字幕test The human insights missing from big data

东郭翰音
2023-12-01

1
00:00:12,705 --> 00:00:14,250
In ancient Greece,

2
00:00:15,256 --> 00:00:19,199
when anyone from slaves to soldiers,
poets and politicians,

3
00:00:19,223 --> 00:00:23,227
needed to make a big decision
on life's most important questions,

4
00:00:23,251 --> 00:00:24,642
like, "Should I get married?"

5
00:00:24,666 --> 00:00:26,523
or "Should we embark on this voyage?"

6
00:00:26,547 --> 00:00:29,475
or "Should our army
advance into this territory?"

7
00:00:29,499 --> 00:00:32,078
they all consulted the oracle.

8
00:00:32,840 --> 00:00:34,280
So this is how it worked:

9
00:00:34,304 --> 00:00:37,416
you would bring her a question
and you would get on your knees,

10
00:00:37,440 --> 00:00:39,311
and then she would go into this trance.

11
00:00:39,335 --> 00:00:40,884
It would take a couple of days,

12
00:00:40,908 --> 00:00:43,071
and then eventually
she would come out of it,

13
00:00:43,095 --> 00:00:45,631
giving you her predictions as your answer.

14
00:00:46,730 --> 00:00:49,296
From the oracle bones of ancient China

15
00:00:49,320 --> 00:00:51,665
to ancient Greece to Mayan calendars,

16
00:00:51,689 --> 00:00:53,985
people have craved for prophecy

17
00:00:54,009 --> 00:00:57,146
in order to find out
what's going to happen next.

18
00:00:58,336 --> 00:01:01,575
And that's because we all want
to make the right decision.

19
00:01:01,599 --> 00:01:03,144
We don't want to miss something.

20
00:01:03,712 --> 00:01:05,455
The future is scary,

21
00:01:05,479 --> 00:01:08,196
so it's much nicer
knowing that we can make a decision

22
00:01:08,220 --> 00:01:10,202
with some assurance of the outcome.

23
00:01:10,899 --> 00:01:12,510
Well, we have a new oracle,

24
00:01:12,534 --> 00:01:14,679
and it's name is big data,

25
00:01:14,703 --> 00:01:18,642
or we call it "Watson"
or "deep learning" or "neural net."

26
00:01:19,160 --> 00:01:23,172
And these are the kinds of questions
we ask of our oracle now,

27
00:01:23,196 --> 00:01:27,118
like, "What's the most efficient way
to ship these phones

28
00:01:27,142 --> 00:01:28,965
from China to Sweden?"

29
00:01:28,989 --> 00:01:30,789
Or, "What are the odds

30
00:01:30,813 --> 00:01:34,176
of my child being born
with a genetic disorder?"

31
00:01:34,772 --> 00:01:38,016
Or, "What are the sales volume
we can predict for this product?"

32
00:01:39,928 --> 00:01:43,975
I have a dog. Her name is Elle,
and she hates the rain.

33
00:01:43,999 --> 00:01:47,305
And I have tried everything
to untrain her.

34
00:01:47,329 --> 00:01:50,100
But because I have failed at this,

35
00:01:50,124 --> 00:01:53,410
I also have to consult
an oracle, called Dark Sky,

36
00:01:53,434 --> 00:01:55,069
every time before we go on a walk,

37
00:01:55,093 --> 00:01:58,670
for very accurate weather predictions
in the next 10 minutes.

38
00:02:01,355 --> 00:02:02,658
She's so sweet.

39
00:02:03,647 --> 00:02:09,354
So because of all of this,
our oracle is a $122 billion industry.

40
00:02:09,826 --> 00:02:13,202
Now, despite the size of this industry,

41
00:02:13,226 --> 00:02:15,682
the returns are surprisingly low.

42
00:02:16,162 --> 00:02:18,656
Investing in big data is easy,

43
00:02:18,680 --> 00:02:20,613
but using it is hard.

44
00:02:21,801 --> 00:02:25,841
Over 73 percent of big data projects
aren't even profitable,

45
00:02:25,865 --> 00:02:28,296
and I have executives
coming up to me saying,

46
00:02:28,320 --> 00:02:30,109
"We're experiencing the same thing.

