1 00:00:17,879 --> 00:00:21,320 Speaker 1: And we are back with another edition of the Federalist 2 00:00:21,440 --> 00:00:25,280 Radio Hour. I'm Matt Kittle, senior Elections correspondent at the 3 00:00:25,280 --> 00:00:30,160 Federalist and your experience Shirpa on today's quest for Knowledge. 4 00:00:30,559 --> 00:00:33,600 As always, you can email the show at radio at 5 00:00:33,640 --> 00:00:38,039 the Federalist dot com, follow us on x at FDR LST, 6 00:00:38,840 --> 00:00:42,880 make sure to subscribe wherever you download your podcast, and 7 00:00:43,000 --> 00:00:46,399 of course to the premium version of our website as well. 8 00:00:46,960 --> 00:00:50,520 Our guest today is Taylor Barkley, Director of public Policy 9 00:00:50,679 --> 00:00:55,280 for the Abundance Institute. Taylor joins us to talk about 10 00:00:55,479 --> 00:01:02,840 artificial intelligence in US elections, particular the regulation of AIS. 11 00:01:02,960 --> 00:01:09,319 Some truly remarkable times these with some very concerning legislation 12 00:01:09,519 --> 00:01:13,120 for defenders of the First Amendment and free speech. Taylor, 13 00:01:13,159 --> 00:01:15,560 thank you so much for joining us today on this 14 00:01:15,719 --> 00:01:18,319 episode of the Federalist Radio Hour. 15 00:01:19,280 --> 00:01:20,799 Speaker 2: It's great to be here. Matt, thanks so much for 16 00:01:20,840 --> 00:01:21,239 having me on. 17 00:01:21,840 --> 00:01:25,400 Speaker 1: Well, it looks like Congress won't be passing any legislation 18 00:01:25,920 --> 00:01:29,280 on this front before election day. I mean by that 19 00:01:29,680 --> 00:01:35,840 artificial intelligence regulation as it relates to elections. You note 20 00:01:36,560 --> 00:01:41,480 that Congressional inaction is quite okay and a relief because 21 00:01:41,519 --> 00:01:45,200 many of these bills pose significant free speech issues. 22 00:01:45,239 --> 00:01:45,920 Speaker 2: I want to get. 23 00:01:45,760 --> 00:01:50,519 Speaker 1: Into the bills before Congress now, but I think the 24 00:01:50,879 --> 00:01:55,200 vote pressing issue at this point is what happened in California. 25 00:01:56,519 --> 00:02:01,400 My colleague Tristan Justice writes at the Federal List. California's 26 00:02:01,439 --> 00:02:05,480 far left governor celebrated Constitution Day with a series of 27 00:02:05,519 --> 00:02:10,400 new laws to quote cracked down on free speech articulated 28 00:02:10,520 --> 00:02:17,280 via artificially generated content. On Tuesday, the California Democrat governor 29 00:02:17,479 --> 00:02:22,120 officially outlawed the creation and distribution of images or videos 30 00:02:22,120 --> 00:02:28,639 created with artificial intelligence known as deep fakes. Let's begin there. 31 00:02:28,800 --> 00:02:33,680 What does this new law do and how will this 32 00:02:33,960 --> 00:02:37,000 impact the election and ultimately free speech? 33 00:02:38,400 --> 00:02:42,000 Speaker 2: These are all great questions, and I think my top 34 00:02:42,039 --> 00:02:46,520 line here takeaway is that AI and election policies are 35 00:02:46,680 --> 00:02:51,039 problems in search of problems. As you mentioned, and I'm 36 00:02:51,080 --> 00:02:54,599 glad you kine Of wrote that article, every level of 37 00:02:54,599 --> 00:02:59,520 government is looking at this issue. There's an SCC proceeding 38 00:02:59,599 --> 00:03:03,360 right now regulating they would seek to regulate AI and 39 00:03:03,400 --> 00:03:07,560 political advertisements. The deadline to submit comments, actually unpreceding is today. 40 00:03:07,680 --> 00:03:10,000 If any member of the public can submit a comment 41 00:03:10,000 --> 00:03:14,159 on these proposed regulations. Congress is proceeding and is unlikely 42 00:03:14,240 --> 00:03:17,560 to pass anything. And then in the states, there's been 43 00:03:17,800 --> 00:03:21,439 there have been numerous laws passed, and I think I 44 00:03:21,479 --> 00:03:24,360 counted twenty different states have passed different laws, as you mentioned, 45 00:03:24,400 --> 00:03:27,080 yet California was the most recent. I'll also add on 46 00:03:27,759 --> 00:03:30,080 you know, one of the reasons Governor Newsom was so 47 00:03:30,159 --> 00:03:33,360 eager to sign these bills into law was because he 48 00:03:33,560 --> 00:03:37,639 noticed a parody Kamala Harris video in the summer, and 49 00:03:38,199 --> 00:03:40,879 it was famous for you know, quote tweeting it and saying, 50 00:03:40,919 --> 00:03:43,280 I'm eager to sign legislation that would prevent this kind 51 00:03:43,280 --> 00:03:46,159 of thing. Turns out he did, and then those bills. 52 00:03:46,199 --> 00:03:48,560 He signed those bills into law. I think it was yesterday. 53 00:03:48,759 --> 00:03:52,000 The creator of that parody of video sued the government 54 00:03:52,080 --> 00:03:56,360 of California for First and fourteenth Amendment violations because, as 55 00:03:56,400 --> 00:04:00,479 you mentioned, there are just tremendous free speech implications here 56 00:04:01,080 --> 00:04:03,960 for a number of reasons. I mean, you know, first 57 00:04:03,960 --> 00:04:08,120 of all, you know, it's the constitution permits uh, you know, 58 00:04:08,599 --> 00:04:12,680 say fake fake speech, maybe not outright fraud, but in 59 00:04:12,719 --> 00:04:15,400 some cases just has to be narrowly tailored. But you know, 60 00:04:15,439 --> 00:04:19,680 we can say, uh, you know, fanciful things about even 61 00:04:20,240 --> 00:04:24,959 real people in history. Uh, there's the Constitution protects such speech. 62 00:04:25,000 --> 00:04:28,040 And a lot of these laws, including the three Bills 63 00:04:28,040 --> 00:04:30,319 of the Governor's signed in California, are are written in 64 00:04:30,360 --> 00:04:33,240 such a way that could just open the floodgates for 65 00:04:33,360 --> 00:04:38,319 privolous litigation on say, you know there's a upstart candidate 66 00:04:38,360 --> 00:04:43,000 who you know, uses AI generated content to do just 67 00:04:43,160 --> 00:04:47,279 very like normal things, say make campaign posters, or even 68 00:04:47,319 --> 00:04:51,560 doesn't use AI to generate images or content. And the 69 00:04:51,959 --> 00:04:56,519 incumb Uh, the incumbent candidate could well well funded, connected 70 00:04:56,560 --> 00:05:00,120 to the right people, could sue this this challenger or 71 00:05:00,279 --> 00:05:02,399 just to tie them up, you have them spend more 72 00:05:02,399 --> 00:05:05,519 money than they should and you know, either you know, 73 00:05:05,639 --> 00:05:10,360 tank their likelihood of getting elected because they're busy chasing 74 00:05:10,399 --> 00:05:14,800 down these frivolous lawsuits or not speaking as much as 75 00:05:14,800 --> 00:05:18,439 they would otherwise because out of fear of maybe getting 76 00:05:18,879 --> 00:05:24,040 these kind of potentially game ending lawsuits filed against them. 77 00:05:24,079 --> 00:05:26,560 Speaker 1: The chilling of speech. You know, that is a big 78 00:05:26,639 --> 00:05:30,279 part list as well. Again, as my colleague get the 79 00:05:30,560 --> 00:05:34,600 Federalist Tristan Justice wrote newsome as you noted, sign two 80 00:05:34,600 --> 00:05:38,639 other pieces of legislation requiring campaigns and social media platforms 81 00:05:38,639 --> 00:05:44,120 to disclose whether their content was created with artificial intelligence. 