1 00:00:18,399 --> 00:00:22,320 We're back with another edition of the Federalist Radio Hour. I'm Emily Jasinski, 2 00:00:22,359 --> 00:00:25,359 culture editor here at the Federalist. As always, you can email the show 3 00:00:25,359 --> 00:00:29,320 at radio at the Federalist dot com, follow us on Twitter at fdr LST. 4 00:00:29,600 --> 00:00:33,439 Make sure to subscribe wherever you download your podcasts as well. Today we're 5 00:00:33,520 --> 00:00:36,759 joined by Neil Chilson. He is a senior research fellow at the Center for 6 00:00:36,799 --> 00:00:41,240 Growth and Opportunity. He's also at the former chief technologist at the FTC, 7 00:00:41,520 --> 00:00:45,439 the author of the book Getting out of Control, Emergent Leadership in a Complex 8 00:00:45,479 --> 00:00:48,320 World, and he's got a subset as well. We're going to talk about 9 00:00:48,320 --> 00:00:50,799 that in a bit. Neil, thank you so much for joining us. 10 00:00:51,399 --> 00:00:54,840 It's great to be here. So the context here is we're talking the day 11 00:00:54,920 --> 00:01:00,320 after Sam Altman's very first congressional testimony on generative AI. Obviously, he's the 12 00:01:00,359 --> 00:01:03,159 CEO of Open AI, which is the parent company of chat GPT. So 13 00:01:03,359 --> 00:01:08,239 there's a lot of conversation very suddenly, a lot of lobbying money being spent 14 00:01:08,359 --> 00:01:14,079 very suddenly here in Washington, DC in regards to generative AI. Neil, 15 00:01:14,159 --> 00:01:17,959 you wrote a subseack post on a modest proposal. When it comes to AI, 16 00:01:18,000 --> 00:01:22,560 you're own sort of Jonathan Swift take on what could happen or what should 17 00:01:22,560 --> 00:01:26,040 happen, perhaps regular on the regulatory front when it comes to AI, first 18 00:01:26,079 --> 00:01:30,120 of all, tell us where people can find the substack post and flush out 19 00:01:30,120 --> 00:01:34,680 the argument a little bit. Sure, So the substack post is at out 20 00:01:34,719 --> 00:01:41,280 of Control dot substack dot com and it's the top post there and it will 21 00:01:41,319 --> 00:01:44,159 probably be the pin post for a little while, so check it out. 22 00:01:45,200 --> 00:01:51,480 The context for this was I was somewhat air quotes inspired by yesterday Senate hearing 23 00:01:51,480 --> 00:01:59,000 where Sam Altman and a couple others testified to the Judiciary Committee about the role 24 00:01:59,079 --> 00:02:04,920 of AI and the government in AI. And no, Sam and open AI 25 00:02:05,079 --> 00:02:09,080 have introduced artificial intelligence to millions of people for the first time, I think 26 00:02:09,080 --> 00:02:15,840 in a concrete way by opening up chat GPT last fall. They really started 27 00:02:15,879 --> 00:02:21,759 a Cambrian explosion of innovation in this space, even getting some of the bigger 28 00:02:22,199 --> 00:02:27,159 tech players who wake up and chick off the dust and try to jump into 29 00:02:27,199 --> 00:02:30,840 the AI space where they had long had investments but maybe hadn't delivered products really 30 00:02:30,919 --> 00:02:37,400 yet. So so that's great, and there's so much happening in that space, 31 00:02:38,479 --> 00:02:46,599 and it was surprising to see yesterday to have Altman and his fellow testifiers, 32 00:02:46,639 --> 00:02:51,800 as well as several of the Senators, essentially agree that there should be 33 00:02:51,960 --> 00:02:57,319 a digital regulator or some sort of regulator for AI. I don't know that 34 00:02:57,360 --> 00:03:00,800 I've ever seen in an industry that's his sort of nascent to the public mind 35 00:03:01,360 --> 00:03:05,080 go in front of Congress and immediately be like, hey, yeah, but 36 00:03:05,120 --> 00:03:09,560 also regulate us. We would really like that and actual AI agency right Like, 37 00:03:09,639 --> 00:03:14,479 that's an open AI and he's not alone. Other people from open AI 38 00:03:14,520 --> 00:03:17,759 have said we need like a regulatory agency for general AI. Yeah. I 