47
00:02:30,133 --> 00:02:31,886
We invested in some big data system,

48
00:02:31,910 --> 00:02:34,878
and our employees aren't making
better decisions.

49
00:02:34,902 --> 00:02:38,064
And they're certainly not coming up
with more breakthrough ideas."

50
00:02:38,734 --> 00:02:41,918
So this is all really interesting to me,

51
00:02:41,942 --> 00:02:43,952
because I'm a technology ethnographer.

52
00:02:44,450 --> 00:02:47,014
I study and I advise companies

53
00:02:47,038 --> 00:02:49,521
on the patterns
of how people use technology,

54
00:02:49,545 --> 00:02:52,223
and one of my interest areas is data.

55
00:02:52,247 --> 00:02:57,440
So why is having more data
not helping us make better decisions,

56
00:02:57,464 --> 00:03:00,247
especially for companies
who have all these resources

57
00:03:00,271 --> 00:03:02,007
to invest in these big data systems?

58
00:03:02,031 --> 00:03:04,429
Why isn't it getting any easier for them?

59
00:03:05,810 --> 00:03:08,444
So, I've witnessed the struggle firsthand.

60
00:03:09,194 --> 00:03:12,678
In 2009, I started
a research position with Nokia.

61
00:03:13,052 --> 00:03:14,210
And at the time,

62
00:03:14,234 --> 00:03:17,392
Nokia was one of the largest
cell phone companies in the world,

63
00:03:17,416 --> 00:03:20,618
dominating emerging markets
like China, Mexico and India --

64
00:03:20,642 --> 00:03:23,144
all places where I had done
a lot of research

65
00:03:23,168 --> 00:03:25,844
on how low-income people use technology.

66
00:03:25,868 --> 00:03:28,198
And I spent a lot of extra time in China

67
00:03:28,222 --> 00:03:30,814
getting to know the informal economy.

68
00:03:30,838 --> 00:03:33,239
So I did things like working
as a street vendor

69
00:03:33,263 --> 00:03:35,837
selling dumplings to construction workers.

70
00:03:35,861 --> 00:03:37,219
Or I did fieldwork,

71
00:03:37,243 --> 00:03:40,201
spending nights and days
in internet cafés,

72
00:03:40,225 --> 00:03:42,771
hanging out with Chinese youth,
so I could understand

73
00:03:42,795 --> 00:03:45,079
how they were using
games and mobile phones

74
00:03:45,103 --> 00:03:48,473
and using it between moving
from the rural areas to the cities.

75
00:03:50,155 --> 00:03:54,082
Through all of this qualitative evidence
that I was gathering,

76
00:03:54,106 --> 00:03:56,930
I was starting to see so clearly

77
00:03:56,954 --> 00:04:01,426
that a big change was about to happen
among low-income Chinese people.

78
00:04:02,840 --> 00:04:07,207
Even though they were surrounded
by advertisements for luxury products

79
00:04:07,231 --> 00:04:10,726
like fancy toilets --
who wouldn't want one? --

80
00:04:10,750 --> 00:04:13,640
and apartments and cars,

81
00:04:13,664 --> 00:04:15,484
through my conversations with them,

82
00:04:15,508 --> 00:04:19,349
I found out that the ads
the actually enticed them the most

83
00:04:19,373 --> 00:04:21,369
were the ones for iPhones,

84
00:04:21,393 --> 00:04:24,445
promising them this entry
into this high-tech life.

85
00:04:25,289 --> 00:04:28,452
And even when I was living with them
in urban slums like this one,

86
00:04:28,476 --> 00:04:31,472
I saw people investing
over half of their monthly income

87
00:04:31,496 --> 00:04:33,119
into buying a phone,

88
00:04:33,143 --> 00:04:35,445
and increasingly, they were "shanzhai,"

89
00:04:35,469 --> 00:04:38,857
which are affordable knock-offs
of iPhones and other brands.

90
00:04:40,123 --> 00:04:41,748
They're very usable.

91
00:04:42,710 --> 00:04:44,032
Does the job.

