1
00:00:00,000 --> 00:00:04,860
So our next section is about common misconceptions and false parallels about voting technology.

2
00:00:06,200 --> 00:00:09,660
And our presenters are David Jefferson, who I introduced before.

3
00:00:09,780 --> 00:00:11,000
That's the chairman of Verified Voting.

4
00:00:11,080 --> 00:00:15,600
And Joe Hall, who's the chief technologist and director for Internet Architecture Project

5
00:00:15,600 --> 00:00:17,400
at the Center for Democracy and Technology.

6
00:00:18,080 --> 00:00:19,520
So let's give him a round of applause.

7
00:00:21,520 --> 00:00:22,620
Thank you, Jake.

8
00:00:24,220 --> 00:00:29,400
Okay, what I'm going to talk about is not about particular voting systems or the code

9
00:00:29,400 --> 00:00:32,080
or vulnerabilities or anything like that.

10
00:00:32,180 --> 00:00:35,780
I'm going to talk about the requirements for voting

11
00:00:35,780 --> 00:00:44,800
and how they compare with the requirements for other applications, online applications in particular.

12
00:00:46,280 --> 00:00:51,840
The point I'm going to be making, the bottom line, is that from a technical point of view,

13
00:00:51,920 --> 00:00:59,080
the security, privacy, transparency, reliability, and other properties

14
00:00:59,080 --> 00:00:59,380
that are going to be used for voting,

15
00:00:59,400 --> 00:01:05,200
the requirements of voting as an application are more complex, harder to satisfy,

16
00:01:05,500 --> 00:01:09,820
than any of the other security problems you're likely to have encountered.

17
00:01:10,160 --> 00:01:17,480
I'm going to make a detailed comparison between voting and e-commerce as the prime example.

18
00:01:19,240 --> 00:01:22,720
So let me move on here.

19
00:01:23,320 --> 00:01:29,380
First, I want to just show up this generic slide about the laws and regulations.

20
00:01:29,400 --> 00:01:34,200
There are a large number of technical requirements that have to be satisfied simultaneously

21
00:01:34,200 --> 00:01:37,340
to have a satisfactory public election.

22
00:01:37,540 --> 00:01:38,960
Now, I'm not going to read all this slide.

23
00:01:39,620 --> 00:01:46,900
I'm just going to read the headings here, voter authentication issues, double vote prevention,

24
00:01:47,760 --> 00:01:54,940
ballot selection, meaning the person gets the correct ballot, accurately capturing the voter's choices,

25
00:01:55,280 --> 00:01:58,680
the integrity of the balloting process,

26
00:01:58,680 --> 00:02:02,800
the accuracy of the tally of those ballots,

27
00:02:03,280 --> 00:02:06,980
the two different kinds of privacy, which I will come back to in a moment,

28
00:02:07,680 --> 00:02:14,700
mixed-mode ballots, software certification, software integrity,

29
00:02:15,360 --> 00:02:18,760
the availability and reliability of the voting process,

30
00:02:19,680 --> 00:02:27,800
all forms of security issues, transparency, manageability, and the mission criticality.

31
00:02:28,680 --> 00:02:30,900
Now, I'm going to highlight only a couple of those.

32
00:02:30,960 --> 00:02:32,100
I'm not going to read all of them.

33
00:02:32,160 --> 00:02:38,680
But I wanted to first of all talk about ballot integrity that, of course,

34
00:02:40,320 --> 00:02:45,000
that no ballots are lost, no ballots are corrupted in the process,

35
00:02:45,080 --> 00:02:49,620
and no phony ballots are inserted into the process in the course of an election.

36
00:02:51,420 --> 00:02:58,120
I want to highlight privacy and what I call privacy two, two different notions of privacy,

37
00:02:58,120 --> 00:03:01,680
that is, simultaneously, requirements for a public election.

38
00:03:02,220 --> 00:03:12,200
The first notion is that it should be impossible to associate a cast ballot with the person who cast it,

39
00:03:12,300 --> 00:03:14,660
at least up to the size of a precinct.

40
00:03:14,700 --> 00:03:24,200
So the most you should be able to say is that you cast one of the 150 ballots that are in the ballot box at that precinct this day.

41
00:03:24,340 --> 00:03:28,100
So that's as close as we ought to be able to identify you, the size of a precinct.

42
00:03:28,120 --> 00:03:32,100
And nobody should be able to identify any closer than that.

43
00:03:32,140 --> 00:03:35,540
Not a court, not an election official, nobody.

44
00:03:35,740 --> 00:03:38,340
That's a strong privacy requirement.

45
00:03:39,140 --> 00:03:41,700
The second privacy requirement is a little more subtle.

46
00:03:42,480 --> 00:03:51,080
Even if the voter wants to tell you or tell the world how he voted, he can do so.

47
00:03:51,140 --> 00:03:54,360
He's free to declare how he voted, but he's not free to prove it.

48
00:03:54,500 --> 00:03:56,680
He is not allowed to have any proof.

49
00:03:58,120 --> 00:04:04,480
And the protection that is afforded by that is a voter who can't prove how he voted

50
00:04:05,000 --> 00:04:10,060
is less subject to coercion than a voter who has proof,

51
00:04:10,260 --> 00:04:13,480
and so the coercer could demand that proof as to how he voted,

52
00:04:13,580 --> 00:04:17,040
and a voter who can't prove how he voted can't sell his vote.

53
00:04:17,220 --> 00:04:21,620
He can't sell one vote, and you can't create a market for vote buying and selling.

54
00:04:21,860 --> 00:04:25,780
So this second privacy criterion is different from the first one,

55
00:04:25,880 --> 00:04:27,760
but both are critical to...

56
00:04:28,120 --> 00:04:30,260
to strong election integrity.

57
00:04:31,420 --> 00:04:36,160
Now, I mentioned reliability and security and transparency.

58
00:04:36,660 --> 00:04:40,580
The point about reliability is that all those previous properties

59
00:04:40,580 --> 00:04:48,180
have to hold, even in the case of partial failures of the voting system,

60
00:04:48,240 --> 00:04:49,500
partial technical failures.

61
00:04:49,680 --> 00:04:57,580
If a voting machine goes down, or if the local area network

62
00:04:58,120 --> 00:05:00,800
or the voting precinct goes down, or the poll book goes down,

63
00:05:00,860 --> 00:05:04,060
or something like that, you have to be able to finish the election anyway.

64
00:05:04,540 --> 00:05:09,640
It has to be robust against all kinds of partial system failures.

65
00:05:09,760 --> 00:05:11,660
Of course, nothing will be robust against, you know,

66
00:05:11,740 --> 00:05:18,420
a regional power failure, perhaps, but at least partial system failures.

67
00:05:18,800 --> 00:05:26,560
And security talks about all of the other good properties that the election has to have have to...

68
00:05:26,560 --> 00:05:27,300
have to hold.

69
00:05:27,300 --> 00:05:27,460
And security talks about all of the other good properties that the election has to have have to hold,

70
00:05:27,480 --> 00:05:33,960
even in the case of a limited, at least, conspiracy of election officials,

71
00:05:34,260 --> 00:05:38,080
programmers, voters, third parties, to undermine the election, okay?

72
00:05:38,440 --> 00:05:40,100
Even if they're trying to do so.

73
00:05:41,280 --> 00:05:46,640
We need to be able to conduct an election and guarantee the integrity, even in the face of that.

74
00:05:47,860 --> 00:05:48,380
Okay.

75
00:05:50,140 --> 00:05:51,520
Now let me move on.

76
00:05:53,500 --> 00:05:54,420
So I'm going to...

77
00:05:54,420 --> 00:05:57,280
I said I was going to talk to you about how voting is dispensable.

78
00:05:57,300 --> 00:06:00,480
It's distinctly different from other security problems.

79
00:06:00,660 --> 00:06:06,560
And the comparison I'm going to make is online commerce, let's say, buying things from Amazon,

80
00:06:06,800 --> 00:06:08,840
versus online voting.

81
00:06:10,140 --> 00:06:13,800
You know, voting from your home computer.

82
00:06:14,180 --> 00:06:14,960
Now, we don't...

83
00:06:14,960 --> 00:06:18,280
We don't have a lot of online voting in the United States.

84
00:06:18,380 --> 00:06:19,100
We have some.

85
00:06:19,740 --> 00:06:26,220
Online voting is an extremely dangerous idea, but I'm not here to rail against it from that point of view.

86
00:06:26,220 --> 00:06:28,340
I'm talking about requirements here.

87
00:06:29,030 --> 00:06:35,360
And people have a tendency to think in terms that if I can shop safely online,

88
00:06:35,740 --> 00:06:39,060
why can't I vote safely online?

89
00:06:39,360 --> 00:06:41,700
After all, the two are very similar.

90
00:06:41,980 --> 00:06:44,860
They both involve a simple online transaction.

91
00:06:45,140 --> 00:06:47,060
You navigate to the correct website.

92
00:06:47,320 --> 00:06:48,980
You identify yourself.

93
00:06:49,400 --> 00:06:54,180
Once you're identified, you make your choices, your purchases, or your votes.

94
00:06:54,180 --> 00:06:58,460
When you've looked over your purchases and your votes, and you're happy,

95
00:06:58,540 --> 00:07:03,340
you press that, you know, buy button or that vote button, and you're done.

96
00:07:04,600 --> 00:07:05,200
It's a...

97
00:07:05,200 --> 00:07:11,620
So superficially, from just an online transaction point of view, they sound very similar.