82 00:05:44,759 --> 00:05:46,839 And this is what you were getting into before. And 83 00:05:46,879 --> 00:05:51,240 I think this is hilarious. For everybody will have a 84 00:05:51,240 --> 00:05:55,360 different view about the content, but I thought that. I 85 00:05:55,399 --> 00:06:02,319 thought the AI ad with Kamala Harris is absolutely hilarious 86 00:06:02,360 --> 00:06:04,800 and quite frankly, I knew it was a joke. 87 00:06:06,000 --> 00:06:06,319 Speaker 2: Yeah, but. 88 00:06:07,920 --> 00:06:11,560 Speaker 1: Most most people get sarcasm, they get parody, you know, 89 00:06:11,639 --> 00:06:16,600 they get you know, they know what weird Al Yankovic 90 00:06:16,680 --> 00:06:20,199 has done the popular music over the years. For the 91 00:06:20,240 --> 00:06:24,079 look I got, most people get satire, and that's exactly 92 00:06:24,199 --> 00:06:27,360 what it was. But you know, for us of a 93 00:06:27,399 --> 00:06:33,519 certain philosophical, political, philosophical point of view, that stuff seemed 94 00:06:33,519 --> 00:06:38,439 to ring true in any ways about Kamala Harris and 95 00:06:38,480 --> 00:06:40,680 where she is at with her policies. But I got 96 00:06:40,839 --> 00:06:45,560 first and foremost the fact that it was a satire 97 00:06:45,680 --> 00:06:51,279 and b that it was you know, protected speech but 98 00:06:51,319 --> 00:06:55,879 no longer in California. So elon Musk comes out. I 99 00:06:56,040 --> 00:06:58,879 remember that a lot of this stuff, isn't it driven 100 00:06:59,040 --> 00:07:05,959 by Musk? Actually, you know, pushing this out there, this 101 00:07:05,959 --> 00:07:11,480 this artificial intelligence ad out there and making all of 102 00:07:11,480 --> 00:07:14,680 these threats. Now he's come through on those well. In July, 103 00:07:15,000 --> 00:07:18,879 Musk announced that he would relocate his headquarters for two 104 00:07:18,920 --> 00:07:23,240 of his companies from California to Texas because you know, 105 00:07:23,680 --> 00:07:29,879 he was being basically with these bills targeted for free speech. 106 00:07:29,920 --> 00:07:32,720 He said, it's hard to be a free speech platform 107 00:07:32,759 --> 00:07:36,639 in a state that wants to ban free speech. Isn't 108 00:07:36,639 --> 00:07:39,839 that where we're at right now? With these new laws 109 00:07:39,879 --> 00:07:40,360 in place? 110 00:07:41,560 --> 00:07:44,079 Speaker 2: That seems to be the case. And you know when 111 00:07:44,120 --> 00:07:46,279 I said this is, you know, these policies and these 112 00:07:46,360 --> 00:07:50,319 these laws are problems in search of problems. I'm going 113 00:07:50,399 --> 00:07:53,360 to get the second problem, and this issue of AI 114 00:07:53,519 --> 00:07:57,720 and elections is not really the prognostications and dire predictions 115 00:07:57,720 --> 00:08:00,240 have not come to be. They have not they have 116 00:08:00,319 --> 00:08:03,240 not proven true. And at the Abundance Institute, my team 117 00:08:03,240 --> 00:08:07,399 and I we've been tracking every single media mentioned using 118 00:08:07,519 --> 00:08:11,279 AI machine learning tools, every single media mention of say 119 00:08:11,480 --> 00:08:15,839 US election, AI, Biden, Trump, Harris, you know a number 120 00:08:15,839 --> 00:08:19,160 of keywords and your listeners can actually access the full 121 00:08:19,240 --> 00:08:24,279 database at AI Election Observatory dot com. And we have 122 00:08:24,399 --> 00:08:30,279 some of our writeups and summarizing our findings in multi 123 00:08:30,279 --> 00:08:33,080 part series that we've been doing over the course of 124 00:08:33,080 --> 00:08:36,720 the year at our substack now and next dot substack. 125 00:08:36,960 --> 00:08:39,279 Speaker 1: Yeah, it's a great cury, it really is. This is 126 00:08:40,279 --> 00:08:43,879 the third in the fourth it's a four part series, 127 00:08:43,919 --> 00:08:46,440 I believe. I think this is the third installment, correct. 128 00:08:46,600 --> 00:08:49,120 Speaker 2: That's correct. Yeah, And I will have the fourth I 129 00:08:49,159 --> 00:08:52,240 believe around October sixth which will be around thirty that 130 00:08:52,240 --> 00:08:55,120 will be thirty days until the November fifth election, and 131 00:08:55,240 --> 00:08:58,679 we'll have an after election kind of analysis. And we've 132 00:08:58,720 --> 00:09:02,120 been writing these as just you know, whatever happens, we 133 00:09:02,159 --> 00:09:05,720 want to log it because since I've been writing about 134 00:09:05,720 --> 00:09:07,879 this issue of you know, deep fakes and you know 135 00:09:08,000 --> 00:09:10,559 truths and elections a bit always come into it since 136 00:09:10,600 --> 00:09:13,279 about twenty nineteen, and you know every election since then 137 00:09:13,320 --> 00:09:15,399 has been you know, the last kind of human election 138 00:09:15,679 --> 00:09:20,159 twenty twenty two, twenty twenty four. And you know, we 139 00:09:20,200 --> 00:09:23,080 did these this tracker just to have the receipts. So 140 00:09:23,120 --> 00:09:26,240 what has actually happened. And you know, we're in our 141 00:09:26,320 --> 00:09:30,360 comment which we're filing in this FCC proceeding, we we 142 00:09:30,639 --> 00:09:33,799 we say we have thirty five thousand media mentions about 143 00:09:33,960 --> 00:09:39,240 AI and US election instances, but only four actual instances 144 00:09:39,480 --> 00:09:44,039 of political campaigns using AI generated content. And you know, 145 00:09:44,080 --> 00:09:48,679 even the most famous, arguably the most effective perhaps example 146 00:09:48,720 --> 00:09:52,360 of AI generated content the Biden robocall in New Hampshire 147 00:09:52,399 --> 00:09:58,399 earlier this year, where he robot Biden apparently call called 148 00:09:58,399 --> 00:10:01,399 a number of a few thousand residents in New Hampshire 149 00:10:01,399 --> 00:10:03,240 telling them to not turn up at the polls during 150 00:10:03,240 --> 00:10:07,240 the primary, and according to testimony from the New Hampshire 151 00:10:07,240 --> 00:10:11,480 Elections commissioner, it was the highest turnout ever. And this 152 00:10:11,679 --> 00:10:15,840 robocall ended up being driven by a political activist who 153 00:10:16,960 --> 00:10:19,039 lived out of state and wanted to make a point 154 00:10:19,360 --> 00:10:21,519 and is now being fined six million dollars by the 155 00:10:21,559 --> 00:10:24,960 Federal Communications Commission. But even that, right, it was you know, 156 00:10:25,000 --> 00:10:29,039 supposedly of the more effective means and method didn't impact. 157 00:10:29,080 --> 00:10:31,840 There was no discernible effect. Voters still turned out to vote. 158 00:10:32,240 --> 00:10:34,279 And we see that time and time again, and I 159 00:10:34,279 --> 00:10:36,919 think there's this overestimation and how easy it is to 160 00:10:37,039 --> 00:10:41,519 deceive people. In many cases, the bottleneck is often on 161 00:10:41,600 --> 00:10:43,919 distribution of this material. And you know, you mentioned Elon 162 00:10:44,039 --> 00:10:46,440 Musk sharing the Kamala Harris parity video, and I think 163 00:10:46,440 --> 00:10:49,519 that's an interesting example where you know, maybe if Musk 164 00:10:49,600 --> 00:10:53,639 given his platform the attention he receives from politicians like 165 00:10:53,679 --> 00:10:56,960 Gavin Newsom. Maybe this Kamala Harris parity video wouldn't have 166 00:10:58,399 --> 00:11:00,360 gotten the attention it did, but you know, you you 167 00:11:00,440 --> 00:11:03,399 shared it, You've got a lot of attention. That's where 168 00:11:03,440 --> 00:11:06,080 we see this. I think the other problem with these 169 00:11:06,120 --> 00:11:08,399 policies come in we have the lawmakers all over the 170 00:11:08,480 --> 00:11:10,840 nation at all levels thing, you know, we have to 171 00:11:10,879 --> 00:11:13,679 do something, and that's that's where so many of these 172 00:11:13,679 --> 00:11:16,720 concerns about free speech