92
00:04:44,570 --> 00:04:50,359
And after years of living
with migrants and working with them

93
00:04:50,383 --> 00:04:53,817
and just really doing everything
that they were doing,

94
00:04:53,841 --> 00:04:57,438
I started piecing
all these data points together --

95
00:04:57,462 --> 00:05:00,585
from the things that seem random,
like me selling dumplings,

96
00:05:00,609 --> 00:05:02,413
to the things that were more obvious,

97
00:05:02,437 --> 00:05:05,669
like tracking how much they were spending
on their cell phone bills.

98
00:05:05,693 --> 00:05:08,332
And I was able to create
this much more holistic picture

99
00:05:08,356 --> 00:05:09,512
of what was happening.

100
00:05:09,536 --> 00:05:11,258
And that's when I started to realize

101
00:05:11,282 --> 00:05:14,791
that even the poorest in China
would want a smartphone,

102
00:05:14,815 --> 00:05:19,800
and that they would do almost anything
to get their hands on one.

103
00:05:20,893 --> 00:05:23,297
You have to keep in mind,

104
00:05:23,321 --> 00:05:26,405
iPhones had just come out, it was 2009,

105
00:05:26,429 --> 00:05:28,314
so this was, like, eight years ago,

106
00:05:28,338 --> 00:05:30,775
and Androids had just started
looking like iPhones.

107
00:05:30,799 --> 00:05:33,306
And a lot of very smart
and realistic people said,

108
00:05:33,330 --> 00:05:35,537
"Those smartphones -- that's just a fad.

109
00:05:36,063 --> 00:05:39,059
Who wants to carry around
these heavy things

110
00:05:39,083 --> 00:05:42,570
where batteries drain quickly
and they break every time you drop them?"

111
00:05:44,613 --> 00:05:45,814
But I had a lot of data,

112
00:05:45,838 --> 00:05:48,098
and I was very confident
about my insights,

113
00:05:48,122 --> 00:05:50,951
so I was very excited
to share them with Nokia.

114
00:05:53,152 --> 00:05:55,669
But Nokia was not convinced,

115
00:05:55,693 --> 00:05:58,028
because it wasn't big data.

116
00:05:58,842 --> 00:06:01,246
They said, "We have
millions of data points,

117
00:06:01,270 --> 00:06:05,517
and we don't see any indicators
of anyone wanting to buy a smartphone,

118
00:06:05,541 --> 00:06:09,929
and your data set of 100,
as diverse as it is, is too weak

119
00:06:09,953 --> 00:06:11,667
for us to even take seriously."

120
00:06:12,728 --> 00:06:14,333
And I said, "Nokia, you're right.

121
00:06:14,357 --> 00:06:15,917
Of course you wouldn't see this,

122
00:06:15,941 --> 00:06:19,312
because you're sending out surveys
assuming that people don't know

123
00:06:19,336 --> 00:06:20,495
what a smartphone is,

124
00:06:20,519 --> 00:06:22,885
so of course you're not going
to get any data back

125
00:06:22,909 --> 00:06:25,481
about people wanting to buy
a smartphone in two years.

126
00:06:25,505 --> 00:06:27,623
Your surveys, your methods
have been designed

127
00:06:27,647 --> 00:06:29,669
to optimize an existing business model,

128
00:06:29,693 --> 00:06:32,301
and I'm looking
at these emergent human dynamics

129
00:06:32,325 --> 00:06:33,679
that haven't happened yet.

130
00:06:33,703 --> 00:06:36,141
We're looking outside of market dynamics

131
00:06:36,165 --> 00:06:37,796
so that we can get ahead of it."

132
00:06:39,193 --> 00:06:41,437
Well, you know what happened to Nokia?

133
00:06:41,461 --> 00:06:43,826
Their business fell off a cliff.

134
00:06:44,611 --> 00:06:48,338
This -- this is the cost
of missing something.

135
00:06:48,983 --> 00:06:50,982
It was unfathomable.

136
00:06:51,823 --> 00:06:53,474
But Nokia's not alone.

137
00:06:54,078 --> 00:06:56,659
I see organizations
throwing out data all the time

138
00:06:56,683 --> 00:06:59,244
because it didn't come from a quant model

139
00:06:59,268 --> 00:07:01,036
or it doesn't fit in one.

140
00:07:02,039 --> 00:07:04,087
But it's not big data's fault.