39 00:03:17,759 --> 00:03:21,680 think that's right. And you know, it wasn't one hundred percent clear what 40 00:03:21,719 --> 00:03:23,960 he was asking for. I thought it was a bit more clear what Professor 41 00:03:24,039 --> 00:03:30,639 Marcus was asking for. It wasn't clear what the scope of what Altman was 42 00:03:30,680 --> 00:03:32,479 asking for, but it did seem like they generally agree that there should be 43 00:03:32,520 --> 00:03:37,719 some sort of regulator, and in particular, Altman mentioned a sort of licensing 44 00:03:37,840 --> 00:03:44,439 regime, which raised the hackles for many people who have experienced with occupational licensing 45 00:03:44,520 --> 00:03:49,599 and other issues that have seen the real barriers to competition that they can provide 46 00:03:49,639 --> 00:03:53,159 so so, but I don't want to. I don't want to read too 47 00:03:53,199 --> 00:03:57,240 much into what Altman was supporting there. I mean, his position maybe more 48 00:03:57,319 --> 00:04:00,719 nuanced than it came across in the hearing. But in any case, I 49 00:04:00,759 --> 00:04:06,560 was somewhat inspired by that call for a regulatory agency on AI to to think 50 00:04:06,599 --> 00:04:10,800 about, like, what was the problem that the Congress was trying to solve 51 00:04:10,919 --> 00:04:14,599 here? And there was a lot of talk about AI, but nobody really 52 00:04:14,639 --> 00:04:19,480 drilled into the specifics. Artificial intelligence is an incredibly broad topic. The only 53 00:04:19,519 --> 00:04:23,399 thing that might be as I point out in my post, you know, 54 00:04:23,439 --> 00:04:31,480 AI includes everything from health diagnostic software to autonomous vehicles, to the Dolli image 55 00:04:31,519 --> 00:04:35,519 generation and some of the other tools that we're using to even predictive text on 56 00:04:35,560 --> 00:04:41,199 your phones. I mean, they all kind of count as AI. And 57 00:04:42,199 --> 00:04:46,560 the discussion yesterday didn't really drill into what they meant on AI. And so 58 00:04:46,600 --> 00:04:50,759 it's so broad that the only thing that's probably broader than artificial intelligence is you 59 00:04:50,800 --> 00:04:55,839 know, just playing old regular intelligence. And so so I wrote a sort 60 00:04:55,879 --> 00:05:00,879 of modest proposal in the Swiftian style, calling for an agent SEE to regulate 61 00:05:00,720 --> 00:05:03,360 intelligence. I said, you know, like, let's not. If we're 62 00:05:03,360 --> 00:05:10,079 going to regulate artificial intelligence, let's not discriminate against it's it's uh, it's 63 00:05:10,319 --> 00:05:15,600 it's it's natural counterpart. Yeah, so, you know. So the argument 64 00:05:15,759 --> 00:05:21,720 was essentially a argument from the ridiculous torture ridiculous What would it look like to 65 00:05:21,720 --> 00:05:27,839 have an agency that regulated all intelligence, because that's that's about as general and 66 00:05:27,879 --> 00:05:32,000 so. Using the acronym that some of the open AI folks have have called 67 00:05:32,040 --> 00:05:36,639 for, which is oasis, I uh you know, I ran through and 68 00:05:36,959 --> 00:05:43,959 with the help of chat GPT wrote this essay arguing that if we're worried about 69 00:05:44,040 --> 00:05:47,279 artificial intelligence and the harm that it could cause, how much more should we 70 00:05:47,319 --> 00:05:51,399 be worried about human intelligence? After all, it uh, it created artificial 71 00:05:51,399 --> 00:05:57,639 intelligence and so and has brought us to the brink of disaster many times. 