1
00:00:01,373 --> 00:00:03,117
Is it just me,

2
00:00:03,141 --> 00:00:05,473
or are there other people here

3
00:00:05,497 --> 00:00:07,841
that are a little bit
disappointed with democracy?

4
00:00:08,986 --> 00:00:11,322
(Applause)

5
00:00:12,141 --> 00:00:14,212
So let's look at a few numbers.

6
00:00:14,934 --> 00:00:17,107
If we look across the world,

7
00:00:17,131 --> 00:00:21,023
the median turnout
in presidential elections

8
00:00:21,047 --> 00:00:22,698
over the last 30 years

9
00:00:22,722 --> 00:00:25,344
has been just 67 percent.

10
00:00:26,329 --> 00:00:28,302
Now, if we go to Europe

11
00:00:28,326 --> 00:00:32,754
and we look at people that participated
in EU parliamentary elections,

12
00:00:32,778 --> 00:00:34,849
the median turnout in those elections

13
00:00:34,873 --> 00:00:36,977
is just 42 percent.

14
00:00:38,125 --> 00:00:39,794
Now let's go to New York,

15
00:00:39,818 --> 00:00:44,499
and let's see how many people voted
in the last election for mayor.

16
00:00:44,523 --> 00:00:48,340
We will find that only
24 percent of people showed up to vote.

17
00:00:49,063 --> 00:00:52,158
What that means is that,
if "Friends" was still running,

18
00:00:52,182 --> 00:00:55,530
Joey and maybe Phoebe
would have shown up to vote.

19
00:00:55,554 --> 00:00:56,844
(Laughter)

20
00:00:57,434 --> 00:01:01,860
And you cannot blame them
because people are tired of politicians.

21
00:01:01,884 --> 00:01:05,771
And people are tired of other people
using the data that they have generated

22
00:01:05,795 --> 00:01:07,993
to communicate with
their friends and family,

23
00:01:08,017 --> 00:01:10,111
to target political propaganda at them.

24
00:01:10,519 --> 00:01:13,247
But the thing about this
is that this is not new.

25
00:01:13,271 --> 00:01:16,496
Nowadays, people use likes
to target propaganda at you

26
00:01:16,520 --> 00:01:19,893
before they use your zip code
or your gender or your age,

27
00:01:19,917 --> 00:01:23,487
because the idea of targeting people
with propaganda for political purposes

28
00:01:23,511 --> 00:01:25,103
is as old as politics.

29
00:01:25,530 --> 00:01:27,808
And the reason why that idea is there

30
00:01:27,832 --> 00:01:31,304
is because democracy
has a basic vulnerability.

31
00:01:31,710 --> 00:01:33,642
This is the idea of a representative.

32
00:01:34,049 --> 00:01:37,993
In principle, democracy is the ability
of people to exert power.

33
00:01:38,017 --> 00:01:41,835
But in practice, we have to delegate
that power to a representative

34
00:01:41,859 --> 00:01:44,114
that can exert that power for us.

35
00:01:44,561 --> 00:01:46,386
That representative is a bottleneck,

36
00:01:46,410 --> 00:01:47,707
or a weak spot.

37
00:01:47,731 --> 00:01:51,680
It is the place that you want to target
if you want to attack democracy

38
00:01:51,704 --> 00:01:55,193
because you can capture democracy
by either capturing that representative

39
00:01:55,217 --> 00:01:57,365
or capturing the way
that people choose it.

40
00:01:58,065 --> 00:01:59,481
So the big question is:

41
00:01:59,505 --> 00:02:01,209
Is this the end of history?

42
00:02:01,989 --> 00:02:05,092
Is this the best that we can do

43
00:02:05,878 --> 00:02:08,945
or, actually, are there alternatives?

44
00:02:10,130 --> 00:02:12,484
Some people have been thinking
about alternatives,

45
00:02:12,508 --> 00:02:16,187
and one of the ideas that is out there
is the idea of direct democracy.