come in. You know, a lot 173 00:11:16,720 --> 00:11:20,799 of these like the bills in California requiring platforms to 174 00:11:21,039 --> 00:11:24,240 label content. So now you have to now the owner 175 00:11:24,399 --> 00:11:28,399 onus is on a platform to look through every single picture, 176 00:11:28,879 --> 00:11:32,240 audio file, video to make a determination within you know, 177 00:11:32,279 --> 00:11:34,960 a reasonable amount of time whether it's AI generated or 178 00:11:35,039 --> 00:11:37,759 was it not, and then label it. And of course 179 00:11:37,799 --> 00:11:43,919 this is just California and not a nationwide law, so 180 00:11:44,200 --> 00:11:48,480 there's incredible compliance burdens there. There's also just potential to 181 00:11:48,600 --> 00:11:52,240 muddy the waters, like if everything, if things are over labeled, 182 00:11:52,360 --> 00:11:56,639 then maybe that can create its own kind of falsehoods 183 00:11:56,679 --> 00:12:01,639 and issues, and you know, we could see you know, 184 00:12:01,720 --> 00:12:03,840 speech slowed down, It could take longer for things to 185 00:12:03,879 --> 00:12:08,120 get posted, frivolous lawsuits will fly because you know who 186 00:12:08,120 --> 00:12:11,639 doesn't like suing the big tech companies for whatever reason 187 00:12:11,720 --> 00:12:16,279 on whatever side. And we just go on down the list, Matt, 188 00:12:16,279 --> 00:12:19,200 of all the all the troubling issues with these bills. 189 00:12:19,240 --> 00:12:21,879 And but you know, I really want to hammer home 190 00:12:21,919 --> 00:12:24,919 for your listeners. You know the dominant narrative out there 191 00:12:25,039 --> 00:12:29,240 that AI generated generative AI is going to ruin democracy, 192 00:12:29,559 --> 00:12:32,600 the death of democracy. That story is not proving true. 193 00:12:33,000 --> 00:12:35,879 And it's you know, traditional methods are still out there 194 00:12:35,919 --> 00:12:39,879 of missing disinformation, and politicians are way overreacting and in 195 00:12:40,200 --> 00:12:42,559 most ninety nine percent of the cases that I've. 196 00:12:42,440 --> 00:12:47,679 Speaker 3: Seen, we have a huge hole in our nation's roof. 197 00:12:47,840 --> 00:12:50,600 The Watched Out on Wall Street podcast with Chris Markowski 198 00:12:50,679 --> 00:12:53,519 every day Chris helps unpack the connection between politics and 199 00:12:53,559 --> 00:12:56,039 the economy and how it affects your wallet. There should 200 00:12:56,120 --> 00:12:59,679 be only one priority for our lawmakers. America is spending 201 00:12:59,799 --> 00:13:02,679 three eight billion dollars a day in paying off our 202 00:13:02,720 --> 00:13:05,360 own interest. You only need to be voting for candidates 203 00:13:05,360 --> 00:13:07,960 who can address this problem. Whether it's happening in DC 204 00:13:08,159 --> 00:13:10,759 or down on Wall Street, it's affecting you financially. Be informed, 205 00:13:10,840 --> 00:13:12,879 check out the Watchdowt on Wall Street podcast with Chris 206 00:13:13,000 --> 00:13:15,919 Markowski on Apple, Spotify or wherever you get your podcasts. 207 00:13:18,840 --> 00:13:21,200 Speaker 1: That's what they do in ninety nine percent of the cases, 208 00:13:21,240 --> 00:13:25,200 no matter what. So we can we can apply that 209 00:13:25,279 --> 00:13:28,080 to just about anything. And isn't that that is the case? 210 00:13:28,200 --> 00:13:31,600 I mean, the fire of AI over the last couple 211 00:13:31,600 --> 00:13:35,840 of years in this country, particularly over the last six months, 212 00:13:35,960 --> 00:13:38,960 has been, oh my gosh, we've got to do something, 213 00:13:39,000 --> 00:13:41,639 We've got to do something. That's exactly what Gavin Newsom 214 00:13:41,720 --> 00:13:45,200 is doing. I do believe. Of course, there are you know, 215 00:13:45,759 --> 00:13:51,120 there are leftist agenda items within his conduct and his 216 00:13:51,320 --> 00:13:54,399 movements on this front as well. But you know, that's 217 00:13:54,559 --> 00:13:58,720 basically the case across the country. We have politicians, we 218 00:13:58,799 --> 00:14:03,320 have policymakers, we have bureaucrats trying to do something, once again, 219 00:14:03,399 --> 00:14:08,200 to regulate something, and that's a very dangerous place to be. Now. 220 00:14:08,360 --> 00:14:11,320 Having said all of that, I will tell you this, 221 00:14:13,279 --> 00:14:18,120 I'm not real thrilled about the technology. But then again, 222 00:14:18,519 --> 00:14:23,720 I'm an old journalist. You know, I didn't start off 223 00:14:23,759 --> 00:14:28,120 with a typewriter, certainly, but you know, I started out 224 00:14:28,200 --> 00:14:31,960 when you couldn't go to the internet al Gore hadn't 225 00:14:31,960 --> 00:14:34,879 invented it yet, and so you couldn't go to the 226 00:14:34,879 --> 00:14:38,279 Internet to just basically get the information. It was a 227 00:14:38,320 --> 00:14:41,879 skill to be a reporter. Now some of that is deteriorating, 228 00:14:42,559 --> 00:14:46,120 and it's the impact it will have elsewhere, I think 229 00:14:46,879 --> 00:14:50,919 is clearly it will be substantial. But you know what 230 00:14:50,960 --> 00:14:55,080 I do like is I like the First Amendment, Love 231 00:14:55,240 --> 00:15:00,440 the First Amendment, Love free speech. And I just see 232 00:15:00,519 --> 00:15:04,320 this whole movement going on, the putting the fire out 233 00:15:04,360 --> 00:15:09,399 of AI as yet another assault, the latest in a 234 00:15:09,559 --> 00:15:14,240 multifront assault on the First Amendment, and it is coming 235 00:15:14,320 --> 00:15:17,399 from the people who are supposed to be the defenders 236 00:15:17,679 --> 00:15:21,519 of the First Amendment. In the Federalist story that I 237 00:15:21,639 --> 00:15:27,120 had just mentioned, the AP has actually applauded the California 238 00:15:27,360 --> 00:15:32,279 blueprint for killing free speech. As noted in the piece, 239 00:15:32,360 --> 00:15:36,799 the new laws reaffirm California's position as a leader in 240 00:15:36,919 --> 00:15:42,840 regulating AI in the US, especially in combating election deep fakes. 241 00:15:42,919 --> 00:15:47,559 The AP reported the state was the first in the 242 00:15:47,679 --> 00:15:52,840 US to ban manipulated videos and pictures related to elections 243 00:15:52,919 --> 00:15:58,080 in twenty nineteen. Measures in technology and AI proposed by 244 00:15:58,120 --> 00:16:02,200 California lawmakers have been used as as blueprints for legislators 245 00:16:02,200 --> 00:16:07,960 across the country. Industry experts said, and again, this is 246 00:16:08,000 --> 00:16:10,879 the AP reporting what is happening out there. But it 247 00:16:10,960 --> 00:16:13,600 seems to me that there are a lot of folks 248 00:16:13,720 --> 00:16:18,799 in journalism, particularly corporate journalism, that are leading this story 249 00:16:18,840 --> 00:16:22,240 as well. Are you tracking that? 250 00:16:22,240 --> 00:16:26,519 Speaker 2: That does make sense? You know, it's a typical pattern 251 00:16:26,720 --> 00:16:31,200 with emerging technologies. You know, emerging technologies often have the 252 00:16:31,279 --> 00:16:37,080 effect of challenging incumbent power or entrenched interests, whatever they 253 00:16:37,120 --> 00:16:40,440 might be. You know, the printing press did this, Digital 254 00:16:40,480 --> 00:16:44,559 computing has done this. You know, you think about your 255 00:16:44,559 --> 00:16:48,360 automobiles up ending the buggy making, the buggy whip industry, 256 00:16:48,679 --> 00:16:50,240 on on and on we could go, and you know, 257 00:16:50,279 --> 00:16:55,759 there's this push and pull between industries and individuals who 258 00:16:55,879 --> 00:16:58,159 have done things maybe one way, have certainly made their 259 00:16:58,159 --> 00:17:02,759 livelihoods doing things another way, and then the maybe individuals 260 00:17:02,799 --> 00:17:06,720 and companies that have new technology. So using the generative 261 00:17:06,759 --> 00:17:10,480 AI as an example, I think it makes absolute sense. 