WEBVTT 1 00:00:18.160 --> 00:00:21.000 And we are back with another edition of the Federalist 2 00:00:21.079 --> 00:00:25.399 Radio Hour. I'm Matt Kittle, Senior Elections correspondent at the Federalist, 3 00:00:25.920 --> 00:00:30.760 and your experience sharpa on today's quest for Knowledge. As always, 4 00:00:30.760 --> 00:00:33.640 you can email the show at radio at the Federalist 5 00:00:33.719 --> 00:00:38.359 dot com, follow us on ex at FDRLST, make sure 6 00:00:38.399 --> 00:00:41.560 to subscribe wherever you download your podcast, and of course 7 00:00:41.640 --> 00:00:45.000 to the premium version of our website as well. Our 8 00:00:45.039 --> 00:00:49.640 guest today is Neil Chilson, former FTC Chief Technologist and 9 00:00:49.840 --> 00:00:55.600 currently head of aipolicy with the Abundance Institute. Where are 10 00:00:55.640 --> 00:01:02.520 we taking artificial intelligence? Where is artificial intelligence taking us? Well? 11 00:01:02.520 --> 00:01:05.560 We ask those questions and many more on this edition 12 00:01:05.640 --> 00:01:07.959 of the Federalist Radio Hour. Neil, thank you so much 13 00:01:08.000 --> 00:01:12.159 for joining us. It's great to be here speaking of here. 14 00:01:13.519 --> 00:01:17.400 Is this really you? Or is this AI? Or as 15 00:01:17.439 --> 00:01:23.079 they used to say when I was much younger dating 16 00:01:23.159 --> 00:01:27.439 myself here, is it live or is it memorys? We've 17 00:01:27.439 --> 00:01:31.760 come a long way since cassette tapes, certainly in America. 18 00:01:32.560 --> 00:01:35.519 I jest, of course, this is the real Neil Chilsen, 19 00:01:35.599 --> 00:01:41.400 but it does raise the question of just how much 20 00:01:42.040 --> 00:01:46.640 we are inundated with AI and the difficulty sometimes of 21 00:01:47.239 --> 00:01:54.040 telling artificial intelligence from reality. I think this thing is 22 00:01:54.200 --> 00:01:57.879 only going to intensify, of course, as we move forward. 23 00:01:58.319 --> 00:02:00.400 Where are we going with all of this? 24 00:02:01.840 --> 00:02:04.439 Well, as much as my wife and my kids might 25 00:02:04.480 --> 00:02:09.280 appreciate an AI upgraded dad and husband, they're stuck with me, 26 00:02:09.360 --> 00:02:10.840 and so are you guys, the real me. 27 00:02:13.080 --> 00:02:15.360 But you know that's I say that sort of jokingly. 28 00:02:15.439 --> 00:02:19.360 But where we are right now with AI is continuing 29 00:02:19.439 --> 00:02:25.599 a long trend of adding new capabilities to computers. And 30 00:02:26.080 --> 00:02:30.520 they're surprising in this case, and they're surprising in some ways. 31 00:02:31.159 --> 00:02:34.639 But the history of artificial intelligence, a term that was 32 00:02:34.680 --> 00:02:38.879 coined in the nineteen fifties, has been exactly this, surprising 33 00:02:38.919 --> 00:02:43.800 new things everybody gets excited about. We figure out, oh, 34 00:02:43.879 --> 00:02:48.039 this is not actually the same thing as human intelligence, 35 00:02:48.080 --> 00:02:52.840 not exactly. It's powerful, and then we sort of it 36 00:02:52.840 --> 00:02:55.439 gets adapted. It's in our phones, it's in our computers, 37 00:02:55.439 --> 00:02:58.120 and we move on. So, you know, artificial intelligence. The 38 00:02:58.159 --> 00:03:01.439 cutting edge of artificial intelligence technology when I was first 39 00:03:01.439 --> 00:03:02.879 getting into computers in the. 40 00:03:03.000 --> 00:03:04.919 Nineties was chess playing. 41 00:03:05.680 --> 00:03:09.240 And now everybody's phone can play chess better than ninety 42 00:03:09.319 --> 00:03:11.520 nine point nine percent of humans on the planet, and 43 00:03:11.960 --> 00:03:16.360 we don't think of that as somehow you know, terminator 44 00:03:16.400 --> 00:03:18.280 style artificial intelligence. 