46
00:02:16,790 --> 00:02:19,269
This is the idea of bypassing
politicians completely

47
00:02:19,293 --> 00:02:21,696
and having people vote directly on issues,

48
00:02:21,720 --> 00:02:23,985
having people vote directly on bills.

49
00:02:24,415 --> 00:02:25,751
But this idea is naive

50
00:02:25,775 --> 00:02:28,946
because there's too many things
that we would need to choose.

51
00:02:28,970 --> 00:02:31,752
If you look at the 114th US Congress,

52
00:02:31,776 --> 00:02:34,263
you will have seen that
the House of Representatives

53
00:02:34,287 --> 00:02:37,176
considered more than 6,000 bills,

54
00:02:37,200 --> 00:02:39,856
the Senate considered
more than 3,000 bills

55
00:02:39,880 --> 00:02:42,688
and they approved more than 300 laws.

56
00:02:42,712 --> 00:02:44,291
Those would be many decisions

57
00:02:44,315 --> 00:02:46,502
that each person would have to make a week

58
00:02:46,526 --> 00:02:48,668
on topics that they know little about.

59
00:02:49,229 --> 00:02:51,510
So there's a big cognitive
bandwidth problem

60
00:02:51,534 --> 00:02:55,526
if we're going to try to think about
direct democracy as a viable alternative.

61
00:02:56,205 --> 00:03:00,640
So some people think about the idea
of liquid democracy, or fluid democracy,

62
00:03:00,664 --> 00:03:04,440
which is the idea that you endorse
your political power to someone,

63
00:03:04,464 --> 00:03:06,164
who can endorse it to someone else,

64
00:03:06,188 --> 00:03:08,729
and, eventually, you create
a large follower network

65
00:03:08,753 --> 00:03:12,047
in which, at the end, there's a few people
that are making decisions

66
00:03:12,071 --> 00:03:15,214
on behalf of all of their followers
and their followers.

67
00:03:16,326 --> 00:03:20,455
But this idea also doesn't solve
the problem of the cognitive bandwidth

68
00:03:20,479 --> 00:03:24,356
and, to be honest, it's also quite similar
to the idea of having a representative.

69
00:03:24,795 --> 00:03:28,263
So what I'm going to do today is
I'm going to be a little bit provocative,

70
00:03:28,277 --> 00:03:30,577
and I'm going to ask you, well:

71
00:03:30,601 --> 00:03:37,163
What if, instead of trying
to bypass politicians,

72
00:03:37,187 --> 00:03:39,387
we tried to automate them?

73
00:03:45,871 --> 00:03:48,797
The idea of automation is not new.

74
00:03:48,821 --> 00:03:50,901
It was started more than 300 years ago,

75
00:03:50,925 --> 00:03:53,993
when French weavers decided
to automate the loom.

76
00:03:54,820 --> 00:03:59,180
The winner of that industrial war
was Joseph-Marie Jacquard.

77
00:03:59,204 --> 00:04:00,955
He was a French weaver and merchant

78
00:04:00,979 --> 00:04:03,419
that married the loom
with the steam engine

79
00:04:03,443 --> 00:04:05,633
to create autonomous looms.

80
00:04:05,657 --> 00:04:08,410
And in those autonomous looms,
he gained control.

81
00:04:08,434 --> 00:04:12,319
He could now make fabrics that were
more complex and more sophisticated

82
00:04:12,343 --> 00:04:14,471
than the ones they
were able to do by hand.

83
00:04:15,193 --> 00:04:17,825
But also, by winning that industrial war,

84
00:04:17,849 --> 00:04:21,373
he laid out what has become
the blueprint of automation.

85
00:04:22,135 --> 00:04:25,005
The way that we automate things
for the last 300 years

86
00:04:25,029 --> 00:04:26,411
has always been the same:

87
00:04:27,006 --> 00:04:29,515
we first identify a need,

88
00:04:29,539 --> 00:04:32,723
then we create a tool
to satisfy that need,

89
00:04:32,747 --> 00:04:34,787
like the loom, in this case,

90
00:04:34,811 --> 00:04:37,202
and then we study how people use that tool

91
00:04:37,226 --> 00:04:38,711
to automate that user.