262 00:17:10,559 --> 00:17:13,640 You know that the AP would be for this because 263 00:17:14,240 --> 00:17:17,039 you know what generative AI does is I believe and 264 00:17:17,559 --> 00:17:19,160 I've seen it happen, you know, in my own life 265 00:17:19,200 --> 00:17:24,880 and with others. It's democratized creative writing. It's democratized graphic 266 00:17:24,960 --> 00:17:29,240 design in so many ways. I used Generative Google's new 267 00:17:29,960 --> 00:17:33,680 notebook LM today to summarize a long paper I didn't 268 00:17:33,720 --> 00:17:35,680 have time to read, and it actually converted it into 269 00:17:36,039 --> 00:17:40,920 a podcast style interview between between two AI voices, and 270 00:17:40,960 --> 00:17:44,079 it was a really helpful way to learn quickly about 271 00:17:44,119 --> 00:17:45,759 this this paper that I didn't have time to sit 272 00:17:45,799 --> 00:17:49,000 down and read. And we're seeing that that seems creepy 273 00:17:49,079 --> 00:17:54,000 to me, but trying it out, yeah, it was pretty fun. 274 00:17:55,000 --> 00:17:58,400 And yeah, but you know, there's room for everyone here, Matt, Like, 275 00:17:58,440 --> 00:18:00,799 if you don't want to use it, that's fine, No, no, no. 276 00:18:00,720 --> 00:18:03,680 Speaker 1: And don't And please don't get me wrong, just because 277 00:18:03,960 --> 00:18:09,000 I'm creeped out by some of it doesn't mean as 278 00:18:09,039 --> 00:18:13,200 a first of all, a free market capitalist, I'm I'm 279 00:18:13,279 --> 00:18:17,759 here to stop the next wave of technology, the next 280 00:18:17,799 --> 00:18:21,200 wave of innovation, and I certainly am not here to 281 00:18:21,319 --> 00:18:25,599 stop the expression of of thought in this country. I 282 00:18:25,599 --> 00:18:28,359 think we're in a dangerous place. Just because you know, 283 00:18:28,480 --> 00:18:32,640 something creeps me out doesn't mean that I'm I'm here 284 00:18:32,799 --> 00:18:37,880 to you know, push the administrative state to stop all 285 00:18:37,920 --> 00:18:41,279 of this and to force you know, burden some regulations. 286 00:18:41,920 --> 00:18:44,839 You know, I am I am burdened by what has 287 00:18:44,960 --> 00:18:51,319 been to borrow a phrase from the Democrat presidential candidate, 288 00:18:51,920 --> 00:18:54,599 I am burdened, and we all are burdened by what 289 00:18:54,759 --> 00:18:58,960 has been in terms of regulation in this country. But 290 00:18:59,119 --> 00:19:02,640 you know that that's what I'm seeing through through all 291 00:19:02,720 --> 00:19:06,440 of this is this massive panic, this massive scare, and 292 00:19:06,480 --> 00:19:09,880 what you are ultimately saying is even if you don't 293 00:19:09,920 --> 00:19:16,039 care for the technology, the evidence doesn't support the knee 294 00:19:16,119 --> 00:19:17,440 jerk reaction to it. 295 00:19:18,039 --> 00:19:19,680 Speaker 2: That's correct, That's correct, because I think a lot of 296 00:19:19,720 --> 00:19:22,559 this is motivated by these hypothetical scenarios. And that's like 297 00:19:23,480 --> 00:19:27,079 we're even reading this FCC proposed rule, and they're their motivation. 298 00:19:27,240 --> 00:19:31,359 It's it's hypothetical harms of and the Governor of California 299 00:19:31,400 --> 00:19:34,039 even in his statement it's and the members of the 300 00:19:34,039 --> 00:19:38,119 Assembly who supported these these bills. It's well, someday someone 301 00:19:38,160 --> 00:19:41,640 will make a video of X Y politicians saying, you know, 302 00:19:41,799 --> 00:19:44,000 such and such a thing, and that's going to be 303 00:19:44,079 --> 00:19:46,480 the end of the world, or you know, cause some 304 00:19:46,559 --> 00:19:50,599 sort of harm. And we had plenty of opportunities over 305 00:19:50,640 --> 00:19:53,880 the last decade. I'll say, even just since twenty seventeen 306 00:19:53,960 --> 00:19:58,720 for manipulated you know, AI generated content that's convincing enough 307 00:19:58,720 --> 00:20:01,200 to cause harm. I think, you know, a key factor 308 00:20:01,240 --> 00:20:05,160 of you know, even miss It will say, you know, disinformation. 309 00:20:05,359 --> 00:20:08,000 This is someone you know who's creating information to to 310 00:20:08,240 --> 00:20:13,720 sow you know, confusion or or cause harm. You know, 311 00:20:13,880 --> 00:20:17,200 Donald Trump was perhaps maybe the best person because he 312 00:20:17,279 --> 00:20:19,160 was so unpredictable. You know, we never know what he's 313 00:20:19,200 --> 00:20:20,920 going to say next, and I think that is a 314 00:20:21,000 --> 00:20:26,480 critical ingredient for disinformation. But no one, there was no instance. 315 00:20:26,519 --> 00:20:29,200 I mean, people did create fake content with Donald Trump, 316 00:20:29,279 --> 00:20:32,240 but nothing rose to the level of these these hypothetical 317 00:20:32,319 --> 00:20:35,960 doom and gloom scenarios where say, you know, we went 318 00:20:36,000 --> 00:20:38,279 to war with another country, or you know, half the 319 00:20:38,319 --> 00:20:40,720 population decided not to go vote because they heard the 320 00:20:40,720 --> 00:20:43,359 president say that the polls were closed or something like that. 321 00:20:43,359 --> 00:20:46,640 That has not happened. It's really really hard to do 322 00:20:47,000 --> 00:20:51,000 at that scale. Even though you know, yes, generative AI 323 00:20:51,079 --> 00:20:55,279 has democratized kind of you know, top tier level graphic design, 324 00:20:55,359 --> 00:20:59,559 video editing content, and it will only get better from here, Matt, 325 00:20:59,599 --> 00:21:02,960 I think that that always goes into this prognostication it 326 00:21:03,000 --> 00:21:06,799 will only get better from here, even with you know, 327 00:21:07,279 --> 00:21:10,799 nation states who are well funded, well motivated, have the outreach. 328 00:21:11,519 --> 00:21:14,160 There's nothing been, nothing has risen to this, this doom 329 00:21:14,200 --> 00:21:18,640 and gloom level, and so it's discouraging to see policy 330 00:21:18,640 --> 00:21:22,200 makers reacting and crafting policy which should never be done. 331 00:21:22,200 --> 00:21:25,480 It should never be done from hypotheticals In most cases, 332 00:21:26,839 --> 00:21:28,720 I'll say most cases, it should never be just based 333 00:21:28,720 --> 00:21:31,720 purely and hypotheticals. Policy making should always be based in 334 00:21:32,480 --> 00:21:34,880 the evidence. What has happened, what is the actual harm 335 00:21:34,880 --> 00:21:36,920 a is trying to be prevented here, and what is 336 00:21:36,960 --> 00:21:39,359 the best way to actually prevent that harm? In many 337 00:21:39,480 --> 00:21:42,839 in most instances public policy, it's one tool in the toolbox. 