45 00:03:18.280 --> 00:03:19.319 And so so. 46 00:03:19.319 --> 00:03:22.800 Where we are now is we have these really powerful 47 00:03:23.240 --> 00:03:27.199 tools called large language models that can be used for many, 48 00:03:27.240 --> 00:03:30.159 many different types of things. But one of the ways 49 00:03:30.199 --> 00:03:33.039 they're being used is in a sort of chatbot form 50 00:03:33.199 --> 00:03:37.599 where you can ask questions and get really comprehensive, detailed, 51 00:03:38.000 --> 00:03:44.319 sometimes made up answers, and that's that's really powerful. It 52 00:03:44.400 --> 00:03:49.080 teaches us that there's a lot to learn about humans 53 00:03:50.039 --> 00:03:53.520 that you can that you can collect and gather into 54 00:03:53.599 --> 00:03:57.199 computers in these in these formats and then query in 55 00:03:57.240 --> 00:04:00.639 a way that gets you know, very persuasive, very interesting, 56 00:04:01.719 --> 00:04:06.199 often very entertaining content, whether it be text, video, audio. 57 00:04:06.680 --> 00:04:11.360 I love making songs. Actually, my little girls love making 58 00:04:11.439 --> 00:04:14.240 up new songs using some of the apps out there 59 00:04:14.919 --> 00:04:17.040 where you can just type in a prompt and get like, 60 00:04:17.240 --> 00:04:19.560 you know, a song about unicorns or a song about 61 00:04:19.600 --> 00:04:23.959 you know, their kids club running around the neighborhoods as 62 00:04:24.279 --> 00:04:26.319 kids spies or something like that, and so like we 63 00:04:26.399 --> 00:04:29.199 love that stuff. And so yeah, there's a lot of entertainment. 64 00:04:29.240 --> 00:04:31.160 There's a lot of power in these tools, and there's 65 00:04:31.160 --> 00:04:34.079 some risks and people need to be aware of that, 66 00:04:34.120 --> 00:04:35.560 and policymakers do as well. 67 00:04:36.160 --> 00:04:39.000 Some of the uses that you mentioned, you know, are 68 00:04:39.079 --> 00:04:42.839 fun to play around with. Certainly, like you said making 69 00:04:42.920 --> 00:04:47.160 up songs, I've seen that in play and the technology 70 00:04:47.279 --> 00:04:51.720 is really quite good and it takes you literally seconds 71 00:04:51.759 --> 00:04:55.240 to do what the Beatles spent, you know, the better 72 00:04:55.279 --> 00:04:58.480 part of a year doing to make Sergeant Pepper's Lonely 73 00:04:58.560 --> 00:05:01.079 Hearts Club band. I'm not sure saying that the artistry 74 00:05:01.279 --> 00:05:04.800 is the same, but I'm you know, the production value 75 00:05:04.920 --> 00:05:08.079 is certainly there. Yeah, the risk that you talk about 76 00:05:08.120 --> 00:05:14.480 in this arena copyright infringement, Where are we with that today? 77 00:05:14.560 --> 00:05:19.480 And where are we heading in terms of intellectual property? 78 00:05:20.120 --> 00:05:23.279 So there's two concerns on the copyright front. One is 79 00:05:23.959 --> 00:05:26.759 on the training side, So when a company is building 80 00:05:26.800 --> 00:05:30.079 these models and they use a bunch of different content 81 00:05:30.199 --> 00:05:33.639 in order to train on, what are the legal restrictions 82 00:05:33.680 --> 00:05:36.240 on that? And then the other side of it is 83 00:05:36.279 --> 00:05:39.040 on the output side. So when a user types in 84 00:05:39.079 --> 00:05:41.839 a prompt and says, you know, give me a picture 85 00:05:41.839 --> 00:05:46.040 of Mickey Mouse, you know, battling the Marvel team or 86 00:05:46.079 --> 00:05:49.399 something like that, and the model puts out something that 87 00:05:49.519 --> 00:05:56.040 includes you know, maybe some sort of intellectual property protection 88 00:05:56.439 --> 00:05:57.600 protect and content. 89 00:05:58.519 --> 00:06:01.120 Who's responsible for that? Is it the user, is it 90 00:06:01.240 --> 00:06:03.000 the model? And so on. 91 00:06:03.120 --> 00:06:05.639 The first thing about like the sort of ingestion of 92 00:06:05.680 --> 00:06:08.000 this content, the use of this content. We've had some 93 00:06:08.079 --> 00:06:13.000 court decisions and there is a category of this that 94 00:06:13.680 --> 00:06:16.480 judges are thinking of as fair use right. And so 95 00:06:16.519 --> 00:06:19.639 you're training these models the way you might train you know, 96 00:06:19.759 --> 00:06:23.439 anybody reading a book, and so the content of the 97 00:06:23.439 --> 00:06:27.079 model or the content isn't copied into the model, It 98 00:06:27.160 --> 00:06:29.319 is used to train the model, and then the model 99 00:06:29.439 --> 00:06:32.720 understands that content in the context of you know, lots 100 00:06:32.759 --> 00:06:35.600 of other content. And so there are ways in which 101 00:06:35.600 --> 00:06:39.480 the courts are saying that that is fair use. There 102 00:06:39.519 --> 00:06:43.399 was an anthropic settlement in this space where the big 103 00:06:43.439 --> 00:06:47.040 problem the anthropic had was that they hadn't paid for 104 00:06:47.120 --> 00:06:49.600 the original content, right, and so they hadn't bought the 105 00:06:49.600 --> 00:06:53.439 books that they were scanning. They had just downloaded pirated 106 00:06:53.439 --> 00:06:55.199 copies of the book. And the court said, no, no, no, 107 00:06:55.240 --> 00:06:57.519 you can't do that. You need to buy the content. 