141
00:07:04,762 --> 00:07:08,669
It's the way we use big data;
it's our responsibility.

142
00:07:09,550 --> 00:07:11,461
Big data's reputation for success

143
00:07:11,485 --> 00:07:15,244
comes from quantifying
very specific environments,

144
00:07:15,268 --> 00:07:20,181
like electricity power grids
or delivery logistics or genetic code,

145
00:07:20,205 --> 00:07:24,523
when we're quantifying in systems
that are more or less contained.

146
00:07:24,547 --> 00:07:27,516
But not all systems
are as neatly contained.

147
00:07:27,540 --> 00:07:30,798
When you're quantifying
and systems are more dynamic,

148
00:07:30,822 --> 00:07:34,621
especially systems
that involve human beings,

149
00:07:34,645 --> 00:07:37,071
forces are complex and unpredictable,

150
00:07:37,095 --> 00:07:40,581
and these are things
that we don't know how to model so well.

151
00:07:41,024 --> 00:07:43,837
Once you predict something
about human behavior,

152
00:07:43,861 --> 00:07:45,716
new factors emerge,

153
00:07:45,740 --> 00:07:48,105
because conditions
are constantly changing.

154
00:07:48,129 --> 00:07:49,932
That's why it's a never-ending cycle.

155
00:07:49,956 --> 00:07:51,420
You think you know something,

156
00:07:51,444 --> 00:07:53,686
and then something unknown
enters the picture.

157
00:07:53,710 --> 00:07:57,032
And that's why just relying
on big data alone

158
00:07:57,056 --> 00:07:59,905
increases the chance
that we'll miss something,

159
00:07:59,929 --> 00:08:03,706
while giving us this illusion
that we already know everything.

160
00:08:04,226 --> 00:08:08,082
And what makes it really hard
to see this paradox

161
00:08:08,106 --> 00:08:10,765
and even wrap our brains around it

162
00:08:10,789 --> 00:08:14,480
is that we have this thing
that I call the quantification bias,

163
00:08:14,504 --> 00:08:18,426
which is the unconscious belief
of valuing the measurable

164
00:08:18,450 --> 00:08:20,044
over the immeasurable.

165
00:08:21,042 --> 00:08:24,326
And we often experience this at our work.

166
00:08:24,350 --> 00:08:27,000
Maybe we work alongside
colleagues who are like this,

167
00:08:27,024 --> 00:08:29,452
or even our whole entire
company may be like this,

168
00:08:29,476 --> 00:08:32,022
where people become
so fixated on that number,

169
00:08:32,046 --> 00:08:34,113
that they can't see anything
outside of it,

170
00:08:34,137 --> 00:08:38,085
even when you present them evidence
right in front of their face.

171
00:08:38,943 --> 00:08:42,314
And this is a very appealing message,

172
00:08:42,338 --> 00:08:44,681
because there's nothing
wrong with quantifying;

173
00:08:44,705 --> 00:08:46,135
it's actually very satisfying.

174
00:08:46,159 --> 00:08:50,521
I get a great sense of comfort
from looking at an Excel spreadsheet,

175
00:08:50,545 --> 00:08:51,946
even very simple ones.

176
00:08:51,970 --> 00:08:52,984
(Laughter)

177
00:08:53,008 --> 00:08:54,160
It's just kind of like,

178
00:08:54,184 --> 00:08:57,688
"Yes! The formula worked. It's all OK.
Everything is under control."

179
00:08:58,612 --> 00:09:01,002
But the problem is

180
00:09:01,026 --> 00:09:03,687
that quantifying is addictive.

181
00:09:03,711 --> 00:09:05,093
And when we forget that

182
00:09:05,117 --> 00:09:08,155
and when we don't have something
to kind of keep that in check,

183
00:09:08,179 --> 00:09:10,297
it's very easy to just throw out data

184
00:09:10,321 --> 00:09:13,039
because it can't be expressed
as a numerical value.

185
00:09:13,063 --> 00:09:15,984
It's very easy just to slip
into silver-bullet thinking,

186
00:09:16,008 --> 00:09:18,587
as if some simple solution existed.

