was no evidence
of criminal wrongdoing.

113
00:06:20,610 --> 00:06:23,544
In many ways, how we behave online

114
00:06:23,568 --> 00:06:26,682
is a reflection of how
we behave in the world.

115
00:06:27,117 --> 00:06:31,062
But what we don't want to do
is create an easy-to-use system

116
00:06:31,086 --> 00:06:35,249
that can amplify bias
and deepen racial disparities,

117
00:06:36,129 --> 00:06:38,395
rather than dismantling them.

118
00:06:38,863 --> 00:06:42,292
So the cofounder of Nextdoor
reached out to me and to others

119
00:06:42,316 --> 00:06:44,447
to try to figure out what to do.

120
00:06:44,471 --> 00:06:48,417
And they realized that
to curb racial profiling on the platform,

121
00:06:48,441 --> 00:06:50,363
they were going to have to add friction;

122
00:06:50,387 --> 00:06:53,045
that is, they were going
to have to slow people down.

123
00:06:53,069 --> 00:06:55,264
So Nextdoor had a choice to make,

124
00:06:55,288 --> 00:06:57,766
and against every impulse,

125
00:06:57,790 --> 00:06:59,906
they decided to add friction.

126
00:07:00,397 --> 00:07:03,837
And they did this by adding
a simple checklist.

127
00:07:03,861 --> 00:07:05,531
There were three items on it.

128
00:07:06,111 --> 00:07:09,052
First, they asked users to pause

129
00:07:09,076 --> 00:07:14,193
and think, "What was this person doing
that made him suspicious?"

130
00:07:14,876 --> 00:07:19,409
The category "black man"
is not grounds for suspicion.

131
00:07:19,433 --> 00:07:24,572
Second, they asked users to describe
the person's physical features,

132
00:07:24,596 --> 00:07:27,031
not simply their race and gender.

133
00:07:27,642 --> 00:07:31,025
Third, they realized that a lot of people

134
00:07:31,049 --> 00:07:33,977
didn't seem to know
what racial profiling was,

135
00:07:34,001 --> 00:07:35,960
nor that they were engaging in it.

136
00:07:36,462 --> 00:07:39,656
So Nextdoor provided them
with a definition

137
00:07:39,680 --> 00:07:42,582
and told them that it was
strictly prohibited.

138
00:07:43,071 --> 00:07:45,683
Most of you have seen
those signs in airports

139
00:07:45,707 --> 00:07:49,409
and in metro stations,
"If you see something, say something."

140
00:07:49,928 --> 00:07:52,742
Nextdoor tried modifying this.

141
00:07:53,584 --> 00:07:56,156
"If you see something suspicious,

142
00:07:56,180 --> 00:07:58,253
say something specific."

143
00:07:59,491 --> 00:08:03,937
And using this strategy,
by simply slowing people down,

144
00:08:03,961 --> 00:08:09,652
Nextdoor was able to curb
racial profiling by 75 percent.

145
00:08:10,496 --> 00:08:12,586
Now, people often will say to me,

146
00:08:12,610 --> 00:08:17,323
"You can't add friction
in every situation, in every context,

147
00:08:17,347 --> 00:08:21,993
and especially for people who make
split-second decisions all the time."

148
00:08:22,730 --> 00:08:25,293
But it turns out we can add friction

149
00:08:25,317 --> 00:08:27,593
to more situations than we think.

150
00:08:28,031 --> 00:08:30,105
Working with the Oakland Police Department

151
00:08:30,129 --> 00:08:31,546
in California,

152
00:08:31,570 --> 00:08:35,426
I and a number of my colleagues
were able to help the department

153
00:08:35,450 --> 00:08:38,121
to reduce the number of stops they made

154
00:08:38,145 --> 00:08:41,745
of people who were not
committing any serious crimes.

155
00:08:41,769 --> 00:08:44,134
And we did this by pushing officers

156
00:08:44,158 --> 00:08:48,601
to ask themselves a question
before each and every stop they made:

157
00:08:49,451 --> 00:08:52,466
"Is this stop intelligence-led,

158
00:08:52,490 --> 00:08:53,941
yes or no?"

159
00:08:55,353 --> 00:08:56,749
In other words,

160
00:08:57,621 --> 00:09:02,105
do I have prior information
to tie this particular person

161
00:09:02,129 --> 00:09:03,730
to a specific crime?

162
00:09:04,587 --> 00:09:06,045
By adding that question

163
00:09:06,069 --> 00:09:09,148
to the form officers complete
during a stop,

164
00:09:09,172 --> 00:09:10,981
they slow down, they pause,

165
00:09:11,005 --> 00:09:15,225
they think, "Why am I considering
pulling this person over?"

166
00:09:16,721 --> 00:09:22,282
In 2017, before we added that
intelligence-led question to the form,

167
00:09:23,655 --> 00:09:27,601
officers made about 32,000 stops
across the city.

168
00:09:27,625 --> 00:09:31,740
In that next year,
with the addition of this question,

169
00:09:31,764 --> 00:09:34,208
that fell to 19,000 stops.

170
00:09:34,232 --> 00:09:39,193
African-American stops alone
fell by 43 percent.

171
00:09:39,905 --> 00:09:44,343
And stopping fewer black people
did not make the city any more dangerous.

172
00:09:44,367 --> 00:09:47,101
In fact, the crime rate continued to fall,

173
00:09:47,125 --> 00:09:50,462
and the city became safer for everybody.

174
00:09:50,486 --> 00:09:55,841
So one solution can come from reducing
the number of unnecessary stops.

175
00:09:56,285 --> 00:10:00,555
Another can come from improving
the quality of the stops

176
00:10:00,579 --> 00:10:01,884
officers do make.

177
00:10:02,512 --> 00:10:05,108
And technology can help us here.