338 00:21:42,880 --> 00:21:45,599 I think that's one of my bugaboos. We have this 339 00:21:45,680 --> 00:21:48,000 kneejerk reaction to say, oh, there's a problem in the world, 340 00:21:48,000 --> 00:21:50,240 will we got to pass a law when you know, 341 00:21:50,599 --> 00:21:54,319 I wrote this article for this new organization, the Center 342 00:21:55,200 --> 00:22:00,240 for News, Technology and Innovation, based on this research is 343 00:22:00,279 --> 00:22:02,599 tracking we've been doing or this article. You know, the 344 00:22:02,599 --> 00:22:08,119 instances we've seen institutions, civil society journalists included have done 345 00:22:08,160 --> 00:22:10,839 a good job about filtering out, you know, back from 346 00:22:10,880 --> 00:22:15,359 fiction you will take. I think it was Ron DeSantis 347 00:22:15,680 --> 00:22:19,599 campaign affiliated affiliate created what appeared to be some generative 348 00:22:19,640 --> 00:22:24,240 AI images of I think Trump hung hugging Fauci, and 349 00:22:24,720 --> 00:22:29,519 you know, journalists, uh, news sites sprung into action, fact 350 00:22:29,559 --> 00:22:32,759 checking orgs sprung into action, citizens sprung into action, community 351 00:22:32,799 --> 00:22:35,920 notes on X sprung into action, and it was you know, 352 00:22:35,960 --> 00:22:38,359 everyone got involved and said, well, here's what actually happened, 353 00:22:38,359 --> 00:22:41,720 and they provided the information to the public. Hey, just 354 00:22:41,759 --> 00:22:45,119 an fyi, some of these appear AI generated, and that's 355 00:22:45,160 --> 00:22:47,680 exactly what should be happening here. It shouldn't be bands, 356 00:22:47,720 --> 00:22:51,519 it shouldn't be requirements and labeling. And you know, maybe, 357 00:22:51,559 --> 00:22:52,599 you know, if you want to take it that we 358 00:22:52,640 --> 00:22:54,799 can talk about just the issues of defining AI and 359 00:22:54,920 --> 00:22:57,200 all the problems that creates. But you know, I think 360 00:22:57,279 --> 00:23:02,680 that the machine is working, by my estimate, and we 361 00:23:02,799 --> 00:23:05,079 need to trust people. We can trust people on so 362 00:23:05,200 --> 00:23:07,599 much of this legislation and these regulations seems to be 363 00:23:07,640 --> 00:23:09,440 motivated by this lack of trust. You know, people will 364 00:23:09,519 --> 00:23:12,440 just be bamboozled if they see something fake, when that's 365 00:23:12,519 --> 00:23:13,319 not been the case. 366 00:23:14,240 --> 00:23:18,079 Speaker 1: Our guest today is Taylor Barkley, director of public Policy 367 00:23:18,279 --> 00:23:22,839 for the Abundance Institute. Taylor joining us obviously to talk 368 00:23:22,880 --> 00:23:28,839 about artificial intelligence and the impact or perhaps lack thereof 369 00:23:29,119 --> 00:23:34,799 on election season, this critical election season. Obviously, as we 370 00:23:34,799 --> 00:23:39,480 get closer and closer to election day, What about some 371 00:23:39,559 --> 00:23:45,880 of the legislation before Congress? Again, the vast majority of it, 372 00:23:45,960 --> 00:23:51,279 as I see, is really a reaction to the fears 373 00:23:51,359 --> 00:23:58,240 that are driving around AI or being driven by people 374 00:23:58,480 --> 00:24:02,720 afraid of AI. What sort of legislation are we seeing 375 00:24:02,799 --> 00:24:03,920 in Congress today? 376 00:24:05,640 --> 00:24:07,440 Speaker 2: So, I think there are a number of bills. I 377 00:24:07,440 --> 00:24:11,799 think the leading contenders are on the Senate side, mostly 378 00:24:12,440 --> 00:24:17,000 sponsored by Senator Amy Klobuchar, and have gone through the 379 00:24:17,039 --> 00:24:20,599 Senate Rules and Administration Committee, And my colleague Neil Chilson 380 00:24:20,640 --> 00:24:24,440 actually testified and one of the hearings talking about I 381 00:24:24,480 --> 00:24:29,359 think a package of four different Senate bills to prevent 382 00:24:29,440 --> 00:24:32,319 the use of AI generated content or required disclosures of 383 00:24:32,359 --> 00:24:36,160 AI generated content and elections, And you know, his testimony 384 00:24:36,200 --> 00:24:38,759 and the others on the panel just just talked about 385 00:24:39,519 --> 00:24:41,240 you know. So there's a lot a lot of these bills. 386 00:24:41,279 --> 00:24:43,599 They I'll say, the federal bills and the state bills. 387 00:24:43,720 --> 00:24:46,599 They start with definitions. That's if you look at legislation, 388 00:24:46,720 --> 00:24:51,000 you'll notice that and that is one of the main 389 00:24:52,000 --> 00:24:55,240 problems with this legislation, whether the federal level of the 390 00:24:55,240 --> 00:24:56,359 state level, and all of it. You know, we can 391 00:24:56,359 --> 00:24:58,599 say this about the Senate bills because that's what my 392 00:24:58,680 --> 00:25:01,519 colleague talked about, and I totally with him. They define 393 00:25:01,640 --> 00:25:04,519 artificial intelligence in such a way that it wraps up 394 00:25:05,079 --> 00:25:10,799 all basically all computer software. So say one of these 395 00:25:10,839 --> 00:25:13,599 Senate bills were passed as written and there wasn't a 396 00:25:14,200 --> 00:25:17,359 narrowly defined definition of AI. And I'll just say here 397 00:25:17,400 --> 00:25:21,279 for your listeners, defining AI is incredibly hard, especially in 398 00:25:21,359 --> 00:25:24,720 a legal context. And as my colleague Neil noted in 399 00:25:24,759 --> 00:25:29,079 his testimony before Center Rules Committee, the leading textbook artificial 400 00:25:29,119 --> 00:25:31,559 Intelligence has a whole chapter devoted about the four ways 401 00:25:31,599 --> 00:25:34,279 to define AI, and even that doesn't come to a conclusion. 402 00:25:34,359 --> 00:25:36,799 So policymakers have a big job when they're trying to 403 00:25:36,839 --> 00:25:39,799 do this. But anyway, you know, broad definitions that wrap 404 00:25:39,880 --> 00:25:43,319 up all software. Again, back to the chilling free speech 405 00:25:44,160 --> 00:25:47,240 we could see now. You know, most content is labeled 406 00:25:47,240 --> 00:25:49,680 as AI generated when it's really not. Because say I 407 00:25:49,839 --> 00:25:52,839 take a picture with my iPhone and make some tweaks, 408 00:25:53,119 --> 00:25:56,000 you know, may say I'm a staffer for you, Matt, 409 00:25:56,039 --> 00:25:57,799 and you're running for office, and I take a picture 410 00:25:57,839 --> 00:26:01,599 of you giving a speech with my iPhone sixteen pro 411 00:26:01,960 --> 00:26:04,400 and I just, you know, your face is a little dark, 412 00:26:04,440 --> 00:26:07,559 so I'm just gonna hit the button brighten it. That 413 00:26:07,559 --> 00:26:10,440 could count as AI and if that goes out into 414 00:26:10,480 --> 00:26:12,240 the world, and that could have to be labeled even 415 00:26:12,279 --> 00:26:14,200 though it's just a photo. So we can get these 416 00:26:14,279 --> 00:26:18,200 real tricky areas of like where is that line between 417 00:26:18,240 --> 00:26:20,759 AI generated and just you know what we would consider 418 00:26:21,200 --> 00:26:23,680 quote a normal photo. And really there's no such thing, 419 00:26:24,119 --> 00:26:27,319 you know, think about since the introduction of photography, you know, 420 00:26:27,480 --> 00:26:30,480 editing photos pretty much happened at the very start, and 421 00:26:30,519 --> 00:26:33,640 we've been dealing with an issue of manipulated media, you know, 422 00:26:33,640 --> 00:26:36,119 throughout human history, frankly but certainly with all of the 423 00:26:36,160 --> 00:26:40,359 history of photography. And there's no real reason to single 424 00:26:40,400 --> 00:26:43,359 out generative AI as the problem or a problem here. 425 00:26:43,920 --> 00:26:47,640 And so these federal bills that Klobuchar and others are supporting, 426 00:26:47,839 --> 00:26:49,759 I think Josh Holly is a co sponsor of one 427 00:26:49,799 --> 00:26:55,599 of them, requiring disclosures on political ads that use AI. 428 00:26:56,799 --> 00:27:00,200 It's again, it's it's a problem in search of a problem. 429 00:27:00,599 --> 00:27:02,960 You know, there's there's nothing showing that jenerative AI is 430 00:27:03,000 --> 00:27:06,160 particularly harmful, and as they said earlier, I really want 431 00:27:06,160 --> 00:27:08,759 to emphasize jenerit of AI can be a boon to 432 00:27:09,759 --> 00:27:14,759 you know, political challengers, private companies, or individuals challenging the 433 00:27:14,759 --> 00:27:18,039 status quo because they do a pretty good job, much 434 00:27:18,039 --> 00:27:21,440 more cheaply than a team of graphic designers. And you know, 435 00:27:21,480 --> 00:27:25,359 we could see some political power, some corporate power, whatever 436 00:27:25,400 --> 00:27:30,240 the power might be challenged and right ways by individuals 437 00:27:30,279 --> 00:27:34,200 institutions who are the kind of the upstart because they 438 00:27:34,240 --> 00:27:37,880 have these these technologies available to them. That's why we 439 00:27:37,599 --> 00:27:41,799 have very very careful with rules like this. You go ahead. 440 00:27:42,200 --> 00:27:47,119 Speaker 1: Do you think that that's why there are politicians in 441 00:27:47,200 --> 00:27:50,440 some in some degree fighting against this because it is 442 00:27:51,079 --> 00:27:55,160 opening up the field of competition or it's balancing the 443 00:27:55,200 --> 00:27:58,319 field of competition. You don't have the kind of resources 444 00:27:58,400 --> 00:28:01,799 that a big campaign has, certainly an incumbent. You use 445 00:28:01,839 --> 00:28:05,119 this sort of stuff, you start, you know, cutting cost, 446 00:28:05,480 --> 00:28:07,920 you get your message out there a little bit more 447 00:28:08,039 --> 00:28:10,960 than you would have otherwise. Do you think it is 448 00:28:11,000 --> 00:28:15,200 a matter of the big guns trying to quiet competition. 449 00:28:16,640 --> 00:28:19,480 Speaker 2: You know, everyone's motivations are different that could be the 450 00:28:19,519 --> 00:28:22,680 case in some instances. You know, I describe you know, 451 00:28:22,759 --> 00:28:25,240 best motivations is as often as I can. I am 452 00:28:25,279 --> 00:28:27,559 seeing that on a lot of at least what these 453 00:28:27,599 --> 00:28:31,319 politicians are saying. You know that there certainly could be 454 00:28:31,359 --> 00:28:35,599 the case we should be very, very disappointing. But I 455 00:28:35,640 --> 00:28:37,599 would say many of them are motivated by just the 456 00:28:37,799 --> 00:28:40,079 I would say, the conversation. I'll just say that at 457 00:28:40,079 --> 00:28:44,519 the elite's level. You know, it's not a very seldom 458 00:28:44,519 --> 00:28:47,319 where the Atlantics, say, doesn't have an article talking about 459 00:28:47,440 --> 00:28:49,319 the death of truth or the loss of free will 460 00:28:49,440 --> 00:28:52,720 and bringing in AI generated deep fix as one of 461 00:28:52,759 --> 00:28:54,680 the causes of that. So I think it's kind of 462 00:28:54,799 --> 00:28:57,680 just the it's the diet of the media diet they're intaking, 463 00:28:57,759 --> 00:29:00,519 They're they're reading smart people are saying this is going 464 00:29:00,559 --> 00:29:02,839 to be the end of the world, and they apologies 465 00:29:02,880 --> 00:29:04,640 and want to do something about it. And you know, 466 00:29:04,680 --> 00:29:06,880 maybe a second order effect could be like, oh, well, 467 00:29:06,920 --> 00:29:10,000 this could you know, help me as the incumbent versus 468 00:29:10,039 --> 00:29:14,000 my challenger. That could certainly be a perk for some. 469 00:29:14,119 --> 00:29:17,119 But you know, I'm hesitant to describe motives when I 470 00:29:17,240 --> 00:29:19,000 learned not sure entirely positive. 471 00:29:19,480 --> 00:29:23,279 Speaker 1: I certainly understand that defining back to defining AI. Is 472 00:29:23,279 --> 00:29:28,160 it like the old court definition of pornography? You know, 473 00:29:28,359 --> 00:29:31,200 I can't I can't define it, but I know what 474 00:29:31,240 --> 00:29:33,160 when I see it? I mean that? Is that the 475 00:29:33,400 --> 00:29:35,519 argument that we're seeing out there today. 476 00:29:37,000 --> 00:29:39,279 Speaker 2: I think, you know, so we're still figuring it out, Matt. 477 00:29:39,920 --> 00:29:43,440 And so we've been talking about generative AI and political 478 00:29:43,480 --> 00:29:46,720 ad context, but you know, regulating our individual intelligence the 479 00:29:46,720 --> 00:29:51,559 same problem as there absolutely, and legislation is often pulling 480 00:29:51,599 --> 00:29:56,960 from federal spending bills like the NDAA from twenty twenty, 481 00:29:57,640 --> 00:30:01,200 and people analyze that definition of AI. There's all sorts 482 00:30:01,240 --> 00:30:03,880 of problems there. So we've yet to actually arrive at 483 00:30:03,880 --> 00:30:05,920 like what I you know, at least in my opinion, 484 00:30:06,000 --> 00:30:09,279 I think others would agree with me, a good generalized 485 00:30:09,559 --> 00:30:12,240 artificial intelligence definition. I think the way to do it 486 00:30:12,279 --> 00:30:14,480 is to to go narrow. And I think it's easier 487 00:30:14,519 --> 00:30:19,160 to define generative artificial intelligence than just artificial intelligence because 488 00:30:20,200 --> 00:30:23,079 many of the definitions I've read, even about AI would 489 00:30:23,160 --> 00:30:26,160 just you know, they'd wrap up, you know, any like 490 00:30:26,160 --> 00:30:30,279 an Excel spreadsheet, you know, a calculator, or a chess 491 00:30:30,319 --> 00:30:33,079 you know, chess playing program when that's not the intent 492 00:30:33,200 --> 00:30:35,319 of most of these bills. Like what most people and 493 00:30:35,359 --> 00:30:39,799 politicians in particular say, politician policy makers are motivated into 494 00:30:39,839 --> 00:30:43,240 action because of the release of chat GBT and twenty 495 00:30:43,279 --> 00:30:45,640 twenty two, and it's you know, it's wild to think 496 00:30:45,680 --> 00:30:47,960 it was, you know, not that long ago, but that 497 00:30:48,039 --> 00:30:51,599 generative AI system is really spread into action. A lot 498 00:30:51,640 --> 00:30:55,200 of this this energy around regulating AI, whether it be 499 00:30:55,240 --> 00:30:58,640 in the political ad sense or generally, and there's a 500 00:30:58,640 --> 00:31:00,720 long way to go. I think that's the story short 501 00:31:01,279 --> 00:31:05,599 in finding a definition that works, because defining it too 502 00:31:05,599 --> 00:31:09,759 broadly can wrap in way more software and technology than 503 00:31:09,880 --> 00:31:13,359 was intended, and as mentioned earlier, for this little litigation 504 00:31:13,440 --> 00:31:18,200 can result chilling of speech, costs to startups, you know, 505 00:31:18,279 --> 00:31:24,519 the incumbent challengers to medium and large tech companies, and 506 00:31:24,599 --> 00:31:26,200 that's not the world we want to see. We want 507 00:31:26,240 --> 00:31:30,720 to see an innovation ecosystem where anyone can start a company, 508 00:31:30,799 --> 00:31:34,079 can make money, can use these technologies for good and 509 00:31:34,400 --> 00:31:38,440 more often than now, when federal government, state government's getting 510 00:31:38,440 --> 00:31:41,200 involved and trying to you know, define and regulate, it's 511 00:31:41,359 --> 00:31:44,680 it's going overbroad instead of narrowly targeted to what is 512 00:31:44,720 --> 00:31:47,599 the harm They're actually trying to prevent prevent and going 513 00:31:47,640 --> 00:31:49,839 from there and figuring out if you know policy of 514 00:31:49,839 --> 00:31:51,599 old policy is the best solution or not. 515 00:31:54,400 --> 00:31:57,200 Speaker 4: I'm interesting Justice and I'm the Western correspondent to the Federalists. 516 00:31:57,200 --> 00:31:59,759 If you liked last month's book club episode is Myself 517 00:31:59,799 --> 00:32:02,440 and Calm is a Scary when you reviewed Kamala harris 518 00:32:02,519 --> 00:32:05,359 As twenty nineteen memoir The Truths We Hold to join 519 00:32:05,440 --> 00:32:07,799 us on October first for our discussion reviewing former House 520 00:32:07,839 --> 00:32:09,559 Speaker Nancy Pelosi's latest. 521 00:32:09,319 --> 00:32:10,599 Speaker 2: Book, The Art of Power. 522 00:32:11,039 --> 00:32:12,839 Speaker 4: Each month and I will be bringing down to the 523 00:32:12,920 --> 00:32:14,559 latest books that are merged from the top of our 524 00:32:14,599 --> 00:32:15,200 little class. 525 00:32:15,200 --> 00:32:17,519 Speaker 2: So I'd like to invite you to read along with us. 526 00:32:19,119 --> 00:32:22,480 Speaker 1: Well, let's face it, we've been dealing with artificial intelligence 527 00:32:22,720 --> 00:32:26,960 for a very long time before correct concept of digital 528 00:32:27,160 --> 00:32:28,400 artificial intelligence? 529 00:32:28,400 --> 00:32:29,920 Speaker 2: Correct? What have you? I mean? 530 00:32:30,200 --> 00:32:34,599 Speaker 1: You want to look about regulating and investigating a whole 531 00:32:34,720 --> 00:32:39,839 accountable artificial intelligence. Talk to the Central Intelligence Agency. They 532 00:32:40,240 --> 00:32:44,400 have been involved in artificial intelligence for a long time, 533 00:32:44,519 --> 00:32:49,400 as have the FBI and others. But you know that's 534 00:32:49,599 --> 00:32:52,519 that's the problem when we start getting into this deep 535 00:32:52,599 --> 00:32:56,559 fakes have been happening for a long time, and sometimes 536 00:32:56,599 --> 00:32:59,759 they're being done by our own government with our own 537 00:32:59,799 --> 00:33:03,400 tech X dollars. So we need to think about that. 538 00:33:03,440 --> 00:33:07,440 But let me ask the devil's advocate question here, what 539 00:33:07,440 --> 00:33:12,680 what would be so difficult about placing on an AD 540 00:33:13,440 --> 00:33:18,240 that this was that this AD was made with artificial 541 00:33:18,319 --> 00:33:21,960 intelligence or artificial intelligence was a component in the production 542 00:33:22,039 --> 00:33:22,559 of this ad. 543 00:33:24,279 --> 00:33:27,920 Speaker 2: Great question, And you know, I'll say up front, you know, 544 00:33:28,119 --> 00:33:30,319 so there's an assumption there that labels are that will 545 00:33:30,359 --> 00:33:34,519 be effective at I don't know, you know, maybe informing people, 546 00:33:34,599 --> 00:33:36,720 but then you know, is there a reason to inform 547 00:33:36,759 --> 00:33:40,160 a person or not on you know, something is artificially 548 00:33:40,160 --> 00:33:44,160 generated or not. And then there's assumption so it will 549 00:33:44,160 --> 00:33:47,559 be effective. And research shows that, you know, labels are 550 00:33:47,720 --> 00:33:50,839 actually not that you know, but they don't They don't 551 00:33:50,920 --> 00:33:52,839 change people's opinion one way or the other that much. 552 00:33:53,000 --> 00:33:56,880 Speaker 1: They're kind of look at morbid obesis this country, Taylor, 553 00:33:57,000 --> 00:33:59,599 that should say, you know that all the labels that 554 00:33:59,640 --> 00:34:04,039 were placed on all of the fast food value meals 555 00:34:04,039 --> 00:34:07,839 in this country, nobody's a lot of folks aren't really 556 00:34:08,440 --> 00:34:09,920 paying attention to those, and. 557 00:34:10,039 --> 00:34:12,960 Speaker 2: It kind of becomes it becomes so noisy, right, like 558 00:34:13,000 --> 00:34:16,679 seeing the calories and you know Domino's pizza menu or whatever, like, 559 00:34:16,960 --> 00:34:18,880 it's just another number up there, and I just end 560 00:34:19,000 --> 00:34:20,440 up ignoring it because I just want to know what's 561 00:34:20,480 --> 00:34:22,679 the price, what's with the ingredients, and okay, there's a 562 00:34:22,719 --> 00:34:24,480 bunch of other stuff. Yeah. So I think that's a 563 00:34:24,480 --> 00:34:29,039 great point there, and you know, so your question, I 564 00:34:29,079 --> 00:34:33,719 think the difficulty of it is, uh, it's it's that 565 00:34:33,800 --> 00:34:36,679 line again of you know, what what counts as AI generated? 566 00:34:36,679 --> 00:34:38,239 And some of these bills, you know, try to get 567 00:34:38,239 --> 00:34:42,599 explicit about, you know, depicting fake like what a reasonable 568 00:34:42,639 --> 00:34:48,880 person would you know, take to be fake fictional conduct 569 00:34:48,960 --> 00:34:53,400 or behavior by a real person. But even even that definition, man, 570 00:34:53,440 --> 00:34:56,519 gets real tricky and hairy fast, like what is what 571 00:34:56,559 --> 00:35:00,199 do we mean like a reasonable reasonable person? This? 572 00:35:00,360 --> 00:35:03,519 Speaker 1: Yeah, in this environment, who is a reasonable person anymore? 573 00:35:03,559 --> 00:35:06,400 That's where the root of the definition problem comes in. 574 00:35:06,519 --> 00:35:09,480 Speaker 2: I think, Yeah, what counts as you know, ridiculous content 575 00:35:09,599 --> 00:35:12,320 or context content or you know, fake or you know, 576 00:35:12,440 --> 00:35:15,800 could someone could view as fictional? So you have to 577 00:35:16,840 --> 00:35:19,559 then there's the technical side. So say someone you know, 578 00:35:19,599 --> 00:35:23,480 say I post something on on X that I've used 579 00:35:23,719 --> 00:35:27,159 you know, generative AI to produce and it's maybe it's 580 00:35:27,199 --> 00:35:29,960 pulling in like you know, I always use the Rhnda 581 00:35:30,000 --> 00:35:33,280 Santis you know video from twenty twenty three. It's using 582 00:35:33,320 --> 00:35:37,119 real footage interspersed with generated what took what appeared to 583 00:35:37,119 --> 00:35:41,159 be generative AI images. So it's it's not like this 584 00:35:41,199 --> 00:35:45,360 one hundred percent totally generated thing. Maybe that you know, 585 00:35:45,440 --> 00:35:48,880 Kamala Harris parody video that was close to one hundred percent. 586 00:35:48,920 --> 00:35:53,800 I believe generative AI, but you know, it's determining, so 587 00:35:53,840 --> 00:35:55,519 you have that determination has to be made. And then 588 00:35:55,559 --> 00:35:57,719 when does a label actually get applied? Is it, you know, 589 00:36:00,000 --> 00:36:02,880 percent of the content is generated of AI? Does that 590 00:36:03,000 --> 00:36:05,400 need a label or not? Like what actually helps people? 591 00:36:05,440 --> 00:36:07,239 But then it gets to like the effectiveness of labels. 592 00:36:07,239 --> 00:36:10,000 And I'll just a personal example. I wrote an op 593 00:36:10,119 --> 00:36:13,920 ed and submitted it to a mainstream news outlet summarizing 594 00:36:14,000 --> 00:36:15,840 you know AI news of twenty twenty three, and I 595 00:36:16,239 --> 00:36:18,440 said at the top of the article, I used chat 596 00:36:18,440 --> 00:36:21,199 GPT to help me write this, and it was denied 597 00:36:21,239 --> 00:36:25,039 because I used a generative AI to write my article. 