108 00:06:57.920 --> 00:07:00.000 You need to buy you the copy of the content 109 00:07:00.040 --> 00:07:03.439 and use that to train. You can't just use pirated content. 110 00:07:04.120 --> 00:07:07.560 But overall, that doesn't that doesn't say that much about 111 00:07:08.439 --> 00:07:10.360 you know, I think that still leaves the door open 112 00:07:10.399 --> 00:07:12.480 for companies to train on a lot of this data, 113 00:07:12.879 --> 00:07:15.319 which I think is probably the right decision. I think 114 00:07:15.360 --> 00:07:19.040 it is very analogous to training to reading a book. 115 00:07:20.639 --> 00:07:23.800 And therefore, you know, we don't have copyright that says 116 00:07:23.800 --> 00:07:26.959 you can't learn from a book without permission from the 117 00:07:27.000 --> 00:07:30.639 copyright holder, and so on the other side, we don't know. 118 00:07:30.720 --> 00:07:34.439 We don't know yet. Those cases are still ongoing. We 119 00:07:34.519 --> 00:07:38.439 don't know how courts are going to partition liability for 120 00:07:39.000 --> 00:07:44.160 copyright infringement between the user who writes the prompt and 121 00:07:44.319 --> 00:07:47.639 the company who is running the model or trained the model. 122 00:07:48.480 --> 00:07:51.800 There's some really complicated questions there about what is infringement 123 00:07:52.439 --> 00:07:55.519 even on that side, but then splitting it up between 124 00:07:55.560 --> 00:07:57.959 the user who's asking for something and the model who's 125 00:07:57.959 --> 00:08:01.240 generating it, it gets pretty complicated there. You know, we 126 00:08:01.279 --> 00:08:05.279 don't we don't sue pencil companies because people draw images 127 00:08:05.319 --> 00:08:09.120 of of you know, Mickey Mouse, sure with using a pencil, 128 00:08:09.399 --> 00:08:10.319 and that wouldn't. 129 00:08:09.959 --> 00:08:10.639 Make a ton of sense. 130 00:08:10.680 --> 00:08:12.680 And so we're just trying to figure out the right 131 00:08:12.720 --> 00:08:15.720 analogies here and what makes sense economically. 132 00:08:15.720 --> 00:08:19.959 Still, that's very complicated in the same legal neighborhood. What 133 00:08:20.079 --> 00:08:25.160 about reputational damage. We've already seen some cases on that front. 134 00:08:25.240 --> 00:08:30.639 And listen, there are AI companies you know that are 135 00:08:31.040 --> 00:08:35.679 I think very responsible AI companies. They're doing what they 136 00:08:35.759 --> 00:08:39.120 can and they're investing a lot of money to make 137 00:08:39.159 --> 00:08:42.960 sure that these systems are working right and they don't 138 00:08:43.000 --> 00:08:49.039 cause damage. But sometimes they do reputational damage. What about 139 00:08:49.080 --> 00:08:54.120 the liability issue on that front? Where are the courts 140 00:08:54.279 --> 00:08:55.480 on that particular matter? 141 00:08:56.080 --> 00:08:59.559 So the big, the big challenge. The courts have looked 142 00:08:59.600 --> 00:09:02.159 at this a couple of ways. There's a bunch of 143 00:09:02.240 --> 00:09:07.240 challenges when you're talking about defamation or libel. The biggest 144 00:09:07.320 --> 00:09:10.759 challenge is that often this involves public figures, and public 145 00:09:10.799 --> 00:09:15.000 figure liability requires showing that there was a sort of 146 00:09:15.039 --> 00:09:16.360 malicious intent. 147 00:09:17.919 --> 00:09:18.679 By the party. 148 00:09:19.080 --> 00:09:21.600 And the question again, the same question comes up here 149 00:09:21.639 --> 00:09:24.039 that comes up in the generation who is the party 150 00:09:24.360 --> 00:09:26.399 at fault here? Is it the person who typed the 151 00:09:26.440 --> 00:09:29.360 prompt who said, like, tell me everything you know about 152 00:09:29.399 --> 00:09:33.279 you know X public figure? Or is it the model generation? 