178
00:10:05,132 --> 00:10:07,547
We all know about George Floyd's death,

179
00:10:08,499 --> 00:10:13,271
because those who tried to come to his aid
held cell phone cameras

180
00:10:13,295 --> 00:10:18,726
to record that horrific, fatal
encounter with the police.

181
00:10:18,750 --> 00:10:23,781
But we have all sorts of technology
that we're not putting to good use.

182
00:10:23,805 --> 00:10:26,308
Police departments across the country

183
00:10:26,332 --> 00:10:29,885
are now required to wear body-worn cameras

184
00:10:29,909 --> 00:10:35,839
so we have recordings of not only
the most extreme and horrific encounters

185
00:10:35,863 --> 00:10:38,617
but of everyday interactions.

186
00:10:38,641 --> 00:10:41,418
With an interdisciplinary
team at Stanford,

187
00:10:41,442 --> 00:10:44,129
we've begun to use
machine learning techniques

188
00:10:44,153 --> 00:10:47,520
to analyze large numbers of encounters.

189
00:10:47,544 --> 00:10:52,155
This is to better understand
what happens in routine traffic stops.

190
00:10:52,179 --> 00:10:54,334
What we found was that

191
00:10:54,358 --> 00:10:58,020
even when police officers
are behaving professionally,

192
00:10:58,860 --> 00:11:03,322
they speak to black drivers
less respectfully than white drivers.

193
00:11:04,052 --> 00:11:08,127
In fact, from the words
officers use alone,

194
00:11:08,151 --> 00:11:13,313
we could predict whether they were talking
to a black driver or a white driver.

195
00:11:13,337 --> 00:11:19,099
The problem is that the vast majority
of the footage from these cameras

196
00:11:19,123 --> 00:11:21,210
is not used by police departments

197
00:11:21,234 --> 00:11:23,510
to understand what's
going on on the street

198
00:11:23,534 --> 00:11:25,777
or to train officers.

199
00:11:26,554 --> 00:11:28,012
And that's a shame.

200
00:11:28,796 --> 00:11:33,585
How does a routine stop
turn into a deadly encounter?

201
00:11:33,609 --> 00:11:36,279
How did this happen
in George Floyd's case?

202
00:11:37,588 --> 00:11:39,670
How did it happen in others?

203
00:11:39,694 --> 00:11:43,090
When my eldest son was 16 years old,

204
00:11:43,114 --> 00:11:46,253
he discovered that
when white people look at him,

205
00:11:46,277 --> 00:11:47,838
they feel fear.

206
00:11:49,123 --> 00:11:51,784
Elevators are the worst, he said.

207
00:11:52,313 --> 00:11:54,644
When those doors close,

208
00:11:54,668 --> 00:11:57,751
people are trapped in this tiny space

209
00:11:57,775 --> 00:12:02,242
with someone they have been taught
to associate with danger.

210
00:12:02,744 --> 00:12:05,964
My son senses their discomfort,

211
00:12:05,988 --> 00:12:09,145
and he smiles to put them at ease,

212
00:12:09,169 --> 00:12:10,938
to calm their fears.

213
00:12:11,351 --> 00:12:13,296
When he speaks,

214
00:12:13,320 --> 00:12:15,003
their bodies relax.

215
00:12:15,442 --> 00:12:17,345
They breathe easier.

216
00:12:17,369 --> 00:12:19,900
They take pleasure in his cadence,

217
00:12:19,924 --> 00:12:22,241
his diction, his word choice.

218
00:12:22,986 --> 00:12:24,829
He sounds like one of them.

219
00:12:24,853 --> 00:12:29,583
I used to think that my son
was a natural extrovert like his father.

220
00:12:29,607 --> 00:12:33,157
But I realized at that moment,
in that conversation,

221
00:12:34,143 --> 00:12:39,221
that his smile was not a sign
that he wanted to connect

222
00:12:39,245 --> 00:12:41,209
with would-be strangers.

223
00:12:41,920 --> 00:12:45,572
It was a talisman he used
to protect himself,

224
00:12:45,596 --> 00:12:51,815
a survival skill he had honed
over thousands of elevator rides.

225
00:12:52,387 --> 00:12:57,558
He was learning to accommodate the tension
that his skin color generated

226
00:12:59,026 --> 00:13:01,693
and that put his own life at risk.

227
00:13:02,619 --> 00:13:06,402
We know that the brain is wired for bias,

228
00:13:06,426 --> 00:13:10,891
and one way to interrupt that bias
is to pause and to reflect

229
00:13:10,915 --> 00:13:13,220
on the evidence of our assumptions.

230
00:13:13,244 --> 00:13:14,999
So we need to ask ourselves:

231
00:13:15,023 --> 00:13:19,688
What assumptions do we bring
when we step onto an elevator?

232
00:13:21,776 --> 00:13:23,087
Or an airplane?

233
00:13:23,532 --> 00:13:28,131
How do we make ourselves aware
of our own unconscious bias?

234
00:13:28,155 --> 00:13:30,506
Who do those assumptions keep safe?

235
00:13:32,615 --> 00:13:34,547
Who do they put at risk?

236
00:13:35,649 --> 00:13:38,003
Until we ask these questions

237
00:13:38,978 --> 00:13:43,602
and insist that our schools
and our courts and our police departments

238
00:13:43,626 --> 00:13:46,168
and every institution do the same,

239
00:13:47,835 --> 00:13:51,664
we will continue to allow bias

240
00:13:51,688 --> 00:13:52,966
to blind us.

241
00:13:53,348 --> 00:13:54,757
And if we do,

242
00:13:56,066 --> 00:13:59,274
none of us are truly safe.

243
00:14:02,103 --> 00:14:03,411
Thank you.