98
00:07:11,820 --> 00:07:16,680
And what I'm about to try to convince you of now is that they are not remotely similar,

99
00:07:16,920 --> 00:07:19,280
except in that superficial sense.

100
00:07:20,140 --> 00:07:23,180
So on the left side there are the categories.

101
00:07:23,180 --> 00:07:26,240
These are the categories of differences that I'm going to be talking about,

102
00:07:26,440 --> 00:07:30,880
and the middle column is shopping, and the right-hand column would be voting.

103
00:07:32,120 --> 00:07:36,720
For shopping, there's no eligibility requirement of who's allowed to buy from Amazon.

104
00:07:36,900 --> 00:07:37,740
You can be a kid.

105
00:07:37,780 --> 00:07:39,100
You can be a non-citizen.

106
00:07:39,200 --> 00:07:40,160
You can be a felon.

107
00:07:40,240 --> 00:07:40,760
We don't care.

108
00:07:40,860 --> 00:07:42,360
Amazon does not care.

109
00:07:42,700 --> 00:07:44,700
They don't have an eligibility requirement.

110
00:07:45,080 --> 00:07:49,780
But we do have strong eligibility requirements for who is allowed to vote in a U.S. election,

111
00:07:49,940 --> 00:07:51,180
and we have to ascertain them.

112
00:07:52,360 --> 00:07:56,680
We do that in all states at voter registration time,

113
00:07:56,900 --> 00:07:59,040
and in a growing number of states now.

114
00:07:59,120 --> 00:08:05,420
We also do it, voter ID states, at voting time for, at least for in-person voters,

115
00:08:05,520 --> 00:08:06,980
if not for absentee voters.

116
00:08:07,200 --> 00:08:10,140
So there's this eligibility issue that distinguishes the two.

117
00:08:10,740 --> 00:08:12,500
Then there's the double vote transaction.

118
00:08:12,860 --> 00:08:15,700
Everybody knows that you can't vote twice in the same election.

119
00:08:15,960 --> 00:08:21,020
But Amazon doesn't care if you buy two pairs or four pairs of socks in the same transaction.

120
00:08:21,180 --> 00:08:25,100
Or if you come back and buy more things ten times in the same day.

121
00:08:25,200 --> 00:08:32,020
There is no analog to the notion of double vote prohibition in the e-commerce world.

122
00:08:32,200 --> 00:08:39,520
Again, it's more an additional requirement or structural difference more difficult in the voting world.

123
00:08:40,440 --> 00:08:41,520
Proxy transactions.

124
00:08:42,240 --> 00:08:49,920
A proxy transaction is when you authorize explicitly or implicitly somebody else to use your Amazon account.

125
00:08:49,920 --> 00:08:54,040
Your kid, your spouse, or you give somebody your credit card number,

126
00:08:54,160 --> 00:08:55,220
or you give them your password.

127
00:08:55,400 --> 00:08:57,540
Amazon doesn't care if you do that.

128
00:08:57,600 --> 00:08:58,560
That's perfectly legal.

129
00:08:58,740 --> 00:08:59,780
They're okay with that.

130
00:08:59,940 --> 00:09:01,140
They don't even know.

131
00:09:01,760 --> 00:09:06,640
However, in the voting world, proxy voting is not allowed.

132
00:09:06,820 --> 00:09:10,300
You cannot authorize somebody to vote for you under any circumstances.

133
00:09:11,280 --> 00:09:12,800
In any of the states.

134
00:09:13,060 --> 00:09:13,620
Period.

135
00:09:16,920 --> 00:09:19,900
So, the consequence of those first three things.

136
00:09:19,920 --> 00:09:28,720
The first three differences between shopping and voting is that it appears in the issue of voter identification and authentication.

137
00:09:29,240 --> 00:09:37,140
In the case of Amazon and e-commerce, they are satisfied with weak authentication of who the buyer is.

138
00:09:37,300 --> 00:09:40,480
The true identity of the buyer is not really required.

139
00:09:40,940 --> 00:09:45,000
When you sign up for an Amazon account, they don't ask for your driver's license number.

140
00:09:45,060 --> 00:09:46,260
They don't ask for your passport.

141
00:09:46,540 --> 00:09:47,620
They don't ask for your picture.

142
00:09:48,440 --> 00:09:51,840
They don't really care who you are.

143
00:09:51,980 --> 00:09:54,700
What they care about is that when you later do a transaction,

144
00:09:55,000 --> 00:10:00,800
that the transaction is probably good and the credit card is definitely valid at the moment of the transaction.

145
00:10:01,040 --> 00:10:02,080
That's what they care about.

146
00:10:02,220 --> 00:10:07,080
But the identity of exactly who's buying, you or your spouse or your kid,

147
00:10:07,580 --> 00:10:08,960
they don't really care.

148
00:10:09,040 --> 00:10:16,380
I mean, they'd like it for a marketing information, but it's not a barrier to the purchase if they don't actually know who you are.

149
00:10:16,380 --> 00:10:21,560
But in voting, because we do have to know that you're eligible to vote,

150
00:10:21,680 --> 00:10:23,320
we can't allow you to vote twice,

151
00:10:23,480 --> 00:10:26,620
and we can't allow you to vote for somebody else,

152
00:10:26,780 --> 00:10:34,200
we need strong voter identification and authentication to guarantee those properties.

153
00:10:34,360 --> 00:10:37,940
And it's even harder online, if we're talking about online voting,

154
00:10:38,060 --> 00:10:45,760
because we are worried about large-scale attacks through weaknesses in authentication.

155
00:10:45,760 --> 00:10:49,680
You need strong authentication of ballots.

156
00:10:50,020 --> 00:10:51,100
Sorry, of voters.

157
00:10:52,760 --> 00:10:55,400
Additional distinctions between shopping and voting.

158
00:10:56,920 --> 00:10:57,520
Privacy.

159
00:10:57,580 --> 00:10:59,340
The privacy rules for shopping are,

160
00:10:59,480 --> 00:11:02,880
well, both parties, the buyer and the seller,

161
00:11:03,120 --> 00:11:06,060
know everything about all parts of the transaction.

162
00:11:06,300 --> 00:11:08,100
They know what you're buying, how much it costs,

163
00:11:08,320 --> 00:11:13,280
how to return it or what happens if the transaction somehow goes bad.

164
00:11:14,860 --> 00:11:15,460
And,

165
00:11:15,760 --> 00:11:18,740
you actually do get a receipt in some sense for the transaction,

166
00:11:18,980 --> 00:11:22,020
which you can show to a third party and you can prove to a court,

167
00:11:22,180 --> 00:11:24,440
for example, if the transaction goes bad,

168
00:11:24,560 --> 00:11:26,640
that you have the details and you have the proof of it.

169
00:11:27,520 --> 00:11:28,500
But in voting,

170
00:11:30,080 --> 00:11:34,940
who casts what vote is an absolute secret, as I said earlier.

171
00:11:36,280 --> 00:11:39,200
And a voter knows how she voted,

172
00:11:39,340 --> 00:11:42,820
but she must not be able to prove it to a third party.

173
00:11:42,900 --> 00:11:45,680
It's a different, much more subtle privacy requirement.

174
00:11:45,680 --> 00:11:48,200
The privacy requirements for shopping

175
00:11:48,200 --> 00:11:50,500
are policy requirements,

176
00:11:50,980 --> 00:11:53,660
business and contractual arrangements

177
00:11:53,660 --> 00:11:57,180
that the seller has with the buyer.

178
00:11:57,540 --> 00:12:00,780
They are encased in law in voting.

179
00:12:03,620 --> 00:12:04,620
Integrity.

180
00:12:04,820 --> 00:12:09,800
Integrity means that the transaction is actually correctly done.

181
00:12:10,200 --> 00:12:14,740
So, in the case of shopping,

182
00:12:15,840 --> 00:12:18,580
Amazon will not go out of business

183
00:12:18,580 --> 00:12:22,400
if 1% of all of its transactions are fraudulent.

184
00:12:22,600 --> 00:12:24,600
And it's a good thing because they are.

185
00:12:25,340 --> 00:12:27,040
But in the voting world,

186
00:12:27,200 --> 00:12:29,420
we cannot tolerate a 1%.

187
00:12:29,420 --> 00:12:32,200
We can't tolerate a 1 tenth of 1% error.

188
00:12:32,760 --> 00:12:37,980
We really need systems that are architected and designed and intended

189
00:12:37,980 --> 00:12:39,520
and as far as humanly possible,

190
00:12:39,680 --> 00:12:44,080
100% accurate recording of every transaction.

191
00:12:44,080 --> 00:12:46,640
Especially in this era of closer and closer elections.

192
00:12:50,380 --> 00:12:52,340
Nonetheless, errors occur anyway.

193
00:12:52,740 --> 00:12:54,680
So let's talk about error detectability.

194
00:12:55,720 --> 00:12:57,200
In the commerce world,

195
00:12:57,900 --> 00:13:02,160
failures and frauds are going to be detectable eventually

196
00:13:02,160 --> 00:13:05,820
because of double entry bookkeeping

197
00:13:05,820 --> 00:13:08,820
and the presence of receipts in both directions

198
00:13:08,820 --> 00:13:09,900
and a paper trail.

199
00:13:10,180 --> 00:13:11,940
If nothing else,

200
00:13:12,040 --> 00:13:13,920
some account is going to go to zero

201
00:13:14,080 --> 00:13:15,680
someday and you're going to detect it.

202
00:13:16,920 --> 00:13:18,660
But in the voting world,

203
00:13:21,220 --> 00:13:25,580
modified vote transactions may be completely undetectable.

204
00:13:28,200 --> 00:13:30,040
And privacy violations,

205
00:13:30,340 --> 00:13:32,940
the requirement that votes be private

206
00:13:32,940 --> 00:13:34,160
in the two ways that I described,

207
00:13:34,480 --> 00:13:36,740
are almost always undetectable.

208
00:13:36,980 --> 00:13:38,200
So we have this,

209
00:13:38,280 --> 00:13:42,080
one of the burdens that we have

210
00:13:42,080 --> 00:13:44,160
in architecting election systems

211
00:13:44,160 --> 00:13:49,740
is the fact that many errors are fundamentally undetectable.

212
00:13:49,860 --> 00:13:52,740
Because the voting officials,

213
00:13:52,940 --> 00:13:54,940
they don't know how you intended to vote.

214
00:13:55,320 --> 00:13:58,320
And if you claim that they recorded your vote incorrectly,

215
00:13:58,720 --> 00:14:02,100
you can't prove it to them as a third party.

216
00:14:02,380 --> 00:14:05,980
So it's just really not fundamentally possible

217
00:14:05,980 --> 00:14:09,140
in many cases to actually detect that an error happened at all.

218
00:14:10,260 --> 00:14:11,940
Even if you did detect it,

219
00:14:11,940 --> 00:14:13,260
let's talk about recoverability.

220
00:14:13,580 --> 00:14:15,320
In the case of the shopping world,

221
00:14:15,440 --> 00:14:17,300
there are all kinds of ways to recover

222
00:14:17,940 --> 00:14:19,280
from bad transactions.

223
00:14:19,620 --> 00:14:23,200
Many times they're reversible with returns or refunds,

224
00:14:23,220 --> 00:14:27,240
other ameliorations like insurance and tax deductions

225
00:14:27,240 --> 00:14:30,480
and just price increases to recover the losses.

226
00:14:31,060 --> 00:14:33,680
Those are all kinds of ways to deal with the risk

227
00:14:33,680 --> 00:14:37,740
of faulty or fraudulent transactions in the commerce world.

228
00:14:38,020 --> 00:14:41,760
But even if you detected in the voting world

229
00:14:41,940 --> 00:14:43,600
that there was a fraudulent vote cast,

230
00:14:43,880 --> 00:14:45,840
you wouldn't be able to do anything about it.

231
00:14:45,900 --> 00:14:47,800
Because if I know there's a fraudulent vote cast,

232
00:14:48,340 --> 00:14:50,680
I don't know what the vote should have been.

233
00:14:50,840 --> 00:14:52,740
I don't know who cast it.

234
00:14:52,920 --> 00:14:55,960
I can't bring them to justice.

235
00:14:57,880 --> 00:15:00,920
The recoverability issue is far more troubling,

236
00:15:01,120 --> 00:15:03,420
both detection and recoverability,

237
00:15:03,580 --> 00:15:08,200
in the elections world than it is in the commerce world.

238
00:15:09,200 --> 00:15:10,240
And finally,

239
00:15:10,240 --> 00:15:12,980
additional, this is my last slide here,

240
00:15:14,280 --> 00:15:14,760
availability.

241
00:15:16,800 --> 00:15:19,620
Amazon needs to be up almost all the time.

242
00:15:19,740 --> 00:15:22,560
But if they were down for an hour a year,

243
00:15:24,020 --> 00:15:26,480
they wouldn't, you know, they wouldn't fail.

244
00:15:26,640 --> 00:15:28,280
Or if they were down for 10 hours a year,

245
00:15:28,340 --> 00:15:28,880
they wouldn't fail.

246
00:15:28,980 --> 00:15:30,740
They would fire their IT team maybe,

247
00:15:30,820 --> 00:15:34,080
but they would not suffer catastrophic losses.

248
00:15:36,040 --> 00:15:40,060
However, voting on election day must be up.

249
00:15:40,240 --> 00:15:40,400
Period.

250
00:15:40,860 --> 00:15:46,820
Because whereas Amazon would suffer a certain loss of business

251
00:15:46,820 --> 00:15:49,500
or most likely just only delayed business,

252
00:15:50,380 --> 00:15:54,060
there are statutory requirements for the date that election day is.

253
00:15:54,140 --> 00:15:55,280
And you can't postpone it.

254
00:15:55,360 --> 00:15:57,020
And you can't ask, well,

255
00:15:57,980 --> 00:15:59,320
voter, come back tomorrow.

256
00:15:59,320 --> 00:16:00,980
We're fixing our systems today.

257
00:16:01,120 --> 00:16:02,100
That's just not an option.

258
00:16:03,280 --> 00:16:07,580
So the damage from downtime in a voting system on election day

259
00:16:07,580 --> 00:16:09,120
is basically unrecoverable.

260
00:16:10,240 --> 00:16:12,640
The transparency requirements are different, too.

261
00:16:13,500 --> 00:16:14,960
In the shopping world,

262
00:16:15,140 --> 00:16:19,280
buyers have only the right to examine and question

263
00:16:19,280 --> 00:16:20,980
the details of their own transactions.

264
00:16:21,420 --> 00:16:22,020
Period.

265
00:16:22,180 --> 00:16:23,080
Nobody else's.

266
00:16:23,360 --> 00:16:24,780
But in the voting world,

267
00:16:25,000 --> 00:16:28,740
voters and candidates and the press and everybody else,

268
00:16:28,880 --> 00:16:30,240
at least in principle,

269
00:16:30,520 --> 00:16:33,960
has a right to examine and observe and audit

270
00:16:33,960 --> 00:16:36,200
every detail about the election.

271
00:16:36,860 --> 00:16:37,460
Period.

272
00:16:37,540 --> 00:16:38,460
From soup to nuts

273
00:16:38,460 --> 00:16:43,140
up to as long as they preserve the privacy properties,

274
00:16:43,380 --> 00:16:48,040
you know, no identification of which voter cast which ballot.

275
00:16:48,200 --> 00:16:51,120
But everything else is supposed to be transparent,

276
00:16:51,420 --> 00:16:52,840
at least in principle.

277
00:16:52,920 --> 00:16:54,000
In practice it's not,

278
00:16:54,120 --> 00:16:55,800
but I'm talking in principle here.

279
00:16:58,660 --> 00:17:01,480
The motivations of attackers are completely different, too.

280
00:17:01,540 --> 00:17:06,140
People who attack Amazon transactions

281
00:17:06,140 --> 00:17:08,440
are usually motivated by money.

282
00:17:08,460 --> 00:17:10,140
They're trying to steal your personal data.

283
00:17:10,200 --> 00:17:11,580
They're trying to steal your credit card number.

284
00:17:11,680 --> 00:17:12,680
But in the voting world,

285
00:17:12,860 --> 00:17:16,760
the desire is almost always political.

286
00:17:17,100 --> 00:17:19,240
They're trying to fix the outcome of the election

287
00:17:19,240 --> 00:17:21,020
or they're trying to disrupt the election.

288
00:17:21,620 --> 00:17:23,720
They may also be making money doing that

289
00:17:23,720 --> 00:17:25,020
if they are fixing the election

290
00:17:25,020 --> 00:17:26,860
on behalf of somebody else who's paying them.

291
00:17:28,140 --> 00:17:28,940
And finally,

292
00:17:30,600 --> 00:17:32,580
the consequences of a successful attack

293
00:17:32,580 --> 00:17:36,760
are completely different in the voting and e-commerce world.

294
00:17:36,760 --> 00:17:38,720
Somebody loses money,

295
00:17:38,960 --> 00:17:39,920
usually the merchant,

296
00:17:40,040 --> 00:17:40,940
not the shopper,

297
00:17:41,000 --> 00:17:45,920
with a fraudulent or faulty transaction in the shopping world.

298
00:17:46,220 --> 00:17:47,680
But in the voting world,

299
00:17:47,920 --> 00:17:49,840
the wrong people may be elected,

300
00:17:49,980 --> 00:17:53,360
the wrong initiatives that should have passed fail

301
00:17:53,360 --> 00:17:54,620
or should have failed pass.

302
00:17:55,000 --> 00:17:59,720
And the legitimacy of democratic government is undermined

303
00:17:59,720 --> 00:18:02,240
and our national security is imperiled.

304
00:18:02,620 --> 00:18:04,880
So the bottom line here is

305
00:18:04,880 --> 00:18:06,240
voting,

306
00:18:06,240 --> 00:18:06,540
security,

307
00:18:07,360 --> 00:18:07,680
privacy,

308
00:18:07,920 --> 00:18:08,360
transparency,

309
00:18:08,740 --> 00:18:09,100
availability,

310
00:18:09,500 --> 00:18:09,960
reliability,

311
00:18:10,160 --> 00:18:13,540
all in every dimension more difficult

312
00:18:14,090 --> 00:18:18,160
than the security problems

313
00:18:18,160 --> 00:18:20,960
that you may have encountered in other application domains.

314
00:18:21,340 --> 00:18:21,680
Joe.

315
00:18:23,890 --> 00:18:24,900
Oh, we're about to switch.

316
00:18:26,010 --> 00:18:26,580
That's okay.

317
00:18:40,040 --> 00:18:40,440
We,

318
00:18:41,720 --> 00:18:43,340
you would think we would have solved

319
00:18:43,340 --> 00:18:45,480
the projector problems by now

320
00:18:46,280 --> 00:18:46,820
in the,

321
00:18:46,820 --> 00:18:47,380
in the world.

322
00:18:49,920 --> 00:18:50,600
You're close.

323
00:18:51,460 --> 00:18:52,240
You got it.

324
00:18:53,480 --> 00:18:54,320
There you go.

325
00:18:58,780 --> 00:18:59,500
Thank you, sir.

326
00:18:59,580 --> 00:19:01,000
And it's such a joy to share the stage

327
00:19:01,000 --> 00:19:02,380
with people like David and Candace

328
00:19:02,380 --> 00:19:03,840
and Matt Blaze after this.

329
00:19:03,860 --> 00:19:05,380
There, we've been working on this stuff for,

330
00:19:06,240 --> 00:19:09,000
for a long time and they are mentors

331
00:19:09,000 --> 00:19:10,560
and role models for,

332
00:19:10,580 --> 00:19:12,700
for me and for a ton of other people.

333
00:19:12,760 --> 00:19:13,340
And I'll stop gushing.

334
00:19:13,520 --> 00:19:13,920
All right.

335
00:19:13,960 --> 00:19:14,620
Hi, my name's Joe.

336
00:19:14,780 --> 00:19:16,060
You may have seen me earlier today.

337
00:19:16,160 --> 00:19:17,720
I'm from the Center for Democracy and Technology

338
00:19:17,720 --> 00:19:18,540
in Washington, DC.

339
00:19:19,160 --> 00:19:22,020
I should mention I am hiring an election technology fellow.

340
00:19:22,120 --> 00:19:23,380
So if you know something about cybersecurity

341
00:19:23,380 --> 00:19:24,480
and election technology,

342
00:19:24,960 --> 00:19:26,040
please come talk to me.

343
00:19:26,380 --> 00:19:27,440
I'll be here all weekend.

344
00:19:28,940 --> 00:19:30,760
So what I'm going to do is talk about

345
00:19:30,760 --> 00:19:33,520
a number of what I would consider common misconceptions.

346
00:19:33,860 --> 00:19:35,100
Things you hear people shout.

347
00:19:35,100 --> 00:19:36,620
Why don't we just do this?

348
00:19:36,620 --> 00:19:37,220
Or why don't we just do that?

349
00:19:37,280 --> 00:19:37,520
Or aha,

350
00:19:37,740 --> 00:19:40,540
I have a solution to this extremely complicated problem

351
00:19:40,540 --> 00:19:41,740
that you've never thought of about before.

352
00:19:42,920 --> 00:19:45,500
Don't dare talk about blockchain and Bitcoin

353
00:19:45,500 --> 00:19:46,960
or cryptocurrencies to me.

354
00:19:47,600 --> 00:19:49,340
You don't put anything on the blockchain

355
00:19:49,340 --> 00:19:51,440
you expect to be secret for 80 years.

356
00:19:51,600 --> 00:19:52,320
Just okay?

357
00:19:52,500 --> 00:19:53,320
Like that's it.

358
00:19:53,440 --> 00:19:53,780
You know,

359
00:19:53,780 --> 00:19:55,600
we're using public key cryptography for this crap.

360
00:19:55,800 --> 00:19:57,120
So that's not on here,

361
00:19:57,160 --> 00:19:58,740
but I wanted to rant a little bit about that.

362
00:19:59,120 --> 00:20:00,860
So I'm going to go through five misconceptions.

363
00:20:01,140 --> 00:20:02,200
I'm going to go pretty quickly,

364
00:20:02,300 --> 00:20:03,380
just because you may,

365
00:20:03,920 --> 00:20:04,320
one,

366
00:20:04,320 --> 00:20:05,960
have beef with my interpretation,

367
00:20:06,220 --> 00:20:06,660
or two,

368
00:20:06,800 --> 00:20:09,100
have your own misconception that you'd like to raise

369
00:20:09,100 --> 00:20:10,600
and I'd love to just get into that

370
00:20:10,600 --> 00:20:11,620
and talk about that stuff

371
00:20:11,620 --> 00:20:12,680
because that's where the action is.

372
00:20:12,800 --> 00:20:14,260
You don't want to hear me talk too much.

373
00:20:14,600 --> 00:20:16,340
So just use hand counted paper ballots

374
00:20:16,340 --> 00:20:17,660
is often what you hear people say.

375
00:20:17,900 --> 00:20:19,980
Why can't we just do it like how Canada does

376
00:20:19,980 --> 00:20:21,040
or how England does it?

377
00:20:21,100 --> 00:20:22,680
And that's because they have a parliamentary system

378
00:20:22,680 --> 00:20:24,500
where you make exactly one choice on the ballot.

379
00:20:24,620 --> 00:20:26,840
We have extremely complicated ballots.

380
00:20:27,240 --> 00:20:28,980
The Netherlands has pretty complicated ballots,

381
00:20:29,100 --> 00:20:29,500
but then again,

382
00:20:29,760 --> 00:20:29,880
it's,

383
00:20:29,880 --> 00:20:30,440
anyway,

384
00:20:31,000 --> 00:20:32,420
we have very complicated ballots.

385
00:20:32,540 --> 00:20:32,700
All,

386
00:20:32,700 --> 00:20:33,400
you know,

387
00:20:33,440 --> 00:20:34,260
in California,

388
00:20:34,440 --> 00:20:37,720
there are mosquito abatement district contests.

389
00:20:37,720 --> 00:20:39,560
It's almost a dog catcher on your ballot,

390
00:20:39,720 --> 00:20:40,960
which is great

391
00:20:40,960 --> 00:20:43,440
because you want people to be able to elect things

392
00:20:43,440 --> 00:20:44,020
that they need to elect.

393
00:20:44,260 --> 00:20:47,000
Maybe we make too many decisions in the voting box,

394
00:20:47,100 --> 00:20:48,740
but that's more of a political statement.

395
00:20:49,000 --> 00:20:51,340
We also try our best to enfranchise

396
00:20:51,340 --> 00:20:52,620
as many people as possible.

397
00:20:52,720 --> 00:20:52,980
Many,

398
00:20:52,980 --> 00:20:55,680
many other classes of people do we want to try

399
00:20:55,680 --> 00:20:58,460
and allow the ability to vote independently

400
00:20:58,460 --> 00:21:00,640
and privately by themselves.

401
00:21:00,740 --> 00:21:01,800
So people with disabilities,

402
00:21:02,040 --> 00:21:02,680
voters,

403
00:21:02,700 --> 00:21:03,500
people with vision impairment,

404
00:21:03,720 --> 00:21:04,320
it's very difficult.

405
00:21:04,480 --> 00:21:07,340
People with motor disabilities is also difficult.

406
00:21:07,580 --> 00:21:09,060
If you have compound disabilities,

407
00:21:09,460 --> 00:21:12,000
so if you have hearing disabilities

408
00:21:12,000 --> 00:21:13,940
and a sight disability,

409
00:21:14,320 --> 00:21:15,660
that's extremely hard.

410
00:21:15,860 --> 00:21:17,580
There have been actually a couple of voting systems,

411
00:21:17,740 --> 00:21:18,340
maybe just one,

412
00:21:18,500 --> 00:21:21,880
that targeted that particular kind of compound disability.

413
00:21:22,440 --> 00:21:25,820
There are Native American languages

414
00:21:25,820 --> 00:21:27,100
that have no written form,

415
00:21:27,180 --> 00:21:29,480
that even though these people are literate and speak,

416
00:21:29,660 --> 00:21:31,820
they would prefer to vote in the language

417
00:21:31,820 --> 00:21:32,720
that they speak at home,

418
00:21:32,800 --> 00:21:35,000
and we try our best to enfranchise them.

419
00:21:35,280 --> 00:21:36,720
L.A. County, for example,

420
00:21:36,860 --> 00:21:38,400
I think has 11 languages,

421
00:21:38,640 --> 00:21:39,580
things like Laotian.

422
00:21:39,760 --> 00:21:41,640
It's not even included in the Voting Rights Act,

423
00:21:41,740 --> 00:21:43,080
which is the thing that tends to tell you

424
00:21:43,080 --> 00:21:44,320
what kind of languages you have to support

425
00:21:44,320 --> 00:21:46,920
given traditionally disenfranchised populations.

426
00:21:47,560 --> 00:21:48,860
But what we want to do is make sure,

427
00:21:48,940 --> 00:21:49,460
if you,

428
00:21:50,020 --> 00:21:50,960
as my dad would say,

429
00:21:51,020 --> 00:21:51,840
don't dream in English,

430
00:21:52,000 --> 00:21:53,060
we want to make sure you can vote

431
00:21:53,060 --> 00:21:54,240
in the language that you dream in.

432
00:21:55,400 --> 00:21:56,000
And third,

433
00:21:56,600 --> 00:21:57,900
it's just faster and easier

434
00:21:57,900 --> 00:21:59,060
and it can be much more usable

435
00:21:59,060 --> 00:22:00,920
to use a touchscreen for many people.

436
00:22:00,920 --> 00:22:02,900
Paper, when designed correctly, is awesome.

437
00:22:03,120 --> 00:22:04,880
But imagine trying to put 11 languages

438
00:22:04,880 --> 00:22:06,160
on a single piece of paper.

439
00:22:07,180 --> 00:22:08,220
If you can do that,

440
00:22:08,820 --> 00:22:10,160
man, you can make a lot of money.

441
00:22:11,000 --> 00:22:12,920
But that's not to say that there's no place

442
00:22:12,920 --> 00:22:14,360
for paper ballots at all.

443
00:22:14,500 --> 00:22:15,120
For example,

444
00:22:15,220 --> 00:22:16,680
the voter verified paper audit trail.

445
00:22:17,200 --> 00:22:18,880
Optical scan ballots are exactly

446
00:22:18,880 --> 00:22:20,240
what we're talking about.

447
00:22:20,340 --> 00:22:22,480
That's rooting trust and reliability

448
00:22:22,480 --> 00:22:24,760
and indelibility in a physical object.

449
00:22:25,240 --> 00:22:26,760
A colleague of ours, Doug Jones,

450
00:22:26,980 --> 00:22:28,120
prefers etched aluminum,

451
00:22:28,340 --> 00:22:30,280
but we haven't found a way to do that

452
00:22:30,280 --> 00:22:30,740
in a,

453
00:22:30,920 --> 00:22:32,660
as long as it would be cheap and quick

454
00:22:32,660 --> 00:22:34,100
and all these other things,

455
00:22:34,160 --> 00:22:36,040
it would last for hundreds of years.

456
00:22:36,940 --> 00:22:38,800
Which is maybe not such a good thing,

457
00:22:38,860 --> 00:22:39,200
I'm not sure.

458
00:22:39,440 --> 00:22:40,840
We also do recounts.

459
00:22:40,840 --> 00:22:42,140
These are the heavily lawyered up,

460
00:22:42,240 --> 00:22:43,580
we're going to count every single ballot

461
00:22:43,580 --> 00:22:45,280
and see if we come up with a different result

462
00:22:45,800 --> 00:22:46,220
and then,

463
00:22:46,260 --> 00:22:47,780
stuff that we worked on in my post-doc

464
00:22:47,780 --> 00:22:48,800
at Princeton

465
00:22:48,800 --> 00:22:50,880
and that Phillip Stark has really

466
00:22:50,880 --> 00:22:51,560
sort of led

467
00:22:51,560 --> 00:22:52,740
from Berkeley

468
00:22:52,740 --> 00:22:54,920
are these things called risk limiting audits

469
00:22:54,920 --> 00:22:56,740
and these are statistical methods

470
00:22:56,740 --> 00:22:59,160
of counting a random sample of ballots,

471
00:22:59,360 --> 00:23:00,760
comparing it to the corresponding

472
00:23:00,760 --> 00:23:04,820
data structure. So you're looking at here's the thing the person marked, here's the data structure

473
00:23:04,820 --> 00:23:11,620
that interprets what the person marked. And those are statistical tests designed to make sure that

474
00:23:11,620 --> 00:23:17,600
if you find enough error that would cause the outcome to be different, that you would actually

475
00:23:17,600 --> 00:23:21,380
have announced somebody else, you escalate until you've actually counted all the ballots, right?

476
00:23:21,440 --> 00:23:26,660
Or more practically in some places, you may have to rerun an election, which is easier for smaller

477
00:23:26,660 --> 00:23:31,260
elections, almost impossible legally constitutionally for something like the presidential

478
00:23:31,260 --> 00:23:37,960
election. Damn it. That means I'm taking too much time on each slide, guys. Anyway, so that's

479
00:23:37,960 --> 00:23:42,260
one. Hacking is a new thing in elections. So if you've been in this whole track all day, you may

480
00:23:42,260 --> 00:23:51,080
hear some of this over and over again. Oh, damn it. Cool. Awesome. Thank you, Goon. Cycle, cycle,

481
00:23:51,200 --> 00:23:56,000
cycle. I'll just talk. So there's been a considerable amount of hacking of various kinds.

482
00:23:56,180 --> 00:23:56,640
Hacking is a new thing. It's a new thing. It's a new thing. It's a new thing. It's a new thing.

483
00:23:56,660 --> 00:24:00,520
Hacking is just a horrible word. It's such a big umbrella word, right? You can almost describe

484
00:24:00,520 --> 00:24:07,380
any kind of interference. Or meddling is the word of the now for elections. You can just sort of say,

485
00:24:07,460 --> 00:24:10,140
hey, they did something. We're not sure what they did, but they did something. Or in some cases,

486
00:24:10,160 --> 00:24:15,460
we know what they did. In fact, the U.S. has a pretty substantial history, and I have a lot of

487
00:24:15,460 --> 00:24:20,380
notes on this that I can share with you. If you have any friends that are Chileanos or Chileanos,

488
00:24:20,520 --> 00:24:25,680
in Chile, we did a whole bunch of horrible things to them, using mostly money in elections.

489
00:24:25,680 --> 00:24:31,160
In Italy, to make sure the communists didn't get elected, we did a bunch of things there. In the

490
00:24:31,160 --> 00:24:37,260
Philippines, those weren't like actual direct interventions in the machinery of their democracies,

491
00:24:37,260 --> 00:24:43,080
so to speak, but they were ways of using influence and power to directly change the outcome of

492
00:24:43,080 --> 00:24:49,220
elections. The coolest story is the Ukraine 2014 election, which you may have heard of. And if you

493
00:24:49,220 --> 00:24:55,520
haven't, there's a really cool short paper I can point you at. A hacker group affiliated with the

494
00:24:55,520 --> 00:25:01,680
Russian government, supposedly, broke into the Ukrainian central election administrator's

495
00:25:01,680 --> 00:25:07,480
tabulation server, changed the results about an hour before they were going to be announced.

496
00:25:07,780 --> 00:25:12,680
The Ukrainians found out, rolled everything back, and announced the real results, but Russian

497
00:25:12,680 --> 00:25:18,440
television actually reported the result that the hackers hacked into the machine. And so that's

498
00:25:18,440 --> 00:25:23,820
the only time we've ever seen something that is probably as close to confirmation of an actual

499
00:25:23,820 --> 00:25:24,860
hack that we've ever seen.

500
00:25:25,520 --> 00:25:29,360
The other things are not nearly as sexy. So for example, there's a bunch of denial of service attacks, and

501
00:25:29,360 --> 00:25:34,100
while those are very powerful, that's like everyone's hammer, right? It's always going to be there.

502
00:25:34,220 --> 00:25:38,300
It's always going to be hard to protect against, increasingly so. There was a case, I think in

503
00:25:38,300 --> 00:25:43,580
Illinois, I'm having a hard time tracking this down, but I will find it, where a poll worker ran an

504
00:25:43,580 --> 00:25:48,520
optical, I think it was a couple poll workers, ran an optical scan ballot through the optical scanner

505
00:25:48,520 --> 00:25:53,920
like 200 extra times, trying to inflate the count for one particular contest. But it's really easy

506
00:25:53,920 --> 00:26:00,040
to just count the number of ballots in the bin and recognize, oh geez, there's 200 pieces of paper missing, or

507
00:26:00,040 --> 00:26:06,140
this is off, and just recount the stuff. So that was caught, and they were prosecuted. And then this is one

508
00:26:06,140 --> 00:26:11,520
that David Jefferson told me about, that there was an insider attack on the Canvas database in the

509
00:26:12,440 --> 00:26:17,240
1994 presidential election in South Africa, the first election for Mandela. I didn't even know about

510
00:26:17,240 --> 00:26:23,080
that one, so that's really awesome. I mean, it's good to know about. So hacking has happened before.

511
00:26:23,920 --> 00:26:28,340
And we've covered this a little bit, you know, this strict anonymity requirement. People often say, well,

512
00:26:28,360 --> 00:26:33,140
that's where your problem is. If you didn't have this strict anonymity requirement, you wouldn't have all

513
00:26:33,140 --> 00:26:38,820
these problems. So, for example, in the United Kingdom, when you go vote, you sign your name right next to a

514
00:26:38,820 --> 00:26:43,580
number that's printed on your ballot. So there's an actual mapping between your signature and the number on

515
00:26:43,580 --> 00:26:49,580
your ballot. But that mapping is kept as essentially a state secret that you have to meet very high bar in an

516
00:26:49,580 --> 00:26:53,900
actual election fraud court case, and able to unmask that. So it's a little bit more complicated than that.

517
00:26:53,920 --> 00:27:02,100
Masking. Unmask that mapping. Anyway, so this gets actually kind of interesting. You may not have realized

518
00:27:02,100 --> 00:27:08,720
this, but in the very beginning of at least the United States, and obviously this is a more global room than just the

519
00:27:08,720 --> 00:27:15,240
U.S., but you had to be a white male landowner in order to vote, right? And so you had to own a certain amount of

520
00:27:15,240 --> 00:27:21,320
land, not just any land. It had to be a certain size. You had to be a man, and you had to be a white man at that.

521
00:27:23,920 --> 00:27:28,860
You had to actually raise your hand or announce your vote, viva voce, out loud. That's awesome in terms of

522
00:27:28,860 --> 00:27:33,180
transparency. You could actually sit there and record. Everyone could independently record it and come up with

523
00:27:33,180 --> 00:27:39,540
their own results. But what that means is the coercion and the forms of undue influence, like vote buying, is

524
00:27:39,540 --> 00:27:44,860
still right there. You know, a father can say to the son, hey, you're not going to inherit that land that you want so

525
00:27:44,860 --> 00:27:50,160
bad if you don't show me exactly how you voted before you cast a ballot. And I should mention, there's only one

526
00:27:50,160 --> 00:27:53,900
state in the U.S. that allows you to show your ballot to someone before you cast a ballot. And that's a

527
00:27:53,900 --> 00:27:55,920
state in the U.S. that allows you to show your ballot to someone before you cast it. And it's West Virginia. For

528
00:27:55,920 --> 00:28:00,380
some reason, in the Constitution, it has this thing about open voting, which allows you to show your

529
00:28:00,380 --> 00:28:04,660
ballot to someone. So if you were going to buy a ballot, it's still illegal to purchase votes in West

530
00:28:04,660 --> 00:28:09,320
Virginia. But if you were going to do it, that would be the mechanism to do it. You're not going to get a lot of

531
00:28:09,320 --> 00:28:17,360
votes, I don't think. And so before about 1900, Election Day was a payday for most American voters. You

532
00:28:17,360 --> 00:28:23,840
would go to the polls. There would be a bunch of people with very bright and distinctive color

533
00:28:23,840 --> 00:28:28,680
of paper that were partisan ballots that listed for one party all the people. And you would sit there and just play

534
00:28:28,680 --> 00:28:34,980
them off each other until you got enough money. And they could go and watch you, deposit it in, and then give you the

535
00:28:34,980 --> 00:28:40,980
second half of what they would want to pay you. And it was about three months' worth of pay. We adopt this thing

536
00:28:40,980 --> 00:28:48,720
called the Australian ballot. Most states, between 1890 and about 1905, adopted the Australian ballot. This is an

537
00:28:49,480 --> 00:28:53,820
anonymous ballot printed by the government on a uniform stock with everyone's name on it. And it was a ballot that was

538
00:28:53,820 --> 00:29:02,240
printed the same size on the same ballot. When that happened, voter participation rates just plummeted because it was no

539
00:29:02,240 --> 00:29:08,300
longer a payday, and now it was your actual civic duty. You had to actually care. You had to give a shit and go and vote to do that

540
00:29:08,300 --> 00:29:18,340
stuff. And it's also really important to be able to make sure that the votes are anonymous so that you don't have, in days

541
00:29:18,340 --> 00:29:23,800
like now where you don't, you wouldn't go to a polling place necessarily and do that, but there could be, you know, a

542
00:29:23,820 --> 00:29:30,300
coordination mechanisms involving actual cryptocurrencies to pay you, things like Zcash, right? The thing that is the more

543
00:29:30,300 --> 00:29:39,960
anonymous version of Bitcoin. And now, you know, and this is something else that David Jefferson pointed out to me, this is such a tense

544
00:29:39,960 --> 00:29:47,520
political climate we live in that you can imagine that no, you know, never has the anonymity of your vote been more

545
00:29:47,520 --> 00:29:53,800
important. You could be someone in a party who holds an unpopular view all of a sudden, and they may not be the person who's supposed to be

546
00:29:53,820 --> 00:29:58,180
may take retribution on you if they found you voted a way that they didn't want you

547
00:29:58,180 --> 00:30:02,560
to. Okay. So here's ones that are maybe more

548
00:30:02,560 --> 00:30:07,580
interesting to the technical folks in the room. You'll hear people say our voting system

549
00:30:07,580 --> 00:30:14,680
is highly distributed, so it's hard to attack. I should say that monoculture and the absence

550
00:30:14,680 --> 00:30:20,560
of monoculture, it's a weak guard at best. As people have said, there are often similarities

551
00:30:20,560 --> 00:30:26,960
across systems that can be exploited even if they're very different platforms. But it's

552
00:30:26,960 --> 00:30:31,700
only really a benefit, this distribution, to contests that span more than one jurisdiction.

553
00:30:32,040 --> 00:30:36,640
And what do I mean by that? I mean that if you have your proverbial Tony Soprano, I guess

554
00:30:36,640 --> 00:30:42,560
Tony Soprano was never in a proverb, so whatever. Have your fictitious Tony Soprano, if he could

555
00:30:42,560 --> 00:30:48,260
fix a local waste management bond in his favor for 100 grand, say a $5 million waste

556
00:30:48,260 --> 00:30:50,380
management bond or something like that, not a very big number.

557
00:30:50,560 --> 00:30:55,240
But that means the threat model there is for that one jurisdiction, that one county,

558
00:30:55,280 --> 00:30:58,840
that one town, they're going to have to start thinking about stuff like that and planning

559
00:30:58,840 --> 00:31:05,940
for the nasties that are around your town trying to influence your employees and otherwise

560
00:31:05,940 --> 00:31:12,460
get into the systems. And even then, many states are dominated by very populous single

561
00:31:12,460 --> 00:31:19,580
or one or two or three jurisdictions, right? And because voting equipment has a hard time

562
00:31:19,580 --> 00:31:20,540
scaling, it's hard to get into the systems. And so I think it's important to think about

563
00:31:20,560 --> 00:31:24,880
that. It's more than likely that those jurisdictions are going to be running the same voting software

564
00:31:24,880 --> 00:31:29,840
or at least not very different voting machines and software. And so those statewide races

565
00:31:29,840 --> 00:31:34,860
can be influenced by attacking one or two local jurisdictions. And then when you have

566
00:31:34,860 --> 00:31:42,240
a lot of jurisdictions, you can sort of pick the easiest ones to attack. So maybe you could

567
00:31:42,660 --> 00:31:47,100
attack the big jurisdictions, but it turns out that they have really good cybersecurity

568
00:31:47,100 --> 00:31:50,540
practices and regular physical security practices and whatever.

569
00:31:50,560 --> 00:31:54,200
Plato's security practices. If they have all those things set, you might be like, well,

570
00:31:54,220 --> 00:32:00,180
geez, I can go attack a piece of lower hanging fruit. I may have to change more votes. I may have to

571
00:32:00,180 --> 00:32:06,780
disrupt more records. Like something I care a lot about, you know, your voter registrator

572
00:32:06,780 --> 00:32:13,440
databases are online, but they're often the canonical versions are in staged, you know, air gap

573
00:32:13,440 --> 00:32:18,880
networks that are updated with deltas every few days and stuff like that. But the data on voter

574
00:32:18,880 --> 00:32:20,540
registration systems is insane. And so I think it's important to think about that. And so I think

575
00:32:20,560 --> 00:32:20,740
that's a really good point. I think that's a really good point. And I think that's a really good

576
00:32:20,740 --> 00:32:20,980
point. And I think that's a really good point. There's a lot of things that are staged for

577
00:32:20,980 --> 00:32:25,400
electronic poll books. These are the computers we're using instead of spiral brown notebooks

578
00:32:25,400 --> 00:32:31,220
anymore in polling places. You can imagine that staged data is sitting somewhere probably not

579
00:32:31,220 --> 00:32:36,200
very well secured, but we hope so before it's put on those poll books. That's a natural place to

580
00:32:36,200 --> 00:32:41,940
do a very disruptive attack where you single out even a given party or just randomly select people

581
00:32:41,940 --> 00:32:47,100
in a place that have a certain kind of political affiliation. Finally, the systems are not

582
00:32:47,100 --> 00:32:50,540
connected to the Internet. So you may have heard this a lot, too. But the system is not connected

583
00:32:50,560 --> 00:32:51,140
to the Internet. It's not connected to any kind of political affiliation. So I think that's a

584
00:32:51,140 --> 00:32:54,120
really good point. And I love to point out what Dan Wallach has said in the past. It's like, yeah,

585
00:32:54,140 --> 00:32:58,440
we had viruses before the Internet. We had these things called floppy disks. And they would be

586
00:32:58,440 --> 00:33:05,600
transmission vectors for malware. And there's just so many ways to cross air gaps these days.

587
00:33:05,700 --> 00:33:11,120
You know, like I was in my earlier talk today, I talked about air gap networks having rogue

588
00:33:11,120 --> 00:33:17,740
Wi-Fi attached to them for whatever reason to help people out because people don't recognize that

589
00:33:17,740 --> 00:33:20,540
that may be a security vulnerability. But it helps them do that. And so I think that's a really good

590
00:33:20,540 --> 00:33:26,500
their work better. And so, you know, insider attacks is another really good point to make

591
00:33:26,500 --> 00:33:31,040
here is that insiders have the ability to cross the air gap if you get the right insider.

592
00:33:32,800 --> 00:33:36,800
And, you know, there's the machines in the voting village, which I highly recommend you go check

593
00:33:36,800 --> 00:33:41,560
out. The AccuVote TS, which is still being used in Georgia, was one where Ed Felton and his team

594
00:33:41,560 --> 00:33:48,440
designed a virus that you could put on during the primary on one machine. And then between the

595
00:33:48,440 --> 00:33:53,040
primary and the general election, it would propagate to all the other equipment so that

596
00:33:53,040 --> 00:33:58,280
you'd have the hack ready for the general election. So while these requirements are really

597
00:33:58,280 --> 00:34:04,460
hard and it means our security models have to be a little different, the attackers are going to be

598
00:34:04,460 --> 00:34:09,080
just as clever to get around those kinds of things. And I think that's it. So thank you.

599
00:34:09,140 --> 00:34:13,640
There's my GQ signal if you want to talk about any of this or if you want to potentially come

600
00:34:13,640 --> 00:34:16,640
work for me. That would be wonderful. We'd love to take questions.

601
00:34:24,860 --> 00:34:26,380
Just yell and I'll repeat it.

602
00:34:48,900 --> 00:34:56,280
There's a lot of things where it seems like these problems are solvable. Maybe some of it's a

603
00:34:56,280 --> 00:35:03,580
statutory problem. But, I mean, vote by mail. Yeah, someone can see my ballot, someone can see my mail.

604
00:35:04,940 --> 00:35:07,140
But it's hard to deal with these things in scale, isn't it?

605
00:35:10,520 --> 00:35:16,360
Well, Washington until recently was a singular case because of the, well, in Oregon.

606
00:35:18,440 --> 00:35:23,780
I don't have the same confidence in vote by mail that you do. Okay? Number one is authentication of the

607
00:35:23,780 --> 00:35:29,660
voter is very weak. Okay? The best you have is a signature check, which is often not really even done

608
00:35:29,660 --> 00:35:35,720
or is only done with software aid. Usually you don't have a strong way of authenticating who the

609
00:35:35,720 --> 00:35:42,320
voter is. So in most states in the United States, when there is vote fraud, if it's not committed by

610
00:35:42,320 --> 00:35:48,180
election officials, it's committed through mail-in voting. So mail-in voting is definitely weaker.

611
00:35:48,440 --> 00:35:54,540
Yes, you say, you seem to shrug your shoulders when you said that your vote could be lost in the mail, as though

612
00:35:54,540 --> 00:36:00,640
that didn't matter. The fact is you as the voter may not know that, or you may not learn it until late, until too

613
00:36:00,640 --> 00:36:06,740
late. And yes, maybe you can go and vote in person if you discover on time that your vote didn't make it.

614
00:36:06,980 --> 00:36:13,060
But the point is you may have voted by mail because you were out of state or not in a position to go.

615
00:36:13,240 --> 00:36:18,220
I mean, for many people, that's what absentee voting is all about in every other state.

616
00:36:18,440 --> 00:36:26,660
Besides, besides Washington and Oregon and Colorado. So I guess I'm not as cavalier about the strengths of mail-in voting as you are.

617
00:36:27,120 --> 00:36:32,340
I don't think it's a good idea for it to be universal as it is in Washington and Oregon.

618
00:36:33,380 --> 00:36:41,040
And I, I don't think it strengthens democracy to have that be the only mode of voting.

619
00:36:42,020 --> 00:36:48,060
Yeah, I would just comment on that. In Washington, you can actually track your vote from when it was received.

620
00:36:48,440 --> 00:36:55,600
When it was counted. You can actually verify whether or not it happened. And you get a full month, two months to vote.

621
00:36:55,840 --> 00:36:58,440
So there's plenty of time to address all of these problems.

622
00:36:58,980 --> 00:37:06,960
Yes, but most people don't vote a month or two months in advance. They vote in the last month or the last couple of days.

623
00:37:07,600 --> 00:37:09,580
Anyway, let me, let's, let's move on.

624
00:37:09,580 --> 00:37:14,580
Well, and so just to make a quick comment, I vote by mail in a place that has universal vote by mail.

625
00:37:14,620 --> 00:37:21,640
I really appreciate it because I, I, I am so busy that I, that's, but I, I think the computer science community at least,

626
00:37:21,640 --> 00:37:28,800
that works on elections, recognizes that, I think I'm speaking for many people at once that I don't have the liberty to,

627
00:37:28,880 --> 00:37:35,360
but I'm going to, and say that, that I think we have the opposite view, that vote by mail tends to relax a lot of these requirements.

628
00:37:35,520 --> 00:37:38,920
Like, for example, I know a guy, I know a guy.

629
00:37:38,920 --> 00:37:46,220
In San Francisco who gives his lawyer his ballot with his signature on it and $200 and has his lawyer vote for him,

630
00:37:46,340 --> 00:37:51,200
which is like, it's vote, it's weird. It's, it's not vote buying. It's not coercion.

631
00:37:51,240 --> 00:37:56,360
It's more like you'll make better decisions than I will because I trust you, Mr. Lawyer and stuff like that.

632
00:37:56,440 --> 00:38:02,140
And so it allows a bunch of things that are sort of wrinkles in this stronger model of things.

633
00:38:02,200 --> 00:38:07,600
And the trick there is the, is the models that sort of fit in place in terms of the,

634
00:38:07,680 --> 00:38:08,900
in place of universal vote by mail.

635
00:38:08,920 --> 00:38:13,560
Universal vote by mail are, are, are, are newer things like vote centers, right?

636
00:38:13,640 --> 00:38:19,300
Where you have, instead of small polling places all over the place, you have big polling places that you can go to any one of them.

637
00:38:19,500 --> 00:38:26,240
The problem with stuff like that is if you have optical scan ballots, you have to have enough copies of every ballot style,

638
00:38:26,600 --> 00:38:30,960
which can, in a primary election with a bunch of parties, can just, can grow geometrically.

639
00:38:31,120 --> 00:38:31,620
And so...

640
00:38:31,620 --> 00:38:32,340
Print on demand.

641
00:38:33,100 --> 00:38:35,140
Print on demand, for sure, works very well.

642
00:38:35,280 --> 00:38:36,860
Some states that's very hard, right?

643
00:38:36,900 --> 00:38:38,100
Because you've got to certify the...

644
00:38:38,920 --> 00:38:43,480
The actual printer and the people doing the printer as being ballot printing entities, right?

645
00:38:43,620 --> 00:38:45,140
But that means we can change the laws.

646
00:38:45,320 --> 00:38:47,140
But print on demand is, is a way to do that.

647
00:38:47,200 --> 00:38:53,440
You don't get some of the features you get from actual bonded printers in print on demand.

648
00:38:53,900 --> 00:38:54,460
But anyway.

649
00:38:54,600 --> 00:38:55,480
And there are things...

650
00:38:55,480 --> 00:38:56,060
I'm sorry, I'll shut up.

651
00:38:56,260 --> 00:38:56,640
Larry?

652
00:38:58,060 --> 00:38:58,440
.

653
00:39:03,200 --> 00:39:03,580
.

654
00:39:11,940 --> 00:39:18,120
So the question was, at what point does the convenience versus security,

655
00:39:18,440 --> 00:39:24,000
what point does turnout get affected by the tension between convenience and security?

656
00:39:24,200 --> 00:39:27,540
So if you make things too secure and not convenient enough, are you lowering turnout?

657
00:39:27,700 --> 00:39:28,420
That's a great question.

658
00:39:28,560 --> 00:39:31,060
I'm not aware of research on that.

659
00:39:31,120 --> 00:39:33,540
Paul Gronke and folks like that who, who, who...

660
00:39:33,540 --> 00:39:37,380
Who do a lot of voter sentiment analysis may have data on that.

661
00:39:38,320 --> 00:39:40,480
There's probably some natural experiments, but I don't know.

662
00:39:41,060 --> 00:39:48,060
It's something we have to plan well for and make sure that elections are as usable as possible and easy as possible,

663
00:39:48,220 --> 00:39:49,540
but no easier, right?

664
00:39:50,980 --> 00:39:51,420
Sir?

665
00:39:52,240 --> 00:39:52,680
.

666
00:40:05,740 --> 00:40:06,160
.

667
00:40:08,740 --> 00:40:11,280
So let me try and repeat the question and let me know if I got it wrong.

668
00:40:11,640 --> 00:40:20,440
Is there a, any theoretical model that preserves anonymity in voting where you can...

669
00:40:20,440 --> 00:40:21,320
Say it again?

670
00:40:22,460 --> 00:40:22,880
.

671
00:40:24,280 --> 00:40:27,760
Treat each vote as an atomic transaction, sort of like acid, right?

672
00:40:28,760 --> 00:40:29,180
.

673
00:40:30,580 --> 00:40:32,960
Are you talking about remote voting or are you talking about...

674
00:40:34,920 --> 00:40:35,360
.

675
00:40:38,400 --> 00:40:43,140
So there are formal methods here and there are people who, who model polling places.

676
00:40:44,760 --> 00:40:50,320
I'm not sure if that exactly would answer your question, but come find me and I can at least point you to some of those papers.

677
00:40:50,420 --> 00:40:52,900
And if it's not there, maybe you can do it.

678
00:40:52,940 --> 00:40:58,260
But I have a feeling that, that the polling place models people use, especially in queuing theory,

679
00:40:58,440 --> 00:41:02,540
which is, you know, how, how often, how often do people arrive at a polling place

680
00:41:02,960 --> 00:41:05,040
to optimize the distribution of voting machines?

681
00:41:05,260 --> 00:41:07,140
They have to deal with that stuff all the time.

682
00:41:07,600 --> 00:41:08,040
.

683
00:41:08,040 --> 00:41:15,100
I would, I would just say that voting at a precinct is an atomic transaction if it, from a, say, computing point of view.

684
00:41:15,460 --> 00:41:19,640
I mean, from beginning to the end of the transaction, you're not interfered with as a voter.

685
00:41:19,760 --> 00:41:25,740
And if I understand you correctly, yes, a voting transaction is intended to be atomic.

686
00:41:27,340 --> 00:41:27,780
.

687
00:41:27,780 --> 00:41:30,380
Maybe that sounds like, sounds like I didn't satisfy you.

688
00:41:31,080 --> 00:41:31,520
.

689
00:41:31,520 --> 00:41:32,020
Come talk to us.

690
00:41:32,280 --> 00:41:32,600
Candace?

691
00:41:34,320 --> 00:41:34,640
.

692
00:41:39,880 --> 00:41:40,200
.

693
00:41:50,800 --> 00:41:51,120
.

694
00:41:51,120 --> 00:41:51,400
.

695
00:41:51,400 --> 00:41:51,480
.

696
00:41:51,480 --> 00:41:51,660
.

697
00:41:53,840 --> 00:41:54,160
.

698
00:41:54,160 --> 00:41:54,260
.

699
00:41:54,260 --> 00:41:54,420
.

700
00:41:54,420 --> 00:41:54,600
.

701
00:41:54,600 --> 00:41:54,620
.

702
00:41:54,620 --> 00:41:54,640
.

703
00:41:54,640 --> 00:41:54,680
.

704
00:41:54,680 --> 00:41:54,840
.

705
00:42:12,960 --> 00:42:15,760
. .

706
00:42:18,720 --> 00:42:20,120
.

707
00:42:20,120 --> 00:42:20,680
.

708
00:42:20,680 --> 00:42:21,300
.

709
00:42:21,300 --> 00:42:21,320
.

710
00:42:21,320 --> 00:42:21,340
.

711
00:42:21,340 --> 00:42:21,700
.

712
00:42:21,700 --> 00:42:24,360
.

713
00:42:24,360 --> 00:42:24,420
.

714
00:42:28,440 --> 00:42:29,840
.

715
00:42:29,840 --> 00:42:31,260
.

716
00:42:31,260 --> 00:42:36,100
somehow. In some places it is counted by hand in very tiny places. But many places they

717
00:42:36,100 --> 00:42:40,920
use optical scanners, sometimes central count optical scanners, which I don't think we have

718
00:42:40,920 --> 00:42:46,680
any here. They're really curious machines. They're very high throughput, expensive, $100,000.

719
00:42:47,020 --> 00:42:51,900
You can think of like Pitney Bowes mail handler kind of things to very quickly go through

720
00:42:51,900 --> 00:42:58,220
these systems. And in fact, in the Ohio study from 2007, we installed Linux on the ES and

721
00:42:58,220 --> 00:43:02,800
S1 just to prove that it was a general purpose computing platform. And they do have

722
00:43:02,800 --> 00:43:09,280
vulnerabilities. And often, you know, the things like the tracking capabilities where you

723
00:43:11,080 --> 00:43:16,340
with your vote by mail ballot have something that allows you to check and see that your vote

724
00:43:16,340 --> 00:43:21,100
by mail ballot has been received, has then been counted and accepted, that's so crucial because

725
00:43:21,100 --> 00:43:25,940
often you can see something happen there like, hey, it was rejected for some reason. And

726
00:43:25,940 --> 00:43:28,200
then you have it. In many places you have a period of time where you have to wait for it to

727
00:43:28,200 --> 00:43:31,360
be corrected. You have a period of time to correct that stuff. What you may not know, and this

728
00:43:31,360 --> 00:43:36,300
happens with regular optical scan, too, is if that scanner correctly interpreted the marks you

729
00:43:36,300 --> 00:43:41,580
made on that ballot, right? And this is where the newer computer vision systems are a lot

730
00:43:41,580 --> 00:43:45,940
better at doing things that older systems couldn't do. For example, if you have an oval and

731
00:43:45,940 --> 00:43:50,040
someone circles the oval instead of filling it in or putting an X in it or something like that,

732
00:43:50,780 --> 00:43:55,440
older systems had a hard time catching this. It sounds like from what I'm hearing from people that

733
00:43:55,440 --> 00:43:58,180
newer systems detect marginalized marks and other things like that. And so, you know, it's a

734
00:43:58,180 --> 00:44:02,380
good thing that we're able to better count those. But it's not perfect. And that's why you need to

735
00:44:02,380 --> 00:44:08,360
have regular audits and actually compare some of these marginal marks and figure out if they would

736
00:44:08,360 --> 00:44:09,700
have affected the race.

737
00:44:11,240 --> 00:44:14,920
And Candace, you said something about correctability. Could you repeat the question?

738
00:44:15,220 --> 00:44:20,280
Correctability on a macro scale versus that of a limited evaluation. I understand what you're

739
00:44:20,280 --> 00:44:28,160
saying about correctability as an individual, though, is not possible. But what about in actual

740
00:44:28,180 --> 00:44:33,300
results, or in actual tallies of the results, that that could be a little bit more expensive?

741
00:44:34,700 --> 00:44:51,420
Well, so that's a complex question. So most of the time, there isn't a second process after the

742
00:44:51,420 --> 00:44:57,840
Canvas to check the initial Canvas results at all. Now, you've heard it several times today, people

743
00:44:57,840 --> 00:45:08,380
have recommended that we have risk-limiting audits of paper ballots. And the central idea there is

744
00:45:08,380 --> 00:45:16,100
that you would take a random sample of the paper ballots and check them by hand. And there's a

745
00:45:16,100 --> 00:45:23,320
statistical criterion to indicate whether or not this confirms the results from the original machine

746
00:45:23,320 --> 00:45:26,320
Canvas, or whether it disconfirms them.

747
00:45:26,520 --> 00:45:27,280
Okay.

748
00:45:27,840 --> 00:45:35,080
And the thought is, though the law has not followed this yet, is that if the risk-limiting audit

749
00:45:35,080 --> 00:45:44,800
afterwards determines that the results were originally incorrectly tabulated, either for procedural

750
00:45:44,800 --> 00:45:51,840
error or software error or malicious code or something like that, then the results of the

751
00:45:52,520 --> 00:45:57,500
risk-limiting audit would legally replace those of the machine counts. And then you would presumably,

752
00:45:57,500 --> 00:46:04,940
that would immediately presumably trigger an investigation, a forensic investigation, of the voting machines,

753
00:46:05,220 --> 00:46:15,940
and exactly why did they not match the results of human counts of these ballots. Right now, this law and these

754
00:46:15,940 --> 00:46:22,820
procedures are not in place, and we would, you know, it's one of the things that we are looking to try to institute in

755
00:46:22,820 --> 00:46:25,500
every state in the union as soon as we can.

756
00:46:26,320 --> 00:46:27,480
And for, you know, I think that's a good question. I think that's a good question. I think that's a good question.

757
00:46:27,480 --> 00:46:38,120
If anyone out there is a real math nerd, have I got something for you. Walter Mabane, who is a political scientist and a total math nerd, has a really cool election

758
00:46:38,120 --> 00:46:48,220
forensics course that all the materials are available online, and all he does is work with totals, with vote totals, at the precinct level or whatever level, and uses various kinds of

759
00:46:48,220 --> 00:46:57,360
statistical tests to spot tampering. And he's used these in many, many a foreign country. Things like Benford's Law, which is, I won't get into it, but look it up, it's fun, it's basically the

760
00:46:57,360 --> 00:47:06,140
distribution of the first number in a distribution of numbers, so like, the one in 1045, or something like that. And then there's a second order, Benford's Law, which is the second one.

761
00:47:06,160 --> 00:47:14,140
Anyway, these all have distributions that you can spot when someone is actually messing with stuff. But what can you do afterwards? You know, you're not going to actually, like,

762
00:47:14,580 --> 00:47:24,600
renormalize the results, like a census kind of a thing where you say, oh, we miscounted something, but we know how we miscounted and we're going to do that. Man, campaign lawyers will never let you get away with that.

763
00:47:25,780 --> 00:47:32,760
I did want to add one other thing, which is that the idea of doing forensic examinations of voting machines

764
00:47:32,760 --> 00:47:39,120
or the back-end systems after an election, even if something is suspected as going wrong,

765
00:47:39,340 --> 00:47:48,060
there are no procedures, there is no law to permit, let alone require, such investigations.

766
00:47:48,500 --> 00:47:51,320
It's just not possible under the law today.

767
00:47:51,500 --> 00:47:53,060
In fact, it would be totally prohibited.

768
00:47:53,060 --> 00:47:59,980
So this is a frontier that I think we have to change in the future.

769
00:48:00,400 --> 00:48:01,900
Yeah, I think we've got time for one more.

770
00:48:03,980 --> 00:48:04,440
Sir?

771
00:48:04,900 --> 00:48:09,440
Do we actually know the quality or magnitude of election temporary right now?

772
00:48:09,780 --> 00:48:10,500
Absolutely not.

773
00:48:13,540 --> 00:48:14,660
I mean, yeah.

774
00:48:16,760 --> 00:48:17,440
I don't know.

775
00:48:18,180 --> 00:48:19,980
There are a few things we can say.

776
00:48:20,160 --> 00:48:23,040
I mean, I mentioned earlier that most of the...

777
00:48:23,040 --> 00:48:28,160
Most of the cases of detected fraud, you know, detected and prosecuted,

778
00:48:28,240 --> 00:48:31,900
have involved either election officials or vote-by-mail.

779
00:48:32,320 --> 00:48:34,480
Now, that doesn't tell you how many cases there are.

780
00:48:34,640 --> 00:48:38,340
It just says almost all of those that we have are of that kind.

781
00:48:40,440 --> 00:48:48,600
The notion of voter ID that's been introduced in many states has been touted as a protection against vote fraud,

782
00:48:48,820 --> 00:48:51,700
but I think most of the people in this room know that that's totally ineffective,

783
00:48:52,040 --> 00:48:53,020
except against...

784
00:48:53,040 --> 00:48:53,120
I think most of the people in this room know that that's totally ineffective,

785
00:48:53,120 --> 00:48:55,340
except against one kind of fraud, which never happens,

786
00:48:55,460 --> 00:49:00,260
which is a voter walking in and impersonating another voter in that same precinct.

787
00:49:00,360 --> 00:49:01,680
It just doesn't happen.

788
00:49:03,400 --> 00:49:04,400
Thank you, everyone.

789
00:49:04,620 --> 00:49:05,380
Thank you.

790
00:49:05,620 --> 00:49:07,240
Thank you.

791
00:49:07,240 --> 00:49:07,700
Thank you.