153 00:09:34.679 --> 00:09:38.399 And then on top of that, unless it's made public, 154 00:09:39.519 --> 00:09:43.720 this type of generation by itself is not. I don't 155 00:09:43.720 --> 00:09:48.519 think it's usually subject to defamation law. So the question 156 00:09:48.720 --> 00:09:51.639 is if I generate, If I type a prompt into 157 00:09:51.720 --> 00:09:56.679 chat GPT and I get inaccurate information back, but then 158 00:09:56.840 --> 00:09:59.759 I publish it on my social media website or in 159 00:09:59.799 --> 00:10:03.799 a newspaper article or something, it's the publication of that 160 00:10:03.799 --> 00:10:05.879 that is the real problem. It's not the fact that 161 00:10:06.639 --> 00:10:09.360 the chat GPT came up with it, because nobody sees 162 00:10:09.399 --> 00:10:11.120 that except the person who asks for it. And so 163 00:10:11.600 --> 00:10:15.399 I do think that these defamation cases against the models 164 00:10:15.440 --> 00:10:18.200 are going to be complicated by that factor. I think 165 00:10:18.240 --> 00:10:20.440 that it's difficult to say that the model is that 166 00:10:20.600 --> 00:10:24.679 fault for the distribution of a falsehood when the model 167 00:10:25.200 --> 00:10:28.320 just generates content and then the person has to make 168 00:10:28.320 --> 00:10:29.120 a choice to then. 169 00:10:29.039 --> 00:10:32.279 Put it out into the world. And so I think 170 00:10:32.279 --> 00:10:34.080 that's a complicating factor. And I think that. 171 00:10:34.000 --> 00:10:37.240 Means while the models are trying very hard to be 172 00:10:37.360 --> 00:10:40.759 accurate on this sort of stuff and users, it's still 173 00:10:40.799 --> 00:10:43.679 really on the users to verify that the content that 174 00:10:43.720 --> 00:10:46.759 they're posting, that they're taking from that and using in 175 00:10:46.799 --> 00:10:49.679 the real world is accurate. And I think that's where 176 00:10:49.720 --> 00:10:51.240 the liability probably should sit. 177 00:10:51.559 --> 00:10:54.240 Yeah, tied end to that. We have some very interesting 178 00:10:54.440 --> 00:10:58.679 cases on free speech. We have some and AI. We 179 00:10:58.759 --> 00:11:07.480 have some some concerning moves by politicians, by lawmakers. I'm 180 00:11:07.519 --> 00:11:09.559 thinking of a guy who would like to be president 181 00:11:09.600 --> 00:11:11.840 of the United States in twenty twenty eight. There's no 182 00:11:11.879 --> 00:11:16.399 doubt about that. Gavin Newsom in California and the battles 183 00:11:16.480 --> 00:11:22.480 that have gone on there legally speaking in terms of 184 00:11:23.360 --> 00:11:29.759 content that is driven for political advertising political parody. Really, 185 00:11:30.480 --> 00:11:36.000 that's the interesting part about this, the brave new world 186 00:11:36.120 --> 00:11:41.279 of political advertising or political parody where, for instance, you 187 00:11:41.399 --> 00:11:47.000 have a famous AI video that came out last year 188 00:11:47.720 --> 00:11:52.559 that had Kamala Harris, then the Vice President of the 189 00:11:52.679 --> 00:11:56.440 United States, and a candidate, the candidate the Democrats candidate 190 00:11:56.440 --> 00:12:01.279 for president. The AI technology made her say some things 191 00:12:01.320 --> 00:12:05.600 that she certainly did not say. It was very amusing 192 00:12:05.879 --> 00:12:10.320 to about half of the country. Wasn't so amusing to 193 00:12:10.360 --> 00:12:13.759 the other half. So Gavin Newsom and crew in California 194 00:12:14.360 --> 00:12:18.360 took that and some other instances and used that as. 195 00:12:20.039 --> 00:12:24.399 Kind of a red flag law for or a red flag. 196 00:12:24.080 --> 00:12:29.679 Incidance for AI and communication and put limits on what 197 00:12:29.799 --> 00:12:32.399 you could do or penalties on what you could do 198 00:12:32.919 --> 00:12:36.200 if you did this. Kind of thing. The courts have 199 00:12:36.519 --> 00:12:39.279 entered into this case. Where does all of that stand today? 