187
00:09:19,420 --> 00:09:23,482
Because this is a great moment of danger
for any organization,

188
00:09:23,506 --> 00:09:26,140
because oftentimes,
the future we need to predict --

189
00:09:26,164 --> 00:09:28,330
it isn't in that haystack,

190
00:09:28,354 --> 00:09:30,892
but it's that tornado
that's bearing down on us

191
00:09:30,916 --> 00:09:32,404
outside of the barn.

192
00:09:34,780 --> 00:09:37,106
There is no greater risk

193
00:09:37,130 --> 00:09:38,796
than being blind to the unknown.

194
00:09:38,820 --> 00:09:40,969
It can cause you to make
the wrong decisions.

195
00:09:40,993 --> 00:09:42,967
It can cause you to miss something big.

196
00:09:43,554 --> 00:09:46,655
But we don't have to go down this path.

197
00:09:47,273 --> 00:09:50,468
It turns out that the oracle
of ancient Greece

198
00:09:50,492 --> 00:09:54,458
holds the secret key
that shows us the path forward.

199
00:09:55,474 --> 00:09:58,069
Now, recent geological research has shown

200
00:09:58,093 --> 00:10:01,657
that the Temple of Apollo,
where the most famous oracle sat,

201
00:10:01,681 --> 00:10:04,765
was actually built
over two earthquake faults.

202
00:10:04,789 --> 00:10:07,675
And these faults would release
these petrochemical fumes

203
00:10:07,699 --> 00:10:09,384
from underneath the Earth's crust,

204
00:10:09,408 --> 00:10:13,274
and the oracle literally sat
right above these faults,

205
00:10:13,298 --> 00:10:16,886
inhaling enormous amounts
of ethylene gas, these fissures.

206
00:10:16,910 --> 00:10:17,918
(Laughter)

207
00:10:17,942 --> 00:10:19,115
It's true.

208
00:10:19,139 --> 00:10:20,156
(Laughter)

209
00:10:20,180 --> 00:10:23,689
It's all true, and that's what made her
babble and hallucinate

210
00:10:23,713 --> 00:10:25,437
and go into this trance-like state.

211
00:10:25,461 --> 00:10:27,231
She was high as a kite!

212
00:10:27,255 --> 00:10:31,716
(Laughter)

213
00:10:31,740 --> 00:10:34,519
So how did anyone --

214
00:10:34,543 --> 00:10:37,573
How did anyone get
any useful advice out of her

215
00:10:37,597 --> 00:10:38,787
in this state?

216
00:10:39,317 --> 00:10:41,698
Well, you see those people
surrounding the oracle?

217
00:10:41,722 --> 00:10:43,601
You see those people holding her up,

218
00:10:43,625 --> 00:10:45,342
because she's, like, a little woozy?

219
00:10:45,366 --> 00:10:47,674
And you see that guy
on your left-hand side

220
00:10:47,698 --> 00:10:49,296
holding the orange notebook?

221
00:10:49,925 --> 00:10:51,655
Well, those were the temple guides,

222
00:10:51,679 --> 00:10:54,695
and they worked hand in hand
with the oracle.

223
00:10:55,904 --> 00:10:58,420
When inquisitors would come
and get on their knees,

224
00:10:58,444 --> 00:11:00,784
that's when the temple guides
would get to work,

225
00:11:00,808 --> 00:11:02,672
because after they asked her questions,

226
00:11:02,696 --> 00:11:04,697
they would observe their emotional state,

227
00:11:04,721 --> 00:11:07,045
and then they would ask them
follow-up questions,

228
00:11:07,069 --> 00:11:09,903
like, "Why do you want to know
this prophecy? Who are you?

229
00:11:09,927 --> 00:11:12,191
What are you going to do
with this information?"

230
00:11:12,215 --> 00:11:15,397
And then the temple guides would take
this more ethnographic,

231
00:11:15,421 --> 00:11:17,577
this more qualitative information,

232
00:11:17,601 --> 00:11:19,676
and interpret the oracle's babblings.

233
00:11:21,248 --> 00:11:23,540
So the oracle didn't stand alone,

234
00:11:23,564 --> 00:11:25,712
and neither should our big data systems.