72 00:05:58,360 --> 00:06:01,519 Um. The main point here is that if you think it's too broad to 73 00:06:01,680 --> 00:06:05,879 regulate all intelligence, to have an agency that would that would look at intelligence 74 00:06:05,920 --> 00:06:10,240 and then try to regulate all its effects, probably too broad to have an 75 00:06:10,279 --> 00:06:14,720 agency that's just looking at artificial intelligence and regulating all its effects. Instead, 76 00:06:14,720 --> 00:06:23,120 we should be focused on the actual uses of artificial intelligence and how those impact 77 00:06:25,399 --> 00:06:29,839 various things, and so, you know, we should probably have an agency, 78 00:06:29,879 --> 00:06:33,959 we should look to an agency that's an expert in a vehicle regulation when 79 00:06:33,959 --> 00:06:41,079 we're talking about autonomous vehicles, etc. And so, and these generative ais, 80 00:06:41,120 --> 00:06:44,680 you know, they're sort of general purpose technologies, and so it's hard 81 00:06:44,680 --> 00:06:47,240 to know what specific agency might focus on them, but it might be something 82 00:06:47,319 --> 00:06:53,519 like the Federal Trade Commission, although they don't have a particularly good recent track 83 00:06:53,600 --> 00:07:00,160 record in in you know, focusing on where consumers are injured, and it 84 00:07:00,399 --> 00:07:03,199 tends to be much more of a politicize agency than it used to be. 85 00:07:03,360 --> 00:07:06,959 Anyways, I'm rambling a little bit, but that's the sort of gist of 86 00:07:08,040 --> 00:07:13,399 the quite short piece. But I hope, I hope your listeners will check 87 00:07:13,399 --> 00:07:15,639 it out. It was fun to do well. It's it's actually really interesting 88 00:07:15,680 --> 00:07:19,959 because that's the kind of chicken or egg dilemma when it comes to artificial intelligence, 89 00:07:20,079 --> 00:07:26,560 is how it can fundamentally be disconnected from human intelligence. When it's a 90 00:07:26,560 --> 00:07:30,279 product of human intelligence, it is fueled by human intelligence, and at a 91 00:07:30,319 --> 00:07:36,279 certain point short it becomes divorced from human control. And I'm saying this as 92 00:07:36,319 --> 00:07:41,399 somebody who's like fairly alarmist about AI We've done a bunch of episodes where I'm 93 00:07:41,439 --> 00:07:44,240 like, man, this stuff is happening really quickly and it's really scary. 94 00:07:44,279 --> 00:07:47,240 But we saw what happened with the restrict Act when it came to TikTok. 95 00:07:47,800 --> 00:07:55,519 Typically, when you have big companies, major multinational companies, lobbying for regulation, 96 00:07:55,759 --> 00:07:59,639 it should be setting off alarm bells because it means that they're they're trying 97 00:07:59,639 --> 00:08:03,279 to stiff competition and curry favor with the government and all of that. So 98 00:08:03,519 --> 00:08:09,519 I wonder Neil, as somebody who's actually been at the FTC, what your 99 00:08:09,560 --> 00:08:16,000 perspective is or what your confidence level is as this technology is. I think 100 00:08:16,040 --> 00:08:20,319 we all are in agreement, developing really rapidly and in some great directions and 101 00:08:20,399 --> 00:08:26,079 some frightening directions. As Altman himself conceded at the hearing on Tuesday, what's 102 00:08:26,079 --> 00:08:30,480 your confidence level in our ability to sort of keep up with it even where 103 00:08:30,600 --> 00:08:35,240 like the most like libertarian type person would look in and say, ah, 104 00:08:35,320 --> 00:08:39,120 yeah, we probably need to do something here because it's like actual fake election 105 00:08:39,200 --> 00:08:43,799 video or something like worst case scenario. What confidence level do you have right 106 00:08:43,840 --> 00:08:52,279 now? So you know, it's it's always challenging when you're looking at an 107 00:08:52,320 --> 00:08:56,840 area of new technology. I think what makes it easier in some cases is 108 00:08:56,879 --> 00:09:03,919 if your focus is on the harts that you're worried about and the harms here. 109 00:09:03,120 --> 00:09:07,000 Let's set aside this sort of existential harms and I'm not poo pooing those. 