200 00:12:40.440 --> 00:12:44.159 So yeah, So California and some other states have looked 201 00:12:44.240 --> 00:12:48.879 at how to govern the use of AI generated content 202 00:12:48.960 --> 00:12:53.000 in advertising, in political advertising in particular, and they faced 203 00:12:53.000 --> 00:12:59.159 some real constitutional challenges here. Political speeches right at the 204 00:12:59.200 --> 00:13:04.519 core of First Amendment rights. Lying in political speech is 205 00:13:04.559 --> 00:13:11.960 not policed by the courts as because they often say, essentially, 206 00:13:12.559 --> 00:13:14.960 now you might be able to bring a defamation case possibly, 207 00:13:15.000 --> 00:13:18.759 but again we're talking about public figures, and even in 208 00:13:18.759 --> 00:13:21.919 the political context that gets even harder. Courts tend to 209 00:13:22.000 --> 00:13:25.159 really say, like, if you're going to do political ads, 210 00:13:25.159 --> 00:13:28.120 it's going to be up to the voters to decide 211 00:13:28.279 --> 00:13:33.200 who's telling the truth. And and AI doesn't really change 212 00:13:33.240 --> 00:13:36.039 that that much. It's always been easy to create false 213 00:13:36.120 --> 00:13:40.120 content right now. What you can do maybe with AI 214 00:13:40.399 --> 00:13:42.919 is the types of deep fakes that you're talking about, 215 00:13:43.000 --> 00:13:46.679 the types of putting words in somebody's mouth in a 216 00:13:46.840 --> 00:13:48.279 very convincing way. 217 00:13:48.879 --> 00:13:52.039 I think deceptive content. 218 00:13:51.759 --> 00:13:55.039 Like that is still it's right for you know, competitors 219 00:13:55.080 --> 00:13:58.799 to call it out. I think that companies are trying 220 00:13:58.799 --> 00:14:02.279 to figure out how to balance at but political parity 221 00:14:02.519 --> 00:14:06.559 is highly protected free speech, and I think any type 222 00:14:06.600 --> 00:14:10.440 of you know, government thumb on the scale about what 223 00:14:10.519 --> 00:14:13.879 people can and can't say is just ripe for abuse. 224 00:14:14.000 --> 00:14:16.639 You'll end up getting police, you know, from one political 225 00:14:16.679 --> 00:14:20.360 perspective or another. One party will will bear the brunt 226 00:14:20.360 --> 00:14:22.399 of this more than the other. And I just think 227 00:14:22.440 --> 00:14:26.000 that it's a super risky road to walk down. The 228 00:14:26.039 --> 00:14:30.159 better the better options here are more speech, not censoring 229 00:14:30.200 --> 00:14:32.600 the ability to create content in the first place. 230 00:14:33.320 --> 00:14:35.840 And I think that some of the companies. 231 00:14:35.399 --> 00:14:39.639 Have had some early experiences in trying to shape the 232 00:14:39.679 --> 00:14:43.519 content that was coming out of their models, and I 233 00:14:43.600 --> 00:14:46.879 think they've dialed that back to say, like, hey, we're 234 00:14:46.919 --> 00:14:50.600 going to largely try to lean on the side of 235 00:14:51.080 --> 00:14:56.200 generating what the user is asking for and protects protect 236 00:14:56.279 --> 00:14:59.159 speech there as long as they're not generating illegal content. 237 00:14:59.240 --> 00:15:01.759 I think that's the better that's the better frame. 238 00:15:02.120 --> 00:15:04.840 I hope that more and more companies move towards that direction. 239 00:15:08.200 --> 00:15:11.639 Did a single company save the stock market from crashing 240 00:15:11.679 --> 00:15:14.519 into a recession? The Watchdot on Wall Street podcast with 241 00:15:14.639 --> 00:15:17.759 Chris Markowski. Every day, Chris helps unpack the connection between 242 00:15:17.840 --> 00:15:20.279 politics and the economy and how it affects your wallet. 243 00:15:20.399 --> 00:15:24.320 Tech powerhouse in Nvidia's earnings report did not disappoint, But 244 00:15:24.399 --> 00:15:26.840 what does that tell you about the value of AI? 245 00:15:27.240 --> 00:15:30.360 This cannot save the market forever. Whether it's happening in 246 00:15:30.440 --> 00:15:32.840 DC or down on Wall Street, it's affecting you financially. 247 00:15:32.919 --> 00:15:33.440 Be informed. 248 00:15:33.519 --> 00:15:35.360 Check out the watch Dot on Wall Street podcast with 249 00:15:35.440 --> 00:15:38.720 Chris Markowski on Apple, Spotify or wherever you get your podcasts. 