235
00:11:26,450 --> 00:11:27,611
Now to be clear,

236
00:11:27,635 --> 00:11:31,094
I'm not saying that big data systems
are huffing ethylene gas,

237
00:11:31,118 --> 00:11:33,471
or that they're even giving
invalid predictions.

238
00:11:33,495 --> 00:11:34,656
The total opposite.

239
00:11:34,680 --> 00:11:36,748
But what I am saying

240
00:11:36,772 --> 00:11:40,604
is that in the same way
that the oracle needed her temple guides,

241
00:11:40,628 --> 00:11:42,916
our big data systems need them, too.

242
00:11:42,940 --> 00:11:47,049
They need people like ethnographers
and user researchers

243
00:11:47,073 --> 00:11:49,579
who can gather what I call thick data.

244
00:11:50,322 --> 00:11:53,313
This is precious data from humans,

245
00:11:53,337 --> 00:11:57,439
like stories, emotions and interactions
that cannot be quantified.

246
00:11:57,463 --> 00:11:59,785
It's the kind of data
that I collected for Nokia

247
00:11:59,809 --> 00:12:02,478
that comes in in the form
of a very small sample size,

248
00:12:02,502 --> 00:12:05,457
but delivers incredible depth of meaning.

249
00:12:05,481 --> 00:12:09,161
And what makes it so thick and meaty

250
00:12:10,265 --> 00:12:14,294
is the experience of understanding
the human narrative.

251
00:12:14,318 --> 00:12:17,957
And that's what helps to see
what's missing in our models.

252
00:12:18,671 --> 00:12:22,716
Thick data grounds our business questions
in human questions,

253
00:12:22,740 --> 00:12:26,302
and that's why integrating
big and thick data

254
00:12:26,326 --> 00:12:28,015
forms a more complete picture.

255
00:12:28,592 --> 00:12:31,473
Big data is able to offer
insights at scale

256
00:12:31,497 --> 00:12:34,144
and leverage the best
of machine intelligence,

257
00:12:34,168 --> 00:12:37,740
whereas thick data can help us
rescue the context loss

258
00:12:37,764 --> 00:12:39,862
that comes from making big data usable,

259
00:12:39,886 --> 00:12:42,067
and leverage the best
of human intelligence.

260
00:12:42,091 --> 00:12:45,643
And when you actually integrate the two,
that's when things get really fun,

261
00:12:45,667 --> 00:12:48,103
because then you're no longer
just working with data

262
00:12:48,127 --> 00:12:49,323
you've already collected.

263
00:12:49,347 --> 00:12:52,084
You get to also work with data
that hasn't been collected.

264
00:12:52,108 --> 00:12:53,827
You get to ask questions about why:

265
00:12:53,851 --> 00:12:55,168
Why is this happening?

266
00:12:55,598 --> 00:12:56,977
Now, when Netflix did this,

267
00:12:57,001 --> 00:13:00,036
they unlocked a whole new way
to transform their business.

268
00:13:01,226 --> 00:13:05,182
Netflix is known for their really great
recommendation algorithm,

269
00:13:05,206 --> 00:13:10,003
and they had this $1 million prize
for anyone who could improve it.

270
00:13:10,027 --> 00:13:11,341
And there were winners.

271
00:13:12,075 --> 00:13:16,398
But Netflix discovered
the improvements were only incremental.

272
00:13:17,224 --> 00:13:19,188
So to really find out what was going on,

273
00:13:19,212 --> 00:13:22,953
they hired an ethnographer,
Grant McCracken,

274
00:13:22,977 --> 00:13:24,523
to gather thick data insights.

275
00:13:24,547 --> 00:13:28,471
And what he discovered was something
that they hadn't seen initially

276
00:13:28,495 --> 00:13:29,850
in the quantitative data.

277
00:13:30,892 --> 00:13:33,620
He discovered that people loved
to binge-watch.

278
00:13:33,644 --> 00:13:35,997
In fact, people didn't even
feel guilty about it.

279
00:13:36,021 --> 00:13:37,276
They enjoyed it.

280
00:13:37,300 --> 00:13:38,326
(Laughter)

281
00:13:38,350 --> 00:13:40,706
So Netflix was like,
"Oh. This is a new insight."