92
00:04:39,242 --> 00:04:42,303
That's how we came
from the mechanical loom

93
00:04:42,327 --> 00:04:44,223
to the autonomous loom,

94
00:04:44,247 --> 00:04:46,367
and that took us a thousand years.

95
00:04:46,391 --> 00:04:48,462
Now, it's taken us only a hundred years

96
00:04:48,486 --> 00:04:51,697
to use the same script
to automate the car.

97
00:04:53,286 --> 00:04:55,738
But the thing is that, this time around,

98
00:04:55,762 --> 00:04:57,891
automation is kind of for real.

99
00:04:57,915 --> 00:05:01,236
This is a video that a colleague of mine
from Toshiba shared with me

100
00:05:01,260 --> 00:05:04,519
that shows the factory
that manufactures solid state drives.

101
00:05:04,543 --> 00:05:06,561
The entire factory is a robot.

102
00:05:06,585 --> 00:05:08,510
There are no humans in that factory.

103
00:05:09,033 --> 00:05:11,254
And the robots are soon
to leave the factories

104
00:05:11,278 --> 00:05:13,300
and become part of our world,

105
00:05:13,324 --> 00:05:15,159
become part of our workforce.

106
00:05:15,183 --> 00:05:16,956
So what I do in my day job

107
00:05:16,980 --> 00:05:20,972
is actually create tools that integrate
data for entire countries

108
00:05:20,996 --> 00:05:24,462
so that we can ultimately have
the foundations that we need

109
00:05:24,486 --> 00:05:28,173
for a future in which we need
to also manage those machines.

110
00:05:29,195 --> 00:05:32,101
But today, I'm not here
to talk to you about these tools

111
00:05:32,125 --> 00:05:33,949
that integrate data for countries.

112
00:05:34,463 --> 00:05:37,085
But I'm here to talk to you
about another idea

113
00:05:37,109 --> 00:05:41,974
that might help us think about how to use
artificial intelligence in democracy.

114
00:05:41,998 --> 00:05:46,731
Because the tools that I build
are designed for executive decisions.

115
00:05:46,755 --> 00:05:50,597
These are decisions that can be cast
in some sort of term of objectivity --

116
00:05:50,621 --> 00:05:52,366
public investment decisions.

117
00:05:52,885 --> 00:05:55,516
But there are decisions
that are legislative,

118
00:05:55,540 --> 00:05:59,327
and these decisions that are legislative
require communication among people

119
00:05:59,351 --> 00:06:01,051
that have different points of view,

120
00:06:01,075 --> 00:06:03,688
require participation, require debate,

121
00:06:03,712 --> 00:06:05,190
require deliberation.

122
00:06:06,241 --> 00:06:09,045
And for a long time,
we have thought that, well,

123
00:06:09,069 --> 00:06:12,529
what we need to improve democracy
is actually more communication.

124
00:06:12,553 --> 00:06:16,262
So all of the technologies that we have
advanced in the context of democracy,

125
00:06:16,286 --> 00:06:19,064
whether they are newspapers
or whether it is social media,

126
00:06:19,088 --> 00:06:21,470
have tried to provide us
with more communication.

127
00:06:22,103 --> 00:06:23,925
But we've been down that rabbit hole,

128
00:06:23,949 --> 00:06:26,697
and we know that's not
what's going to solve the problem.

129
00:06:26,721 --> 00:06:28,717
Because it's not a communication problem,

130
00:06:28,741 --> 00:06:30,489
it's a cognitive bandwidth problem.

131
00:06:30,513 --> 00:06:32,879
So if the problem is one
of cognitive bandwidth,

132
00:06:32,903 --> 00:06:35,490
well, adding more communication to people

133
00:06:35,514 --> 00:06:38,258
is not going to be
what's going to solve it.