250 00:15:43.159 --> 00:15:48.159 Well, let's face it, if the courts really did punish 251 00:15:48.960 --> 00:15:54.799 politicians for lying, our correctional system would be just overbroad. 252 00:15:55.840 --> 00:16:01.519 Yeah, political speech is full of half to true misframeans 253 00:16:02.200 --> 00:16:04.679 and courts just don't want to get into that. I 254 00:16:04.720 --> 00:16:09.039 think that's that's just fraught territory, and I think I 255 00:16:09.039 --> 00:16:11.039 think it's right to leave it to the people to 256 00:16:11.080 --> 00:16:12.759 make the decisions about who they trust. 257 00:16:13.080 --> 00:16:17.759 Yeah, that is America, and that's the bottom line. The information, 258 00:16:17.919 --> 00:16:22.480 of course is it's a different age for all of that. 259 00:16:22.679 --> 00:16:26.600 But we've had the same issues and the same problems 260 00:16:26.919 --> 00:16:28.879 over the two hundred and fifty years of this one. 261 00:16:30.240 --> 00:16:33.840 Let's turn our attention to the emotional aspects of AI. 262 00:16:34.480 --> 00:16:38.679 Harvard Business School has an interesting piece up came out recently. 263 00:16:38.799 --> 00:16:43.879 Feeling lonely an attentive listener is an AI prompt away, 264 00:16:44.399 --> 00:16:49.639 and it delves into the brave new world of companionship 265 00:16:50.559 --> 00:16:55.440 with AI. There are well, I think what we're finding out, Neil, 266 00:16:55.720 --> 00:16:59.159 over the last year or so is that there are 267 00:16:59.519 --> 00:17:02.919 a lot of lonely people out there, and they have 268 00:17:03.039 --> 00:17:09.000 turned AI to solve that loneliness problem. Where does all 269 00:17:09.079 --> 00:17:11.920 of that stand today, and where do you think that's 270 00:17:11.960 --> 00:17:12.519 all going? 271 00:17:13.839 --> 00:17:15.759 Yeah, I mean, I think you're totally right. I mean, 272 00:17:15.799 --> 00:17:19.759 we do have an epidemic of sort of people isolating 273 00:17:19.759 --> 00:17:23.599 themselves from others, and I think this is an outgrowth of, 274 00:17:23.799 --> 00:17:25.799 you know, some really bad policies that we had around 275 00:17:25.839 --> 00:17:33.119 the COVID pandemic, as well as just some general fracturing 276 00:17:33.200 --> 00:17:40.160 of American you know, socialization institutions that we've you know, 277 00:17:40.319 --> 00:17:42.559 historically relied on to bring us together with people. 278 00:17:43.920 --> 00:17:45.000 And so I. 279 00:17:44.920 --> 00:17:49.799 Think that trying to fill this gap with AI chatbots 280 00:17:49.960 --> 00:17:53.799 is at best a sort of temporary measure. 281 00:17:55.480 --> 00:17:56.079 I think that. 282 00:17:57.680 --> 00:18:01.359 I don't know how to exactly measure this. You know, 283 00:18:01.400 --> 00:18:05.119 these chatbots are general purpose. Most of them are general purpose. 284 00:18:05.160 --> 00:18:07.039 They're not aimed at this sort of function. Now, there 285 00:18:07.079 --> 00:18:10.359 are some companies who are offering specifically this type of function, 286 00:18:10.680 --> 00:18:12.960 but most of the chatbots that people are using our 287 00:18:13.079 --> 00:18:15.599 general purpose. People are using them for a wide range 288 00:18:15.599 --> 00:18:19.880 of things, from academic research to generating funny videos, and 289 00:18:19.960 --> 00:18:23.359 some people use them occasionally to talk to about like 290 00:18:23.680 --> 00:18:28.359 personal problems or you know, relationship problems or things like that. 291 00:18:28.839 --> 00:18:29.640 I think some of that. 292 00:18:29.559 --> 00:18:34.079 Can be very useful and very helpful, but certainly would 293 00:18:34.079 --> 00:18:36.799 want to keep an eye out on people replacing time 294 00:18:36.839 --> 00:18:43.039 with other humans with these these chatbots. It's just it's 295 00:18:43.119 --> 00:18:46.680 just not it's not the same obviously, and it doesn't 296 00:18:46.720 --> 00:18:51.680 create the same types of deep connections with community that 297 00:18:51.759 --> 00:18:54.480 I think are essential to human flourishing. 