598 00:36:25,079 --> 00:36:26,440 And I thought like it just kind of posed an 599 00:36:26,440 --> 00:36:28,639 interesting question, like, oh, like if I hadn't disclosed but 600 00:36:28,760 --> 00:36:32,159 they've been okay, And you know if I had disclosed, 601 00:36:32,159 --> 00:36:33,920 like is there like a certain mixture that would have 602 00:36:33,960 --> 00:36:37,679 been fine or not? And I think that's that's the 603 00:36:37,719 --> 00:36:41,440 technical problem, determining you know, what amount of it was 604 00:36:41,440 --> 00:36:44,639 was generated and the people are talking tools that say 605 00:36:44,679 --> 00:36:48,440 they can prove, like, you know, this video was you know, 606 00:36:48,559 --> 00:36:51,960 using facial recognition software. You know, there's some software that 607 00:36:52,000 --> 00:36:55,599 purports to work because it can measure eyeblinks. So if 608 00:36:55,599 --> 00:36:59,679 there's a video out there of me speaking human beings, blink, 609 00:36:59,760 --> 00:37:03,360 I think more often than most AI models generating video, 610 00:37:03,400 --> 00:37:04,920 but that's probably going to be a patch soon, so 611 00:37:04,960 --> 00:37:08,480 that's not going to work forever. So it just gets very, 612 00:37:08,559 --> 00:37:12,239 very difficult to accurately label quickly at scale. It's those 613 00:37:12,280 --> 00:37:14,880 three components, Matt, that are the real tricky part. You 614 00:37:14,920 --> 00:37:18,320 think of the volume of content, the data that individuals 615 00:37:18,320 --> 00:37:22,360 across the globe are posting online every single day on 616 00:37:22,679 --> 00:37:25,679 platforms large and small, you know who is actually doing 617 00:37:25,679 --> 00:37:28,039 a label and how are they doing it? And then 618 00:37:28,159 --> 00:37:31,119 is that actually going to meet the goal of helping 619 00:37:31,159 --> 00:37:35,760 citizens and individuals discern you know, true and correct information, 620 00:37:35,840 --> 00:37:37,599 And then that becomes its own issue of like what 621 00:37:37,679 --> 00:37:40,480 is true and product information. The Hunter Biden laptop is 622 00:37:40,480 --> 00:37:43,079 a great example of how how tricky that can be 623 00:37:43,480 --> 00:37:45,840 and how the story can change over time is more 624 00:37:45,840 --> 00:37:49,480 information is released and you know that's its own I'm 625 00:37:49,519 --> 00:37:51,719 sure you guys have covered that well, as of many 626 00:37:51,719 --> 00:37:55,199 other outlets, but that is I think kind of unpacking. 627 00:37:55,280 --> 00:37:57,360 That's how it unpack and maybe answer that question, it's 628 00:37:57,400 --> 00:38:00,960 it's not just a simple you know, hey, AI was 629 00:38:01,039 --> 00:38:02,559 used here or not. It's you know, you have to 630 00:38:02,559 --> 00:38:05,679 do it accurately, at scale and in a way that 631 00:38:05,760 --> 00:38:08,960 actually helps helps people understand. Absolutely. 632 00:38:09,960 --> 00:38:13,559 Speaker 1: Yeah, and you know, you think about the events, the 633 00:38:13,599 --> 00:38:16,719 major events over the last four plus years in this country. 634 00:38:16,719 --> 00:38:20,000 The Hunter Biden story is a perfect example. You'll have 635 00:38:20,079 --> 00:38:25,159 to forgive me and my fellow Americans, many of them, 636 00:38:25,199 --> 00:38:27,039 when we get a little bit itchy, we get a 637 00:38:27,039 --> 00:38:31,679 little bit concerned about government agencies created to be bureaus 638 00:38:31,880 --> 00:38:38,639 of tracking disinformation, misinformation, and malinformation. We've been down this 639 00:38:38,880 --> 00:38:42,800 road before, so there is some clear concern about the 640 00:38:42,800 --> 00:38:47,679 government stepping in to be the arbiter of what is 641 00:38:48,079 --> 00:38:52,039 correct information, what is good information, and what is bad information. 642 00:38:52,199 --> 00:38:56,280 Let's close with this. It's the what's next questions as 643 00:38:56,320 --> 00:39:00,239 always because we're at a pivotal point in AI and 644 00:39:00,880 --> 00:39:04,199 the election season. What is next on this front? 645 00:39:04,199 --> 00:39:07,519 Speaker 2: As you see, well, my team and I at thebund 646 00:39:07,559 --> 00:39:11,639 Ins Institute, we'll keep tracking all mentions of AI and 647 00:39:11,679 --> 00:39:15,280 the US election and look forward to our report part 648 00:39:15,320 --> 00:39:17,320 four in October, and then we'll have an actor action 649 00:39:17,400 --> 00:39:20,400 after November fifth. And you know who, no one can 650 00:39:20,400 --> 00:39:23,239 break the future accurately. But if if passed is any 651 00:39:23,280 --> 00:39:26,199 sort of prologue, I don't believe that the doom and 652 00:39:26,199 --> 00:39:29,039 gloom scenarios are likely to come true. I don't think 653 00:39:29,079 --> 00:39:32,280 we'll see, you know, people not showing up at the 654 00:39:32,320 --> 00:39:36,760 polls because of some generative AI, audio, video, photo, whatever 655 00:39:36,800 --> 00:39:39,519 the case might be. But it's important to pay attention 656 00:39:39,559 --> 00:39:41,159 in the list and I think, you know, that's why 657 00:39:41,159 --> 00:39:44,079 I've appreciated this, this line of research, and I think 658 00:39:44,079 --> 00:39:47,559 we'll see let's see, you know, the SCC proceeding is 659 00:39:47,559 --> 00:39:50,079 happening right now. As I mentioned, we're going at the 660 00:39:50,119 --> 00:39:53,960 first stage of this that agency promulgating rules on the 661 00:39:54,000 --> 00:39:57,079 use of AI and political ads. Listeners should absolutely take 662 00:39:57,079 --> 00:40:02,039 a look at what they're proposing. Deadline for comment is today, 663 00:40:02,559 --> 00:40:04,639 so Congress is not likely to act because they just 664 00:40:04,639 --> 00:40:07,760 have a couple of weeks left of getting stuff done 665 00:40:07,800 --> 00:40:09,760 and they're wrapped up in funding battles and all that 666 00:40:09,800 --> 00:40:12,199 sort of stuff. And the states, I think pay attention 667 00:40:12,320 --> 00:40:14,800 in twenty twenty five. You know, when state legislative sessions 668 00:40:15,239 --> 00:40:18,559 get back into gear, I believe more states will act. 669 00:40:18,639 --> 00:40:21,639 And I you know, in many ways, you know, California 670 00:40:21,719 --> 00:40:25,440 acting cannot inspire some states to act, but in other 671 00:40:25,480 --> 00:40:28,480 ways that can. You know, some I think some legislators 672 00:40:28,480 --> 00:40:30,800 look to California and say, hmm, we don't want to 673 00:40:30,800 --> 00:40:33,519 be like California, so maybe we won't do that right. 674 00:40:34,079 --> 00:40:37,760 And but I think then and nevertheless, you know, state 675 00:40:37,960 --> 00:40:40,920 lawmakers are and governors are hearing from constituents there again, 676 00:40:41,599 --> 00:40:44,119 and take this media diet of doom and gloom, and 677 00:40:44,159 --> 00:40:46,159 I just want to say, like if I have you know, 678 00:40:46,280 --> 00:40:49,360 a couple of suggestions solutions here is you know, there's 679 00:40:49,360 --> 00:40:52,880 a stop panicking. This is not a story where the 680 00:40:52,920 --> 00:40:57,159 prognostications have proved have proven true. If if anything, it's 681 00:40:57,239 --> 00:41:00,280 just the old old style photoshop up. 682 00:41:01,719 --> 00:41:02,000 Speaker 4: Uh. 683 00:41:02,119 --> 00:41:06,639 Speaker 2: You know, bad actors putting in miss and disinformation into 684 00:41:06,639 --> 00:41:09,400 the into the stream that are the cause of issues. 685 00:41:09,400 --> 00:41:11,360 And even did I'll say the Department of Justice did 686 00:41:11,360 --> 00:41:15,159 a report on you know, like Iranian Russian election interference, 687 00:41:15,199 --> 00:41:17,159 and they said generative AI has not been a major 688 00:41:17,199 --> 00:41:18,920 feature of what they've been doing. They're just sticking to 689 00:41:18,920 --> 00:41:22,840 the old school playbook as far as we know. And 690 00:41:22,880 --> 00:41:24,840 so the panic can create this false sense