134
00:06:38,282 --> 00:06:41,395
What we are going to need instead
is to have other technologies

135
00:06:41,419 --> 00:06:44,465
that help us deal with
some of the communication

136
00:06:44,489 --> 00:06:46,731
that we are overloaded with.

137
00:06:46,755 --> 00:06:48,454
Think of, like, a little avatar,

138
00:06:48,478 --> 00:06:49,817
a software agent,

139
00:06:49,841 --> 00:06:51,719
a digital Jiminy Cricket --

140
00:06:51,743 --> 00:06:52,981
(Laughter)

141
00:06:53,005 --> 00:06:57,017
that basically is able
to answer things on your behalf.

142
00:06:57,759 --> 00:06:59,546
And if we had that technology,

143
00:06:59,570 --> 00:07:02,048
we would be able to offload
some of the communication

144
00:07:02,072 --> 00:07:06,219
and help, maybe, make better decisions
or decisions at a larger scale.

145
00:07:06,860 --> 00:07:10,579
And the thing is that the idea
of software agents is also not new.

146
00:07:10,603 --> 00:07:12,712
We already use them all the time.

147
00:07:13,216 --> 00:07:14,737
We use software agents

148
00:07:14,761 --> 00:07:18,436
to choose the way that we're going
to drive to a certain location,

149
00:07:19,070 --> 00:07:21,171
the music that we're going to listen to

150
00:07:21,758 --> 00:07:24,779
or to get suggestions
for the next books that we should read.

151
00:07:25,994 --> 00:07:28,568
So there is an obvious idea
in the 21st century

152
00:07:28,592 --> 00:07:31,235
that was as obvious as the idea

153
00:07:31,259 --> 00:07:36,840
of putting together a steam engine
with a loom at the time of Jacquard.

154
00:07:37,538 --> 00:07:41,994
And that idea is combining
direct democracy with software agents.

155
00:07:42,849 --> 00:07:44,970
Imagine, for a second, a world

156
00:07:44,994 --> 00:07:48,160
in which, instead of having
a representative that represents you

157
00:07:48,184 --> 00:07:49,758
and millions of other people,

158
00:07:49,782 --> 00:07:52,818
you can have a representative
that represents only you,

159
00:07:53,504 --> 00:07:55,758
with your nuanced political views --

160
00:07:55,782 --> 00:07:59,126
that weird combination
of libertarian and liberal

161
00:07:59,150 --> 00:08:01,542
and maybe a little bit
conservative on some issues

162
00:08:01,566 --> 00:08:03,674
and maybe very progressive on others.

163
00:08:03,698 --> 00:08:06,965
Politicians nowadays are packages,
and they're full of compromises.

164
00:08:06,989 --> 00:08:10,623
But you might have someone
that can represent only you,

165
00:08:10,647 --> 00:08:12,499
if you are willing to give up the idea

166
00:08:12,523 --> 00:08:14,772
that that representative is a human.

167
00:08:15,229 --> 00:08:17,311
If that representative
is a software agent,

168
00:08:17,335 --> 00:08:21,505
we could have a senate that has
as many senators as we have citizens.

169
00:08:21,529 --> 00:08:24,387
And those senators are going to be able
to read every bill

170
00:08:24,411 --> 00:08:27,158
and they're going to be able
to vote on each one of them.

171
00:08:27,822 --> 00:08:30,778
So there's an obvious idea
that maybe we want to consider.

172
00:08:30,802 --> 00:08:33,224
But I understand that in this day and age,

173
00:08:33,248 --> 00:08:35,137
this idea might be quite scary.

174
00:08:36,391 --> 00:08:39,831
In fact, thinking of a robot
coming from the future

175
00:08:39,855 --> 00:08:41,528
to help us run our governments

176
00:08:41,552 --> 00:08:43,183
sounds terrifying.

177
00:08:44,223 --> 00:08:45,874
But we've been there before.