298 00:18:54.559 --> 00:18:54.880 And so. 299 00:18:56.519 --> 00:18:59.799 My hope would be that these systems, as they build out, 300 00:19:00.200 --> 00:19:04.839 are aimed at improving people's ability to engage with other people, 301 00:19:05.279 --> 00:19:07.680 because there certainly are people who are not as practice 302 00:19:07.720 --> 00:19:11.400 at that who maybe spent you know, two years, especially 303 00:19:11.400 --> 00:19:13.200 when we get into the kids sector, who spent maybe 304 00:19:13.240 --> 00:19:17.240 two years like doing zoomed school and need to get 305 00:19:17.279 --> 00:19:21.240 back into the swing of, you know, talking to other people. 306 00:19:21.279 --> 00:19:24.920 And I think these tools could can help give tips 307 00:19:24.920 --> 00:19:25.720 on that sort of thing. 308 00:19:26.000 --> 00:19:26.720 But I hope they don't. 309 00:19:27.400 --> 00:19:29.799 I hope people don't rely on them as a substitute 310 00:19:29.839 --> 00:19:32.960 for reaching out, being brave, talking to people they haven't 311 00:19:32.960 --> 00:19:36.359 talked to before, and getting to know people in their community. 312 00:19:36.920 --> 00:19:40.279 We certainly have learned over the last few years just 313 00:19:40.400 --> 00:19:45.400 how dangerous, how much peril lockdown policies have put our 314 00:19:45.519 --> 00:19:49.960 children in especially, but society in general, when it comes 315 00:19:50.000 --> 00:19:55.160 to picking up with communication and relationships and all of 316 00:19:55.200 --> 00:19:58.640 those sorts of things. You know, Neil, though, on this topic, 317 00:19:58.799 --> 00:20:02.039 they always say the heart wants what the heart wants, 318 00:20:02.480 --> 00:20:05.599 The market wants what the market wants. And there are 319 00:20:05.680 --> 00:20:09.519 some strange areas in the marketplace for AI. Not to 320 00:20:09.559 --> 00:20:14.160 say that, you know, there isn't a market for it, 321 00:20:14.279 --> 00:20:17.000 but the question maybe should there be a market for it. 322 00:20:17.319 --> 00:20:22.400 One of those areas is connecting with the dead, and 323 00:20:22.440 --> 00:20:25.640 this has become an interesting subject of late. But the 324 00:20:25.720 --> 00:20:31.240 AI products that offer people the ability to chat with 325 00:20:31.720 --> 00:20:35.960 or hear the voice again of a loved one who's 326 00:20:36.039 --> 00:20:39.799 passed away, what about all of that? 327 00:20:41.559 --> 00:20:44.440 I mean, grief is a grief is a crazy thing, right, 328 00:20:44.480 --> 00:20:47.039 and so I think that people try to deal with 329 00:20:47.119 --> 00:20:48.640 it in a lot of different ways, some of it 330 00:20:48.640 --> 00:20:50.519 healthy and some of it not healthy. 331 00:20:50.559 --> 00:20:52.279 What I would be worried about. Here would be. 332 00:20:52.720 --> 00:20:56.680 Apps that are you know, claiming some sort of therapeutic 333 00:20:57.279 --> 00:21:00.759 effect on the heart, like misclaiming that, fault claiming that, 334 00:21:01.039 --> 00:21:04.079 or falsely claiming that you know, they're going to provide 335 00:21:04.119 --> 00:21:07.480 some sort of resolution. I think people, like I said, 336 00:21:08.559 --> 00:21:12.000 people might use these as tools to engage with maybe 337 00:21:12.039 --> 00:21:14.920 the content that somebody left behind. And I think that 338 00:21:14.960 --> 00:21:19.160 could be interesting if done in a healthy way. Right, 339 00:21:19.400 --> 00:21:23.079 if you are able to look at like air quote 340 00:21:23.200 --> 00:21:26.960 talk to you know, the all the letters that your 341 00:21:27.079 --> 00:21:30.359 your father or your grandfather left behind, that might be 342 00:21:30.400 --> 00:21:33.160 a really engaging way to learn more about their life 343 00:21:33.599 --> 00:21:36.079 and to think about like what they meant to you. 344 00:21:36.519 --> 00:21:40.920 But I don't think there are healthy ways to do that, 345 00:21:40.960 --> 00:21:42.640 and there are unhealthy ways to do that, And I 346 00:21:42.640 --> 00:21:45.599 would be worried about, you know, apps that are claiming 347 00:21:45.640 --> 00:21:49.240 to provide some sort of therapeutic effect when they're not 348 00:21:49.279 --> 00:21:53.960 doing that, rather than maybe ones that are talking about 349 00:21:53.960 --> 00:21:57.559 how they can help you understand you know, your past 350 00:21:57.599 --> 00:21:59.720 and understand your connection to your family better. 