282
00:13:40,730 --> 00:13:42,668
So they went to their data science team,

283
00:13:42,692 --> 00:13:45,010
and they were able to scale
this big data insight

284
00:13:45,034 --> 00:13:47,621
in with their quantitative data.

285
00:13:47,645 --> 00:13:50,815
And once they verified it
and validated it,

286
00:13:50,839 --> 00:13:55,600
Netflix decided to do something
very simple but impactful.

287
00:13:56,654 --> 00:14:03,146
They said, instead of offering
the same show from different genres

288
00:14:03,170 --> 00:14:07,058
or more of the different shows
from similar users,

289
00:14:07,082 --> 00:14:09,636
we'll just offer more of the same show.

290
00:14:09,660 --> 00:14:11,765
We'll make it easier
for you to binge-watch.

291
00:14:11,789 --> 00:14:13,275
And they didn't stop there.

292
00:14:13,299 --> 00:14:14,773
They did all these things

293
00:14:14,797 --> 00:14:17,756
to redesign their entire
viewer experience,

294
00:14:17,780 --> 00:14:19,538
to really encourage binge-watching.

295
00:14:20,050 --> 00:14:23,291
It's why people and friends disappear
for whole weekends at a time,

296
00:14:23,315 --> 00:14:25,658
catching up on shows
like "Master of None."

297
00:14:25,682 --> 00:14:29,855
By integrating big data and thick data,
they not only improved their business,

298
00:14:29,879 --> 00:14:32,691
but they transformed how we consume media.

299
00:14:32,715 --> 00:14:37,267
And now their stocks are projected
to double in the next few years.

300
00:14:38,100 --> 00:14:41,930
But this isn't just about
watching more videos

301
00:14:41,954 --> 00:14:43,574
or selling more smartphones.

302
00:14:43,963 --> 00:14:48,013
For some, integrating thick data
insights into the algorithm

303
00:14:48,037 --> 00:14:50,300
could mean life or death,

304
00:14:50,324 --> 00:14:52,470
especially for the marginalized.

305
00:14:53,558 --> 00:14:56,992
All around the country,
police departments are using big data

306
00:14:57,016 --> 00:14:58,979
for predictive policing,

307
00:14:59,003 --> 00:15:02,087
to set bond amounts
and sentencing recommendations

308
00:15:02,111 --> 00:15:05,258
in ways that reinforce existing biases.

309
00:15:06,116 --> 00:15:08,539
NSA's Skynet machine learning algorithm

310
00:15:08,563 --> 00:15:14,007
has possibly aided in the deaths
of thousands of civilians in Pakistan

311
00:15:14,031 --> 00:15:16,752
from misreading cellular device metadata.

312
00:15:18,951 --> 00:15:22,354
As all of our lives become more automated,

313
00:15:22,378 --> 00:15:25,458
from automobiles to health insurance
or to employment,

314
00:15:25,482 --> 00:15:27,832
it is likely that all of us

315
00:15:27,856 --> 00:15:30,845
will be impacted
by the quantification bias.

316
00:15:32,792 --> 00:15:35,413
Now, the good news
is that we've come a long way

317
00:15:35,437 --> 00:15:37,887
from huffing ethylene gas
to make predictions.

318
00:15:37,911 --> 00:15:40,981
We have better tools,
so let's just use them better.

319
00:15:41,005 --> 00:15:43,328
Let's integrate the big data
with the thick data.

320
00:15:43,352 --> 00:15:45,613
Let's bring our temple guides
with the oracles,

321
00:15:45,637 --> 00:15:49,013
and whether this work happens
in companies or nonprofits

322
00:15:49,037 --> 00:15:51,506
or government or even in the software,

323
00:15:51,530 --> 00:15:53,322
all of it matters,

324
00:15:53,346 --> 00:15:56,369
because that means
we're collectively committed

325
00:15:56,393 --> 00:15:58,584
to making better data,

326
00:15:58,608 --> 00:16:00,444
better algorithms, better outputs

327
00:16:00,468 --> 00:16:02,111
and better decisions.

328
00:16:02,135 --> 00:16:05,693
This is how we'll avoid
missing that something.

329
00:16:07,042 --> 00:16:10,990
(Applause)

 类似资料: