1
00:00:02,680 --> 00:00:05,760
Speaker 1: Hi, it's Wescasova. We're taking a break this week, so

2
00:00:05,960 --> 00:00:09,320
Speaker 1: here's one of our favorite episodes you might have missed.

3
00:00:09,760 --> 00:00:16,480
Speaker 1: Thanks so much for listening. We've talked quite a lot

4
00:00:16,520 --> 00:00:20,000
Speaker 1: on this show lately about artificial intelligence because of how

5
00:00:20,079 --> 00:00:23,680
Speaker 1: quickly it's inserted itself into so many parts of our lives,

6
00:00:24,200 --> 00:00:27,520
Speaker 1: and how it's becoming more and more difficult to tell

7
00:00:27,720 --> 00:00:33,040
Speaker 1: what's real and what's fabricated. One place where that's increasingly

8
00:00:33,080 --> 00:00:37,320
Speaker 1: becoming a problem politics, especially now as the twenty twenty

9
00:00:37,320 --> 00:00:40,000
Speaker 1: four US presidential campaign heats up.

10
00:00:40,360 --> 00:00:45,920
Speaker 2: EI. Artificial intelligence is now hitting the US presidential campaign trail.

11
00:00:46,720 --> 00:00:50,960
Speaker 2: Donald Trump has been got in a series of aidep figs.

12
00:00:51,520 --> 00:00:55,600
Speaker 2: This time. Ron DeSantis is also in the spotlight. Reports

13
00:00:55,600 --> 00:00:59,680
Speaker 2: say DeSantis published fake images of Donald Trump and Anthony Fauci.

14
00:01:00,120 --> 00:01:05,400
Speaker 1: Already, candidates and their surrogates and even bad actors overseas

15
00:01:05,600 --> 00:01:07,920
Speaker 1: are reaching for every tool they can to try to

16
00:01:08,000 --> 00:01:12,280
Speaker 1: shape voters' perceptions and ultimately influence the outcome.

17
00:01:12,680 --> 00:01:15,839
Speaker 3: There is definitely an arms race when it comes to AI,

18
00:01:16,120 --> 00:01:19,319
Speaker 3: but there is an element of mutually assured destruction.

19
00:01:19,520 --> 00:01:22,200
Speaker 1: And the government is struggling to figure out how to

20
00:01:22,319 --> 00:01:24,880
Speaker 1: police the use of AI in campaigns.

21
00:01:25,319 --> 00:01:28,240
Speaker 4: What makes the FEC really interesting in this is that

22
00:01:28,280 --> 00:01:31,319
Speaker 4: they are by designing a partisan agency with a mix

23
00:01:31,319 --> 00:01:34,120
Speaker 4: of Republicans and Democrats, and so across the board you

24
00:01:34,160 --> 00:01:43,000
Speaker 4: really see them deadlocking on almost every issue.

25
00:01:44,360 --> 00:01:47,360
Speaker 1: I'm west casofa today on the Big Take, Bloomberg's Emily

26
00:01:47,440 --> 00:01:51,680
Speaker 1: Burnbaum and Laura Davison tell us how AI is transforming

27
00:01:52,040 --> 00:02:00,280
Speaker 1: and distorting our political reality. Emily, it seems like it

28
00:02:00,320 --> 00:02:04,880
Speaker 1: took about five minutes after AI really came forward as

29
00:02:04,880 --> 00:02:08,280
Speaker 1: a thing that was easily accessible for the political world

30
00:02:08,360 --> 00:02:11,240
Speaker 1: to say, hey, we can use this for all kinds

31
00:02:11,280 --> 00:02:14,560
Speaker 1: of reasons, some good and some really not so good.

32
00:02:15,960 --> 00:02:19,399
Speaker 3: Yeah, we're still really in the early stages of this.

33
00:02:19,560 --> 00:02:23,360
Speaker 3: We've yet to see the most dramatic examples of political

34
00:02:23,400 --> 00:02:27,120
Speaker 3: misinformation thanks to generative AI, but there are a couple

35
00:02:27,240 --> 00:02:31,520
Speaker 3: really big and well known examples that have been circulating.

36
00:02:31,760 --> 00:02:35,800
Speaker 3: So some far right politicians in Germany used AI to

37
00:02:36,120 --> 00:02:39,680
Speaker 3: generate images of immigrants that look very angry and violent.

38
00:02:40,400 --> 00:02:44,120
Speaker 3: No one was real, just you know, meant to incite

39
00:02:44,400 --> 00:02:48,519
Speaker 3: sort of nationalist fervor. Here back in the US, we're

40
00:02:48,560 --> 00:02:53,600
Speaker 3: seeing a lot more of that. So presidential candidate Ron DeSantis,

41
00:02:54,040 --> 00:02:58,560
Speaker 3: his campaign arm put out a video that included images

42
00:02:58,760 --> 00:03:03,880
Speaker 3: of Anthony Foul hugging former President Donald Trump and they

43
00:03:03,880 --> 00:03:08,040
Speaker 3: were not real. So Desantans's campaign didn't say that they

44
00:03:08,160 --> 00:03:13,040
Speaker 3: were AI generated, but experts kind of circled and identified

45
00:03:13,080 --> 00:03:16,520
Speaker 3: them as fake, and then within a couple hours there

46
00:03:16,600 --> 00:03:18,960
Speaker 3: was a Twitter note appended to it saying these are

47
00:03:19,000 --> 00:03:22,600
Speaker 3: not real. And the other big example is the Republican

48
00:03:22,680 --> 00:03:27,000
Speaker 3: National Committee put together an AI generated video that purported

49
00:03:27,040 --> 00:03:29,680
Speaker 3: to show, you know, here's the future that Biden wants,

50
00:03:29,720 --> 00:03:33,560
Speaker 3: and it was very dystopian and disturbing. They actually said

51
00:03:33,600 --> 00:03:34,800
Speaker 3: it was AI generated.

52
00:03:37,200 --> 00:03:38,960
Speaker 4: Yeah, this video was really quite disturbing.

53
00:03:38,960 --> 00:03:39,120
Speaker 2: You know.

54
00:03:39,200 --> 00:03:42,320
Speaker 4: It showed images of China attacking Taiwan. It showed the

55
00:03:42,360 --> 00:03:45,120
Speaker 4: streets of San Francisco with troops marching in as martial

56
00:03:45,200 --> 00:03:47,880
Speaker 4: law was taking over, and those looked very very real.

57
00:03:48,240 --> 00:03:50,480
Speaker 4: The one tailtale sign at the end of this video

58
00:03:50,640 --> 00:03:53,280
Speaker 4: that it was fake was it showed Joe Biden just

59
00:03:53,320 --> 00:03:56,400
Speaker 4: slumped over his desk, looking really dejected, and you could

60
00:03:56,400 --> 00:03:58,600
Speaker 4: see that his elbows weren't quite connecting with the edge

61
00:03:58,600 --> 00:04:01,040
Speaker 4: of the desk. So it also had a little disclaimer,

62
00:04:01,200 --> 00:04:04,000
Speaker 4: very very tiny in the corner saying this was AI generated.

63
00:04:04,040 --> 00:04:05,760
Speaker 4: But one of the things that consumers are going to

64
00:04:05,840 --> 00:04:08,440
Speaker 4: have to start doing is be really good about determining

65
00:04:08,480 --> 00:04:11,000
Speaker 4: what video is fake and what is real. And there's

66
00:04:11,040 --> 00:04:14,040
Speaker 4: always with AI, the technology is pretty good. But you know,

67
00:04:14,080 --> 00:04:17,479
Speaker 4: sometimes people have six fingers on the Fauci and Trump

68
00:04:17,560 --> 00:04:20,960
Speaker 4: hugging images. You know, it looked very airbrushed. Trump's hair

69
00:04:21,000 --> 00:04:23,640
Speaker 4: looked very weird in certain angles. So that's kind of

70
00:04:23,640 --> 00:04:26,159
Speaker 4: a telltale sign if you think something looks weird and

71
00:04:26,160 --> 00:04:27,600
Speaker 4: you're like, look at the hands, look at the feet,

72
00:04:27,600 --> 00:04:29,640
Speaker 4: look at the limbs, look at some of these features.

73
00:04:29,640 --> 00:04:32,159
Speaker 4: They look a little bit distorted, and that's usually a

74
00:04:32,200 --> 00:04:33,720
Speaker 4: sign that something is a little off.

75
00:04:34,320 --> 00:04:36,919
Speaker 1: I remember in the last election we started thinking about

76
00:04:36,960 --> 00:04:41,320
Speaker 1: deep face, these videos, these pictures which are indistinguishable from

77
00:04:41,560 --> 00:04:45,000
Speaker 1: what's real, and the technology wasn't quite there yet. And now,

78
00:04:45,040 --> 00:04:47,480
Speaker 1: of course we see every day that it is getting

79
00:04:47,520 --> 00:04:49,719
Speaker 1: much better. And I guess that raises this question that

80
00:04:49,880 --> 00:04:53,120
Speaker 1: as the campaign heats up, we're going to see so

81
00:04:53,360 --> 00:04:55,440
Speaker 1: many of these images that it's going to be really

82
00:04:55,520 --> 00:04:58,960
Speaker 1: hard to tell what's real and what's not correct.

83
00:04:59,000 --> 00:05:03,400
Speaker 4: And there's no regulation that requires any person, any advertiser,

84
00:05:03,440 --> 00:05:07,479
Speaker 4: any candidate to declare something as fake. Google has said

85
00:05:07,760 --> 00:05:09,560
Speaker 4: that in their tools that they are going to have

86
00:05:09,720 --> 00:05:11,960
Speaker 4: not a physical marker on it, but sort of a

87
00:05:12,000 --> 00:05:14,840
Speaker 4: digital marker, So if people go and reverse image search

88
00:05:14,880 --> 00:05:16,720
Speaker 4: it, it will flag it as something that's fake. But that's

89
00:05:16,880 --> 00:05:19,440
Speaker 4: how often do you go and actually reverse image search something.

90
00:05:20,000 --> 00:05:22,240
Speaker 4: Not that often, So it's going to really take a

91
00:05:22,240 --> 00:05:25,560
Speaker 4: lot of discerning, both from kind of the truth squatting

92
00:05:25,680 --> 00:05:28,400
Speaker 4: faction on Twitter as well as from individual consumers to

93
00:05:28,440 --> 00:05:30,240
Speaker 4: know if something's real or fake. And as we know,

94
00:05:30,279 --> 00:05:31,720
Speaker 4: a lot of people don't even know who's running for

95
00:05:31,720 --> 00:05:32,520
Speaker 4: president right now.

96
00:05:32,960 --> 00:05:37,480
Speaker 3: Yeah. Just on the topic of regulation not being in place,

97
00:05:37,800 --> 00:05:41,080
Speaker 3: the Federal Election Commission this month tried to vote on

98
00:05:41,279 --> 00:05:44,719
Speaker 3: a measure that would have enabled them to create rules

99
00:05:45,120 --> 00:05:49,839
Speaker 3: around AI generated political ads. Basically, the FEC would say

100
00:05:49,960 --> 00:05:53,440
Speaker 3: you have to declare when AI has generated your ad.

101
00:05:53,800 --> 00:05:56,760
Speaker 3: But they found themselves deadlocked. So there was a deep

102
00:05:56,800 --> 00:05:59,920
Speaker 3: disagreement among the commissioners, not even just a long partisan

103
00:06:00,080 --> 00:06:03,200
Speaker 3: lines about if they even have the authority to do that.

104
00:06:03,680 --> 00:06:09,279
Speaker 3: So the agencies that we have are currently deadlocked, unsure

105
00:06:09,279 --> 00:06:12,239
Speaker 3: how to proceed, and on Capitol Hill they're only beginning

106
00:06:12,240 --> 00:06:16,920
Speaker 3: to have these conversations.

107
00:06:16,120 --> 00:06:19,960
Speaker 1: Emily, why would the Federal Election Commission have any doubts

108
00:06:20,000 --> 00:06:22,039
Speaker 1: about what to do? Is there anyone in favor of

109
00:06:22,080 --> 00:06:25,320
Speaker 1: allowing fake stuff to be spread all over the place

110
00:06:25,320 --> 00:06:26,479
Speaker 1: without having to be labeled.

111
00:06:27,000 --> 00:06:30,239
Speaker 3: It's less about the actual measure and more about if

112
00:06:30,360 --> 00:06:33,240
Speaker 3: the FEC has the authority to do this. So we

113
00:06:33,279 --> 00:06:36,440
Speaker 3: actually run into this with tech related issues and our

114
00:06:36,480 --> 00:06:39,920
Speaker 3: government all the time. Most of these agencies weren't set

115
00:06:40,000 --> 00:06:44,520
Speaker 3: up to deal with issues like deep fakes, misinformation scams online,

116
00:06:44,640 --> 00:06:47,080
Speaker 3: and so they're having to adapt in real time, and

117
00:06:47,120 --> 00:06:50,880
Speaker 3: the FEC, some of the commissioners at least say that

118
00:06:50,920 --> 00:06:54,120
Speaker 3: they just don't have power over this part of the lot.

119
00:06:54,560 --> 00:06:58,359
Speaker 3: There are a lot of protections for political speech that

120
00:06:58,600 --> 00:07:02,520
Speaker 3: is an important bedrock of communication in America, so that

121
00:07:02,640 --> 00:07:05,040
Speaker 3: makes this issue even more difficult to deal with.

122
00:07:05,560 --> 00:07:08,480
Speaker 4: What makes the FEC really interesting in this is that

123
00:07:08,480 --> 00:07:11,520
Speaker 4: they are by designing a partisan agency with a mix

124
00:07:11,560 --> 00:07:14,360
Speaker 4: of Republicans and Democrats, and so across the board you

125
00:07:14,400 --> 00:07:17,840
Speaker 4: really see them deadlocking on almost every issue. The FEC

126
00:07:17,880 --> 00:07:21,000
Speaker 4: has had very little success doing anything, which just points

127
00:07:21,000 --> 00:07:23,880
Speaker 4: to that the FEC realistically isn't going to be able

128
00:07:23,920 --> 00:07:25,880
Speaker 4: to take this one on based on all the other

129
00:07:26,000 --> 00:07:27,680
Speaker 4: issues facing the agency.

130
00:07:27,800 --> 00:07:31,640
Speaker 1: And what about the candidates themselves. It's easy to put

131
00:07:31,680 --> 00:07:34,840
Speaker 1: something out that's fake about your opponent and maybe it

132
00:07:34,840 --> 00:07:36,760
Speaker 1: gets you all kinds of attention, but then you know

133
00:07:36,800 --> 00:07:38,880
Speaker 1: they're going to be doing it about you too, and

134
00:07:38,960 --> 00:07:42,600
Speaker 1: so there's this kind of AI fakery arms race that

135
00:07:42,640 --> 00:07:45,160
Speaker 1: we seem to be setting up. Is there any recognition

136
00:07:45,200 --> 00:07:46,840
Speaker 1: on the part of the campaigns that this is a

137
00:07:46,880 --> 00:07:48,560
Speaker 1: really bad line to cross?

138
00:07:49,920 --> 00:07:53,080
Speaker 3: So there is definitely an arms race when it comes

139
00:07:53,160 --> 00:07:56,120
Speaker 3: to AI, But there is an element of mutually assured

140
00:07:56,160 --> 00:08:00,200
Speaker 3: destruction where as soon as one campaign really goes out

141
00:08:00,360 --> 00:08:02,840
Speaker 3: put some words in their opponent's mouth that makes them

142
00:08:02,840 --> 00:08:07,600
Speaker 3: look bad, distributes images videos widely, then that sort of

143
00:08:07,800 --> 00:08:10,800
Speaker 3: opens this whole can of worms, and so there's going

144
00:08:10,880 --> 00:08:13,600
Speaker 3: to be tons of these kinds of ads, and actually,

145
00:08:13,600 --> 00:08:17,120
Speaker 3: no one really wants that. So there's some internal reckoning

146
00:08:17,200 --> 00:08:21,440
Speaker 3: that's happening among political consultants and campaigns right now.

147
00:08:21,640 --> 00:08:23,840
Speaker 1: I guess they've been coming from these outside groups that

148
00:08:23,920 --> 00:08:26,280
Speaker 1: support campaigns so that they're able to have a certain

149
00:08:26,320 --> 00:08:30,720
Speaker 1: amount of deniability even though it's very clear which candidate

150
00:08:30,720 --> 00:08:31,400
Speaker 1: they're supporting.

151
00:08:31,840 --> 00:08:35,199
Speaker 4: But it creates for what we'd call like dark money groups.

152
00:08:35,200 --> 00:08:37,520
Speaker 4: So five oh on and C four political advacy groups.

153
00:08:37,800 --> 00:08:39,760
Speaker 4: This is a real wide opening for you know, groups

154
00:08:39,760 --> 00:08:42,600
Speaker 4: that aren't necessarily representing a specific person but an interest

155
00:08:42,679 --> 00:08:45,840
Speaker 4: area or a larger party. The dark money groups, they

156
00:08:45,840 --> 00:08:47,640
Speaker 4: can go out and take all these images and they're

157
00:08:47,640 --> 00:08:50,480
Speaker 4: super cheap and easy to produce, so they can react

158
00:08:50,520 --> 00:08:52,720
Speaker 4: in real time. If something is said in a debate

159
00:08:52,760 --> 00:08:54,520
Speaker 4: about a candidate they don't like, they can have an

160
00:08:54,520 --> 00:08:56,960
Speaker 4: image or a video out there online by the end

161
00:08:57,000 --> 00:08:58,840
Speaker 4: of the debate. So that's what we're going to start

162
00:08:58,880 --> 00:09:00,960
Speaker 4: to see this really rapid response. You know, what we

163
00:09:01,040 --> 00:09:03,040
Speaker 4: used to just see with with tweets or emails can

164
00:09:03,120 --> 00:09:05,640
Speaker 4: now be done with images and videos.

165
00:09:08,200 --> 00:09:10,920
Speaker 1: Laurie, we've been talking about how we're seeing this happening

166
00:09:10,960 --> 00:09:14,719
Speaker 1: at the national level, at the presidential level, but it's

167
00:09:14,760 --> 00:09:18,440
Speaker 1: also filtering down throughout politics in the US.

168
00:09:19,080 --> 00:09:19,320
Speaker 2: Yeah.

169
00:09:19,360 --> 00:09:21,679
Speaker 4: So AI is really cheap, and so that makes it

170
00:09:21,720 --> 00:09:24,680
Speaker 4: a really great tool for these smaller campaigns that don't

171
00:09:24,679 --> 00:09:26,040
Speaker 4: have a lot of money, that don't have a lot

172
00:09:26,040 --> 00:09:29,360
Speaker 4: of staff. So there's been in some mayoral elections recently,

173
00:09:29,559 --> 00:09:32,800
Speaker 4: AI used both by campaigns to sort of support themselves

174
00:09:32,920 --> 00:09:35,560
Speaker 4: as well as attack ads. There was a really great

175
00:09:35,559 --> 00:09:38,200
Speaker 4: example in Shreveport, Louisiana, where there was an attack ad

176
00:09:38,280 --> 00:09:41,920
Speaker 4: against the incumbent mayor, Adrian Perkins, and it showed this

177
00:09:42,040 --> 00:09:44,280
Speaker 4: scene where it was you know, it was his face

178
00:09:44,400 --> 00:09:46,840
Speaker 4: on a body and his voice was talking, his mouth

179
00:09:46,960 --> 00:09:48,480
Speaker 4: was moving. It wasn't him, it was an actor, and

180
00:09:48,520 --> 00:09:50,560
Speaker 4: then they had used AI to create it to make

181
00:09:50,600 --> 00:09:52,800
Speaker 4: it look like him. He ended up losing this race,

182
00:09:52,840 --> 00:09:55,280
Speaker 4: not necessarily because of this ad, but just a really

183
00:09:55,280 --> 00:09:57,760
Speaker 4: great example of how, you know, someone a developer in

184
00:09:57,800 --> 00:10:00,440
Speaker 4: town who didn't particularly care for the mayor come in

185
00:10:00,480 --> 00:10:03,079
Speaker 4: and create this ad really quickly, easily and cheaply, and

186
00:10:03,640 --> 00:10:05,520
Speaker 4: get it on the airwaves and really make a splash

187
00:10:05,520 --> 00:10:08,320
Speaker 4: with it. In Toronto we also saw the same thing

188
00:10:08,360 --> 00:10:12,360
Speaker 4: in their mayoral race, where candidate Anthony Fury used artificial

189
00:10:12,400 --> 00:10:15,080
Speaker 4: intelligence images in sort of his campaign materials, you know,

190
00:10:15,160 --> 00:10:18,440
Speaker 4: showing packages of crack pipe kits that were sponsored by

191
00:10:18,440 --> 00:10:20,360
Speaker 4: the City of Toronto, you know, using this to criticize

192
00:10:20,440 --> 00:10:21,959
Speaker 4: policies of his opponents.

193
00:10:22,320 --> 00:10:26,160
Speaker 1: Emily, you talked about how difficult it is for the

194
00:10:26,200 --> 00:10:28,400
Speaker 1: government to try and figure out how and whether they

195
00:10:28,440 --> 00:10:32,600
Speaker 1: can regulate this, is there anything illegal about a candidate

196
00:10:33,320 --> 00:10:38,080
Speaker 1: making another candidate appear to be doing something completely false?

197
00:10:38,840 --> 00:10:43,000
Speaker 3: One use of AI that could be illegal, and we

198
00:10:43,120 --> 00:10:46,200
Speaker 3: can remember this is still an untested area of the

199
00:10:46,280 --> 00:10:49,199
Speaker 3: law in many ways, so we're not positive. But one

200
00:10:49,520 --> 00:10:52,320
Speaker 3: thing that could be illegal is if you use a

201
00:10:52,360 --> 00:10:58,200
Speaker 3: candidate's likeness to get money that the candidate doesn't actually receive.

202
00:10:58,720 --> 00:11:01,960
Speaker 3: So that is one thing that the FBC was discussing

203
00:11:02,240 --> 00:11:05,480
Speaker 3: that seems like a clear violation of our campaign laws

204
00:11:05,559 --> 00:11:08,600
Speaker 3: when you are not allowed to impersonate anyone else and

205
00:11:08,640 --> 00:11:11,680
Speaker 3: not allowed to lie about where money is going. There

206
00:11:11,720 --> 00:11:15,240
Speaker 3: are also laws cropping up in the states across the

207
00:11:15,280 --> 00:11:20,000
Speaker 3: country that ban certain kinds of quote unquote deep fakes.

208
00:11:20,040 --> 00:11:23,800
Speaker 3: So a lot of those laws pertained to pornography. So

209
00:11:23,840 --> 00:11:27,080
Speaker 3: there have been some reports that show over ninety percent

210
00:11:27,320 --> 00:11:30,760
Speaker 3: of the deep fakes on the Internet are pornographic images,

211
00:11:31,120 --> 00:11:36,240
Speaker 3: mostly featuring women, and so a bunch of states have

212
00:11:37,160 --> 00:11:39,840
Speaker 3: rushed in to fill this gap in the law, basically

213
00:11:39,920 --> 00:11:43,920
Speaker 3: just improving victims' rights, so people who were featured in

214
00:11:43,960 --> 00:11:48,080
Speaker 3: pornographic images on the Internet can sue and have the

215
00:11:48,200 --> 00:11:51,000
Speaker 3: right to stop the proliferation of those But there are

216
00:11:51,080 --> 00:11:55,280
Speaker 3: far fewer rules about election related deep fakes. There are

217
00:11:55,360 --> 00:12:01,199
Speaker 3: some they probably won't withstand legal challenges, So it's still

218
00:12:01,280 --> 00:12:04,600
Speaker 3: kind of complicated whenever you try to touch political speech,

219
00:12:04,960 --> 00:12:07,680
Speaker 3: but there are clearer violations.

220
00:12:08,400 --> 00:12:11,920
Speaker 1: After the break. How AI is changing the way campaigns

221
00:12:12,160 --> 00:12:13,559
Speaker 1: get their message.

222
00:12:13,200 --> 00:12:24,840
Speaker 5: Out, Laurie, you hinted a little bit before about how

223
00:12:24,880 --> 00:12:28,679
Speaker 5: campaigns are using AI for non nefarious reasons, that it's

224
00:12:28,720 --> 00:12:31,520
Speaker 5: just like become a useful tool being used by campaigns

225
00:12:31,559 --> 00:12:33,839
Speaker 5: the way it's used by a lot of different industries.

226
00:12:34,240 --> 00:12:36,280
Speaker 4: Yeah, so there's a bunch of different things that campaigns

227
00:12:36,280 --> 00:12:38,319
Speaker 4: are using and some that people would kind of clearly

228
00:12:38,360 --> 00:12:41,680
Speaker 4: put into a non nefarious camp. So things like helping

229
00:12:41,960 --> 00:12:44,680
Speaker 4: use AI to analyze voter roles of people that they

230
00:12:44,720 --> 00:12:48,400
Speaker 4: should be targeting, organizing some of their data, figuring out

231
00:12:48,400 --> 00:12:52,040
Speaker 4: what kind of messages would resonate with which voters. Another

232
00:12:52,120 --> 00:12:54,720
Speaker 4: thing that they're doing is using tools like chat GBT,

233
00:12:54,880 --> 00:12:57,640
Speaker 4: not necessarily chat GPT because that tool has some restrictions

234
00:12:57,679 --> 00:13:00,360
Speaker 4: on political speech, but to write the first draft of

235
00:13:00,440 --> 00:13:04,160
Speaker 4: press releases of fundraising emails. And they found a really

236
00:13:04,200 --> 00:13:08,120
Speaker 4: great cost savings instead of having five people who are

237
00:13:08,200 --> 00:13:11,240
Speaker 4: copywriters on a campaign working on all these messages that

238
00:13:11,280 --> 00:13:14,040
Speaker 4: you'd have one person that is just telling the tool, okay,

239
00:13:14,080 --> 00:13:18,000
Speaker 4: write a fundraising email with a focus on democracy, one

240
00:13:18,040 --> 00:13:19,760
Speaker 4: with a focus on the environment, right, one with a

241
00:13:19,800 --> 00:13:24,040
Speaker 4: focus on defeating Republicans, for example, And then they can

242
00:13:24,120 --> 00:13:26,200
Speaker 4: get something on the page and then just edit it

243
00:13:26,240 --> 00:13:29,760
Speaker 4: and it goes much much quicker. The one campaign consultant

244
00:13:29,800 --> 00:13:32,080
Speaker 4: described this as solving the blank page problem. Anyone who's

245
00:13:32,080 --> 00:13:34,439
Speaker 4: ever written anything can probably sympathize with that if it's

246
00:13:34,480 --> 00:13:36,120
Speaker 4: much easier to get something on the paper and then

247
00:13:36,200 --> 00:13:39,160
Speaker 4: edit it versus just starting from scratch. So they're finding

248
00:13:39,160 --> 00:13:41,200
Speaker 4: really big cost savings in that of they're able to

249
00:13:41,679 --> 00:13:44,640
Speaker 4: both raise more money because they're more effectively targeting people

250
00:13:44,679 --> 00:13:47,120
Speaker 4: as well as cut down on staff cost for that

251
00:13:47,200 --> 00:13:50,720
Speaker 4: particular function of writing. That's been really widespread, and one

252
00:13:50,840 --> 00:13:54,000
Speaker 4: consultant estimated about fifty percent of campaigns are doing this,

253
00:13:54,080 --> 00:13:56,480
Speaker 4: but expect that number to be somewhere closer to one

254
00:13:56,559 --> 00:13:58,320
Speaker 4: hundred by the end of the twenty twenty four cycle.

255
00:13:58,760 --> 00:14:02,240
Speaker 3: But Congress just said new rules on how staffers can

256
00:14:02,360 --> 00:14:05,920
Speaker 3: use chat GPT, and the main concern is that a

257
00:14:05,960 --> 00:14:10,440
Speaker 3: lot of what you put into chat GPT can become

258
00:14:10,720 --> 00:14:13,360
Speaker 3: available in other ways. You know, there are a lot

259
00:14:13,400 --> 00:14:17,200
Speaker 3: of privacy concerns around chat GPT, So the new rules say,

260
00:14:17,679 --> 00:14:20,080
Speaker 3: first of all, you have to use the paid for

261
00:14:20,360 --> 00:14:24,400
Speaker 3: version of chat GPT, which is better protective of privacy.

262
00:14:24,800 --> 00:14:28,520
Speaker 3: Second of all, you can't put confidential information into chat GPT,

263
00:14:28,920 --> 00:14:32,120
Speaker 3: which congressional staffers are often working with. So they're kind

264
00:14:32,200 --> 00:14:34,760
Speaker 3: of like making it up as they go along. But

265
00:14:35,000 --> 00:14:38,320
Speaker 3: congressional staffers work all day and night. They are never

266
00:14:38,400 --> 00:14:40,400
Speaker 3: off the clock, and so anything that cuts back on

267
00:14:40,640 --> 00:14:41,960
Speaker 3: time is really valuable.

268
00:14:45,040 --> 00:14:49,040
Speaker 1: Open AI. The makers of chat GPT, they are kind

269
00:14:49,040 --> 00:14:52,720
Speaker 1: of alarmed at the way campaigns could be using their

270
00:14:52,760 --> 00:14:54,920
Speaker 1: technology and or trying to figure out a way to

271
00:14:55,000 --> 00:14:55,480
Speaker 1: limit it.

272
00:14:55,960 --> 00:14:58,280
Speaker 4: If you put in really anything related to politics and

273
00:14:58,320 --> 00:15:01,680
Speaker 4: chat GEPT, they're sort of the standard disclaimer language coming

274
00:15:01,760 --> 00:15:03,320
Speaker 4: up of saying, you know, look, I don't have an

275
00:15:03,320 --> 00:15:06,040
Speaker 4: opinion on this. There's nuance to this, and so chat

276
00:15:06,080 --> 00:15:09,800
Speaker 4: gibt is being very careful with anything related to politics.

277
00:15:09,840 --> 00:15:11,560
Speaker 4: But if you push it a little bit, like if

278
00:15:11,560 --> 00:15:14,160
Speaker 4: you type who are the wokest Democrats, it will name

279
00:15:14,360 --> 00:15:19,480
Speaker 4: Alexandria Kazi Kortez, Rashida talib Aana Presley Bilhan, omar interesting

280
00:15:19,480 --> 00:15:20,960
Speaker 4: that all like the top four or five names that

281
00:15:21,000 --> 00:15:23,560
Speaker 4: came up were all women. Sort of treading this careful line,

282
00:15:23,600 --> 00:15:26,360
Speaker 4: it's not completely staying out of the political arena. So

283
00:15:26,480 --> 00:15:29,840
Speaker 4: for a lot of political things, chat gibt has basically

284
00:15:29,920 --> 00:15:33,800
Speaker 4: stopped cataloging language from after September twenty twenty one, So

285
00:15:33,880 --> 00:15:36,600
Speaker 4: when you ask chat gibt a political question will be like,

286
00:15:36,600 --> 00:15:39,040
Speaker 4: here is my information as of September twenty twenty one.

287
00:15:39,280 --> 00:15:40,920
Speaker 4: You may need to do more research if you want

288
00:15:40,920 --> 00:15:43,160
Speaker 4: something more recent. So it's kind of just clamping down

289
00:15:43,200 --> 00:15:45,440
Speaker 4: and not feeding the beast, so to speak, with more

290
00:15:45,440 --> 00:15:48,160
Speaker 4: information to make it sort of an obsolete tool when

291
00:15:48,160 --> 00:15:49,280
Speaker 4: it comes to political speech.

292
00:15:49,680 --> 00:15:54,000
Speaker 3: They're also putting together something called classifiers, so open ai

293
00:15:54,240 --> 00:15:57,640
Speaker 3: is able to see what people put into chat gpt,

294
00:15:58,160 --> 00:16:00,880
Speaker 3: and so they're working hard now to figure out what

295
00:16:00,960 --> 00:16:05,840
Speaker 3: are the classifiers or identifying freezes that people use, where

296
00:16:05,920 --> 00:16:07,760
Speaker 3: when they're using those, you know that it's for a

297
00:16:07,800 --> 00:16:11,080
Speaker 3: political purpose, And then if that's happening, how can you

298
00:16:11,320 --> 00:16:15,040
Speaker 3: control what chat gpt says in response, they really don't

299
00:16:15,040 --> 00:16:18,240
Speaker 3: want their product to be used to create lies and

300
00:16:18,280 --> 00:16:19,840
Speaker 3: misinformation that they're blamed for.

301
00:16:20,680 --> 00:16:22,640
Speaker 4: To sort of add those. Chat GIBT is sort of

302
00:16:22,680 --> 00:16:26,200
Speaker 4: the big juggernaut and the industry of generative text AI,

303
00:16:26,480 --> 00:16:28,800
Speaker 4: but there's so many other tools that are being developed

304
00:16:28,840 --> 00:16:32,240
Speaker 4: and tools that are being developed specifically for political uses.

305
00:16:32,280 --> 00:16:35,000
Speaker 4: So usually these are paid tools that campaigns would subscribe to.

306
00:16:35,360 --> 00:16:37,640
Speaker 4: But just because chat GBT is sort of clamping down

307
00:16:37,640 --> 00:16:40,120
Speaker 4: on this, it doesn't mean that other actors can go

308
00:16:40,160 --> 00:16:43,080
Speaker 4: buy these tools that are specifically designed for this to

309
00:16:43,720 --> 00:16:45,480
Speaker 4: create some political misinformation.

310
00:16:46,320 --> 00:16:50,520
Speaker 1: Laura earlier Emily painted this picture of mutual assured destruction

311
00:16:50,720 --> 00:16:53,160
Speaker 1: where if one candidate attacks the other and the other

312
00:16:53,280 --> 00:16:56,040
Speaker 1: attacks back, then everybody gets hurt. And in you're reporting,

313
00:16:56,080 --> 00:17:00,000
Speaker 1: you found that both Democrats and Republicans are using this technology,

314
00:17:00,080 --> 00:17:01,400
Speaker 1: but in slightly different ways.

315
00:17:02,520 --> 00:17:04,800
Speaker 4: So what we've seen in terms of the visual and

316
00:17:04,920 --> 00:17:07,800
Speaker 4: video side as Republicans using this more. We saw big

317
00:17:07,840 --> 00:17:10,600
Speaker 4: first video ad was the RNC attacking Biden. We saw

318
00:17:10,680 --> 00:17:14,040
Speaker 4: DeSantis attacking Trump with this. But both parties are definitely

319
00:17:14,119 --> 00:17:16,679
Speaker 4: using it, and probably Democrats are using it more. When

320
00:17:16,720 --> 00:17:19,239
Speaker 4: it comes to the text and generative AI, they have

321
00:17:19,400 --> 00:17:22,840
Speaker 4: firms that are using the text side a lot more.

322
00:17:23,040 --> 00:17:26,000
Speaker 4: That's really widespread among campaigns. This is a really hard

323
00:17:26,000 --> 00:17:28,560
Speaker 4: thing that's hard to track you know exactly how widespread

324
00:17:28,560 --> 00:17:32,119
Speaker 4: it is within either party. Democrats are definitely more cautious

325
00:17:32,320 --> 00:17:35,399
Speaker 4: on using the visual and image side. They kind of

326
00:17:35,440 --> 00:17:37,800
Speaker 4: like to say, look, we're being responsible here, We're not

327
00:17:37,840 --> 00:17:42,320
Speaker 4: going to create misinformation or deep fakes. Republicans, particularly because

328
00:17:42,320 --> 00:17:44,640
Speaker 4: they're going into a contested primary, so they have more

329
00:17:44,680 --> 00:17:46,600
Speaker 4: going on right now, but they're a little bit more

330
00:17:46,600 --> 00:17:49,040
Speaker 4: willing to try out this tool and see how it works,

331
00:17:49,040 --> 00:17:51,600
Speaker 4: while Democrats are a little bit hanging back on the sidelines.

332
00:17:52,080 --> 00:17:55,080
Speaker 3: I spoke to a spokesperson for the Democratic National Committee

333
00:17:55,359 --> 00:18:00,800
Speaker 3: who said Democrats are wary of misinformation. There's a history

334
00:18:00,840 --> 00:18:04,560
Speaker 3: there that goes back to the twenty sixteen election. There's

335
00:18:04,800 --> 00:18:09,480
Speaker 3: a whole team dedicated to stopping misinformation at the DNC,

336
00:18:10,080 --> 00:18:13,000
Speaker 3: So that makes them extra cautious as they begin to

337
00:18:13,080 --> 00:18:16,800
Speaker 3: implement AI slowly into things like fundraising.

338
00:18:19,080 --> 00:18:22,120
Speaker 1: Emily, that's an interesting point that the Democrats are saying

339
00:18:22,119 --> 00:18:23,959
Speaker 1: that they're going to really need to be monitoring this.

340
00:18:24,040 --> 00:18:25,560
Speaker 1: Do you think that we're going to have now this

341
00:18:25,600 --> 00:18:30,560
Speaker 1: whole campaign industry where groups are doing AI checks similar

342
00:18:30,600 --> 00:18:32,200
Speaker 1: to the way we have fact checkers.

343
00:18:32,920 --> 00:18:38,520
Speaker 3: Yeah, there is this anxiety and wariness among political consultants

344
00:18:38,720 --> 00:18:42,520
Speaker 3: about how AI is going to be deployed. So the

345
00:18:42,680 --> 00:18:48,040
Speaker 3: trade Association for Political Consultants, which exists, shockingly came out

346
00:18:48,080 --> 00:18:51,280
Speaker 3: with this statement saying, we are against the use of

347
00:18:51,320 --> 00:18:54,800
Speaker 3: deep fakes in political campaigns and we really shouldn't be

348
00:18:54,800 --> 00:18:58,640
Speaker 3: messing too much with this technology. It's quite dangerous. So

349
00:18:58,840 --> 00:19:02,800
Speaker 3: I do think that incre there is desire to call

350
00:19:02,880 --> 00:19:07,359
Speaker 3: out misinformation, to call out when generative AI is being used.

351
00:19:07,680 --> 00:19:10,560
Speaker 3: I also think that a lot of technology companies are

352
00:19:10,600 --> 00:19:13,160
Speaker 3: about to make a lot of money off of basically

353
00:19:13,240 --> 00:19:16,640
Speaker 3: tools that allow you to identify what's real. Those are

354
00:19:16,720 --> 00:19:20,760
Speaker 3: going to be commonly used, at least among political campaigns,

355
00:19:20,760 --> 00:19:22,200
Speaker 3: if not the general population.

356
00:19:23,280 --> 00:19:26,360
Speaker 1: When we come back, what can we do to avoid

357
00:19:26,400 --> 00:19:39,520
Speaker 1: getting duped? So what are we supposed to do about this?

358
00:19:39,840 --> 00:19:44,720
Speaker 1: Just as citizens watching a campaign trying to make good decisions,

359
00:19:44,920 --> 00:19:49,520
Speaker 1: trying to figure out what's real and what's not. I mean, Laura,

360
00:19:49,640 --> 00:19:52,760
Speaker 1: you kind of jokingly at the beginning said, oh, look

361
00:19:52,800 --> 00:19:55,760
Speaker 1: at the handslet's eve it as six fingers. But this

362
00:19:55,840 --> 00:19:58,720
Speaker 1: is a pretty serious thing. How are we going to

363
00:19:58,760 --> 00:20:01,080
Speaker 1: be able to discern what's real and what's not.

364
00:20:02,240 --> 00:20:04,240
Speaker 4: This is probably when tech companies are going to have

365
00:20:04,280 --> 00:20:07,160
Speaker 4: to start stepping in. We've seen Twitter and Facebook come

366
00:20:07,200 --> 00:20:11,480
Speaker 4: in with labels on misinformation regarding vaccines or other things.