351 00:22:00.079 --> 00:22:02.039 Some of that could be very interesting and positive. 352 00:22:02.559 --> 00:22:05.400 I can't wait to get an AI letter the kinds 353 00:22:05.440 --> 00:22:08.599 of letters I get from some members of family and friends, 354 00:22:08.640 --> 00:22:13.519 particularly on the older spectrum of family and friends, the 355 00:22:13.559 --> 00:22:18.880 old merry medical Christmas letters where they tell me about 356 00:22:18.920 --> 00:22:22.359 all of their health ailments in the space of two 357 00:22:23.079 --> 00:22:24.160 very long pages. 358 00:22:25.319 --> 00:22:27.759 I suspect, I suspect, I suspect. 359 00:22:27.759 --> 00:22:31.279 If you get Christmas cards this you know, this season, 360 00:22:31.440 --> 00:22:33.400 there's there's a good chance a chunk of them were 361 00:22:34.400 --> 00:22:37.359 helped in being written by AI. But can I give 362 00:22:37.400 --> 00:22:41.759 you one example. My in laws recently came across a 363 00:22:41.799 --> 00:22:47.000 handwritten letter from you know, my father in law's dad. 364 00:22:47.799 --> 00:22:51.920 It was a very it was a terrific piece consoling 365 00:22:51.960 --> 00:22:54.839 somebody else in the loss of their their spouse. But 366 00:22:54.960 --> 00:22:57.480 some of it was actually quite difficult to read because 367 00:22:57.480 --> 00:23:01.079 it was handwritten. I used, you know, I scanned it 368 00:23:01.119 --> 00:23:04.400 into chat GPT really quickly and asked it if it 369 00:23:04.400 --> 00:23:08.880 could create a transcript of it. And that was extremely helpful. 370 00:23:08.880 --> 00:23:11.440 Actually it decrypted some of the words that we had 371 00:23:11.480 --> 00:23:13.799 we were really struggling with and uh, and we were 372 00:23:13.839 --> 00:23:16.160 able to enjoy this this letter in a way that 373 00:23:16.240 --> 00:23:18.880 I don't think we would have struggled more to figure 374 00:23:18.920 --> 00:23:21.680 out what the handwriting said had we had we not 375 00:23:21.799 --> 00:23:22.079 use that. 376 00:23:22.200 --> 00:23:26.440 So there are really positive uses to this technology. But there. 377 00:23:26.480 --> 00:23:30.799 You know, people shouldn't be substituting, you know, moving through 378 00:23:30.839 --> 00:23:33.720 their grief process by you know, pretending that they could 379 00:23:33.720 --> 00:23:36.759 still talk to their their their past relatives. 380 00:23:36.759 --> 00:23:39.200 I think, yeah, I don't I don't doubt that that 381 00:23:39.839 --> 00:23:43.799 kind of positive application and the implications of that. You know, 382 00:23:43.839 --> 00:23:48.240 you talk about, let's face it, cursive writing as a 383 00:23:48.279 --> 00:23:52.279 lost art in America anyway, But there's some beautiful cursive writing, 384 00:23:52.759 --> 00:23:55.839 uh in you know the history of family letters and 385 00:23:55.880 --> 00:23:59.839 those sorts of things. But I think about those, you know, 386 00:24:00.160 --> 00:24:04.160 as a history buff those important historical documents, you know, 387 00:24:04.240 --> 00:24:07.839 the letters from Lincoln that I can't always you know, 388 00:24:07.920 --> 00:24:11.119 I'm looking at the penmanship, and he's got pretty good penmanship, 389 00:24:11.200 --> 00:24:13.559 but there are areas where I can't find it. I 390 00:24:13.559 --> 00:24:17.440 think that's interesting. That's a very interesting application. We're going 391 00:24:17.519 --> 00:24:21.480 to talk more about some of you know, those very 392 00:24:21.680 --> 00:24:25.160 useful applications for AI coming up. Our guest today in 393 00:24:25.240 --> 00:24:29.200 this edition of The Federalist Radio Hour Neil Chilson, former 394 00:24:29.279 --> 00:24:34.039 FDC Chief technologist and currently head of AI policy with 395 00:24:34.160 --> 00:24:38.240 the Abundance Institute. I want to get to the jobs 396 00:24:38.319 --> 00:24:42.519 question though, because that is a huge concern, a growing 397 00:24:42.640 --> 00:24:46.039 concern for a lot of Americans what is fact? What 398 00:24:46.200 --> 00:24:50.920 is fiction? About what AI is doing and is about 399 00:24:51.000 --> 00:24:52.559 to do to the job market. 400 00:24:53.920 --> 00:24:58.079 Well, sorting out the facts can be challenging in sort 401 00:24:58.119 --> 00:25:00.680