178
00:08:45,898 --> 00:08:47,171
(Laughter)

179
00:08:47,195 --> 00:08:49,690
And actually he was quite a nice guy.

180
00:08:51,677 --> 00:08:58,111
So what would the Jacquard loom
version of this idea look like?

181
00:08:58,135 --> 00:09:00,036
It would be a very simple system.

182
00:09:00,060 --> 00:09:03,518
Imagine a system that you log in
and you create your avatar,

183
00:09:03,542 --> 00:09:05,998
and then you're going
to start training your avatar.

184
00:09:06,022 --> 00:09:08,704
So you can provide your avatar
with your reading habits,

185
00:09:08,728 --> 00:09:10,589
or connect it to your social media,

186
00:09:10,613 --> 00:09:13,021
or you can connect it to other data,

187
00:09:13,045 --> 00:09:15,317
for example by taking
psychological tests.

188
00:09:15,341 --> 00:09:18,309
And the nice thing about this
is that there's no deception.

189
00:09:18,333 --> 00:09:21,672
You are not providing data to communicate
with your friends and family

190
00:09:21,696 --> 00:09:24,847
that then gets used in a political system.

191
00:09:24,871 --> 00:09:28,575
You are providing data to a system
that is designed to be used

192
00:09:28,599 --> 00:09:30,715
to make political decisions
on your behalf.

193
00:09:31,264 --> 00:09:35,244
Then you take that data and you choose
a training algorithm,

194
00:09:35,268 --> 00:09:36,831
because it's an open marketplace

195
00:09:36,855 --> 00:09:39,641
in which different people
can submit different algorithms

196
00:09:39,665 --> 00:09:44,059
to predict how you're going to vote,
based on the data you have provided.

197
00:09:44,083 --> 00:09:47,538
And the system is open,
so nobody controls the algorithms;

198
00:09:47,562 --> 00:09:49,674
there are algorithms
that become more popular

199
00:09:49,698 --> 00:09:51,421
and others that become less popular.

200
00:09:51,445 --> 00:09:53,252
Eventually, you can audit the system.

201
00:09:53,276 --> 00:09:55,157
You can see how your avatar is working.

202
00:09:55,181 --> 00:09:57,333
If you like it,
you can leave it on autopilot.

203
00:09:57,357 --> 00:09:59,419
If you want to be
a little more controlling,

204
00:09:59,443 --> 00:10:01,411
you can actually choose that they ask you

205
00:10:01,435 --> 00:10:03,503
every time they're going
to make a decision,

206
00:10:03,527 --> 00:10:05,162
or you can be anywhere in between.

207
00:10:05,186 --> 00:10:07,591
One of the reasons
why we use democracy so little

208
00:10:07,615 --> 00:10:11,183
may be because democracy
has a very bad user interface.

209
00:10:11,207 --> 00:10:13,690
And if we improve the user
interface of democracy,

210
00:10:13,714 --> 00:10:15,841
we might be able to use it more.

211
00:10:16,452 --> 00:10:19,659
Of course, there's a lot of questions
that you might have.

212
00:10:20,473 --> 00:10:22,634
Well, how do you train these avatars?

213
00:10:22,658 --> 00:10:24,552
How do you keep the data secure?

214
00:10:24,576 --> 00:10:27,824
How do you keep the systems
distributed and auditable?

215
00:10:27,848 --> 00:10:29,910
How about my grandmother,
who's 80 years old

216
00:10:29,946 --> 00:10:31,906
and doesn't know how to use the internet?

217
00:10:32,262 --> 00:10:34,483
Trust me, I've heard them all.

218
00:10:34,507 --> 00:10:39,067
So when you think about an idea like this,
you have to beware of pessimists

219
00:10:39,091 --> 00:10:43,410
because they are known to have
a problem for every solution.

220
00:10:43,434 --> 00:10:45,259
(Laughter)

221
00:10:45,283 --> 00:10:48,323
So I want to invite you to think
about the bigger ideas.

222
00:10:48,347 --> 00:10:51,973
The questions I just showed you
are little ideas

223
00:10:51,997 --> 00:10:54,899
because they are questions
about how this would not work.

224
00:10:55,502 --> 00:10:57,483
The big ideas are ideas of:

225
00:10:57,507 --> 00:10:59,314
What else can you do with this

226
00:10:59,338 --> 00:11:01,227
if this would happen to work?

227
00:11:01,774 --> 00:11:05,219
And one of those ideas is,
well, who writes the laws?

228
00:11:05,854 --> 00:11:10,077
In the beginning, we could have
the avatars that we already have,

229
00:11:10,101 --> 00:11:13,598
voting on laws that are written
by the senators or politicians

230
00:11:13,622 --> 00:11:14,973
that we already have.

231
00:11:15,491 --> 00:11:17,205
But if this were to work,

232
00:11:17,902 --> 00:11:20,252
you could write an algorithm

233
00:11:20,276 --> 00:11:22,426
that could try to write a law

234
00:11:22,450 --> 00:11:24,871
that would get a certain
percentage of approval,

235
00:11:24,895 --> 00:11:26,677
and you could reverse the process.

236
00:11:26,701 --> 00:11:30,213
Now, you might think that this idea
is ludicrous and we should not do it,

237
00:11:30,237 --> 00:11:33,023
but you cannot deny that it's an idea
that is only possible

238
00:11:33,047 --> 00:11:36,067
in a world in which direct democracy
and software agents

239
00:11:36,091 --> 00:11:38,747
are a viable form of participation.

240
00:11:40,596 --> 00:11:43,349
So how do we start the revolution?

241
00:11:44,238 --> 00:11:47,548
We don't start this revolution
with picket fences or protests

242
00:11:47,572 --> 00:11:51,762
or by demanding our current politicians
to be changed into robots.

243
00:11:51,786 --> 00:11:53,335
That's not going to work.

244
00:11:53,359 --> 00:11:54,971
This is much more simple,

245
00:11:54,995 --> 00:11:56,154
much slower

246
00:11:56,178 --> 00:11:57,592
and much more humble.

247
00:11:57,616 --> 00:12:01,965
We start this revolution by creating
simple systems like this in grad schools,

248
00:12:01,989 --> 00:12:04,083
in libraries, in nonprofits.

249
00:12:04,107 --> 00:12:06,761
And we try to figure out
all of those little questions

250
00:12:06,785 --> 00:12:08,006
and those little problems

251
00:12:08,030 --> 00:12:11,931
that we're going to have to figure out
to make this idea something viable,

252
00:12:11,955 --> 00:12:14,306
to make this idea something
that we can trust.

253
00:12:14,330 --> 00:12:17,965
And as we create those systems that have
a hundred people, a thousand people,

254
00:12:17,989 --> 00:12:21,759
a hundred thousand people voting
in ways that are not politically binding,

255
00:12:21,783 --> 00:12:23,801
we're going to develop trust in this idea,

256
00:12:23,825 --> 00:12:25,344
the world is going to change,

257
00:12:25,368 --> 00:12:28,243
and those that are as little
as my daughter is right now

258
00:12:28,267 --> 00:12:29,604
are going to grow up.

259
00:12:30,580 --> 00:12:32,949
And by the time my daughter is my age,

260
00:12:32,973 --> 00:12:37,409
maybe this idea, that I know
today is very crazy,

261
00:12:37,433 --> 00:12:41,567
might not be crazy to her
and to her friends.

262
00:12:41,956 --> 00:12:43,793
And at that point,

263
00:12:43,817 --> 00:12:46,420
we will be at the end of our history,

264
00:12:46,444 --> 00:12:49,265
but they will be
at the beginning of theirs.

265
00:12:49,646 --> 00:12:50,829
Thank you.

266
00:12:50,853 --> 00:12:53,895
(Applause)