367
00:20:11,600 --> 00:20:14,680
Speaker 4: Regarding the twenty twenty election. We're probably going to see

368
00:20:14,680 --> 00:20:16,199
Speaker 4: those start to come in. I don't know that the

369
00:20:16,240 --> 00:20:19,240
Speaker 4: tech companies have They don't have fully fledged policies yet

370
00:20:19,280 --> 00:20:21,200
Speaker 4: on how to do that, and part of it is

371
00:20:21,200 --> 00:20:24,280
Speaker 4: that it is really difficult to check. We see companies

372
00:20:24,400 --> 00:20:26,560
Speaker 4: like Google saying, Okay, we're going to put these markers

373
00:20:26,800 --> 00:20:29,640
Speaker 4: in these images. That's probably the next step that we're

374
00:20:29,640 --> 00:20:31,920
Speaker 4: going to see of each company, because they want to

375
00:20:31,960 --> 00:20:34,120
Speaker 4: be seen as responsible and because they don't want an

376
00:20:34,160 --> 00:20:36,359
Speaker 4: image created on their platform to be the thing that

377
00:20:36,560 --> 00:20:39,160
Speaker 4: blows up an election, for example. You're going to see

378
00:20:39,280 --> 00:20:42,120
Speaker 4: pressure in the industry to do more, but it's right

379
00:20:42,160 --> 00:20:44,800
Speaker 4: now it's really incumbent upon the viewer. So it is

380
00:20:44,880 --> 00:20:46,359
Speaker 4: kind of looking at the image, look to see if

381
00:20:46,359 --> 00:20:48,840
Speaker 4: there's a little tag in the corner that mentions it's

382
00:20:48,880 --> 00:20:51,240
Speaker 4: AI generated, or a screen at the very end of

383
00:20:51,280 --> 00:20:53,440
Speaker 4: an ad that says this was created by AI.

384
00:20:53,840 --> 00:20:56,760
Speaker 1: You know, in previous campaigns, we've seen Facebook, We've seen

385
00:20:56,800 --> 00:21:00,199
Speaker 1: Twitter make big public announcements they've hired all of these

386
00:21:00,240 --> 00:21:02,119
Speaker 1: people who are going to be monitoring, they're going to

387
00:21:02,119 --> 00:21:04,159
Speaker 1: try to crack down on fakery, and yet they just

388
00:21:04,359 --> 00:21:06,959
Speaker 1: could not possibly do it. In the case of Twitter,

389
00:21:07,080 --> 00:21:10,480
Speaker 1: we're seeing all kinds of fakery all over the platform,

390
00:21:10,600 --> 00:21:14,480
Speaker 1: especially since Elon Musk took over the company and kind

391
00:21:14,480 --> 00:21:17,600
Speaker 1: of opened it up in a way that previously people

392
00:21:17,600 --> 00:21:18,760
Speaker 1: couldn't post stuff on it.

393
00:21:19,600 --> 00:21:23,160
Speaker 3: Yes, but we're also seeing tools like Twitter's community notes,

394
00:21:23,200 --> 00:21:27,479
Speaker 3: which allows everyday users to weigh in on whether something

395
00:21:27,600 --> 00:21:31,879
Speaker 3: is true or false, that was already under construction before

396
00:21:31,920 --> 00:21:35,000
Speaker 3: Elon Musk came in. It's actually proven to be a

397
00:21:35,040 --> 00:21:40,760
Speaker 3: really effective obstacle for the spread of generative AI misinformation.

398
00:21:40,960 --> 00:21:43,200
Speaker 3: So if we think about the ad Byron de Sampas's

399
00:21:43,200 --> 00:21:46,440
Speaker 3: campaign within hours that had a Twitter note on it

400
00:21:46,680 --> 00:21:49,840
Speaker 3: saying tech experts have looked at this and these images

401
00:21:49,880 --> 00:21:53,040
Speaker 3: are not real. So if you think about what you

402
00:21:53,119 --> 00:21:56,439
Speaker 3: can do as an individual, it's a mitzvah to just

403
00:21:56,720 --> 00:22:02,040
Speaker 3: get out the word that something is fake, that maybe

404
00:22:02,359 --> 00:22:08,360
Speaker 3: tech experts are weighing in and just spreading that information together.

405
00:22:08,840 --> 00:22:10,960
Speaker 3: People create media literacy that way.

406
00:22:11,640 --> 00:22:15,880
Speaker 4: And the thing too is ethicists are less concerned about

407
00:22:15,880 --> 00:22:18,280
Speaker 4: people like Biden and Donald Trump that have swarms of

408
00:22:18,320 --> 00:22:21,000
Speaker 4: press following everywhere they go and everything they do. But

409
00:22:21,080 --> 00:22:24,440
Speaker 4: on the local level, where local news isn't as strong,

410
00:22:24,480 --> 00:22:27,399
Speaker 4: where there are people and candidates and politicians who aren't

411
00:22:27,400 --> 00:22:31,080
Speaker 4: necessarily followed as closely, it's way easier to spread misinformation

412
00:22:31,160 --> 00:22:33,720
Speaker 4: in these areas where there is more of a news vacuum.

413
00:22:34,080 --> 00:22:36,560
Speaker 1: Emily Laura, thanks so much for coming on the show.

414
00:22:36,920 --> 00:22:39,240
Speaker 3: Thanks so much for having us, Thank.

415
00:22:39,080 --> 00:22:42,160
Speaker 1: You, thanks for listening to us here at The Big Take.

416
00:22:42,320 --> 00:22:45,680
Speaker 1: It's a daily podcast from Bloomberg and iHeartRadio. For more

417
00:22:45,680 --> 00:22:49,919
Speaker 1: shows from iHeartRadio, visit the iHeartRadio app, Apple Podcasts, or

418
00:22:49,960 --> 00:22:52,560
Speaker 1: wherever you listen, and we'd love to hear from you.

419
00:22:52,880 --> 00:22:56,040
Speaker 1: Email us questions or comments to Big Take at Bloomberg

420
00:22:56,080 --> 00:22:59,399
Speaker 1: dot net. The supervising producer of The Big Take is

421
00:22:59,480 --> 00:23:04,240
Speaker 1: Vicky verg Our senior producer is Catherine Fink. Federica Romanello

422
00:23:04,359 --> 00:23:09,080
Speaker 1: is our producer. Our associate producer is Zeneb Sidiki. Rafael

423
00:23:09,240 --> 00:23:12,800
Speaker 1: mcili is our engineer. Our original music was composed by

424
00:23:12,880 --> 00:23:16,240
Speaker 1: Leo Sidrin. I'm west Kosova. We'll be back tomorrow with

425
00:23:16,359 --> 00:23:17,320
Speaker 1: another Big Take.