110 00:09:07,039 --> 00:09:13,559 I think there's we should think about those, for sure, But aside 111 00:09:13,600 --> 00:09:16,919 from those, the types of harms that people are worrying about, our harms 112 00:09:16,919 --> 00:09:20,600 that actually we've been talking about for a decade now. Most of them are 113 00:09:20,639 --> 00:09:26,559 the result of digital communications much more so than they are how you might generate 114 00:09:26,600 --> 00:09:31,360 them. Now, there might be a new method through which some of these 115 00:09:31,440 --> 00:09:35,720 like disinformation or bias or you know, woke AI and things like that. 116 00:09:37,039 --> 00:09:39,799 If you looked at the hearing on Tuesday, a lot of the concerns were 117 00:09:39,840 --> 00:09:43,279 basically you could have subbed out AI and just put social media in there, 118 00:09:43,279 --> 00:09:46,480 and it was like all the same. It's the same playbook that we've been 119 00:09:46,519 --> 00:09:50,639 talking about for the last you know, at least five years in Congress. 120 00:09:50,679 --> 00:09:56,000 And so so with that, I think not saying that we've solved those problems 121 00:09:56,039 --> 00:10:00,759 at all, but we do have tools in place for many of those problems. 122 00:10:00,799 --> 00:10:05,000 You know, consumer fraud. Uh, you know, UM lies to 123 00:10:05,360 --> 00:10:09,919 lies to people about like elections, election fraud. Like, we have legal 124 00:10:11,000 --> 00:10:15,480 constraints on those things. We have legal we have agencies that are charged with 125 00:10:15,679 --> 00:10:24,399 policing those Um. Are those agencies effective? I think maybe not? But 126 00:10:24,440 --> 00:10:28,279 that's that's that's sort of separate from the AI question, right, um. 127 00:10:28,559 --> 00:10:31,480 And so the question should be what does AI bring that's new? Does it? 128 00:10:31,720 --> 00:10:35,279 What does it ramp up as far as threats in this space? And 129 00:10:35,440 --> 00:10:41,360 what might we need to do um to increase the capabilities of agencies? And 130 00:10:41,399 --> 00:10:43,799 I think that's an easier question than it is to say, like do we 131 00:10:43,879 --> 00:10:48,000 have to keep on top of every development of AI. Well, no, 132 00:10:48,159 --> 00:10:50,879 maybe we just need to focus on what the effects are and focus on whether 133 00:10:50,960 --> 00:10:54,200 or not we have the tools in place to address those effects. And I 134 00:10:54,240 --> 00:11:00,559 think, um, I think the FTC traditionally has been you know, has 135 00:11:00,600 --> 00:11:03,679 had a focus that is much more on effects. Right, does this cause 136 00:11:03,720 --> 00:11:07,720 consumer injury? Does it harm competition? And if it does either of those 137 00:11:07,759 --> 00:11:13,799 things, then the FTC has the ability to bring tools in Uh. Sorry, 138 00:11:13,840 --> 00:11:18,879 my daughter's really being allowed in the background. Again, It's it's barely 139 00:11:18,879 --> 00:11:22,519 coming up in the audio, so it is okay, great, um, 140 00:11:24,679 --> 00:11:30,399 excellent. So yeah, so so that that would be my my approach. 141 00:11:30,519 --> 00:11:33,279 I mean, that's how I think about keeping up with technology. And that's 142 00:11:33,320 --> 00:11:37,960 why the FTC, despite having a very general statute, has been able to, 143 00:11:39,039 --> 00:11:43,879 you know, become the primary privacy regulator of the US because it's taken 144 00:11:43,919 --> 00:11:48,559 a case by case enforcement approach, which is more flexible than when you write 145 00:11:48,559 --> 00:11:52,960 down a bunch of rules about how a company or an industry should operate. 146 00:11:52,639 --> 00:11:58,480 That's that gets out of date very quickly. Well, and let's talk about 147 00:11:58,559 --> 00:12:03,200 the licensing proposal that we heard come from people in the industry themselves, which 148 00:12:03,240 --> 00:12:05,360 again I always see such a red flag, but it's a great example of 149 00:12:05,480 --> 00:12:09,480 also they're just sort of putting it out in the open. You know, 150 00:12:09,519 --> 00:12:13,120 you have the big guys coming in a asking for regulation and then be asking 151 00:12:13,159 --> 00:12:20,039 for licensing. Is particularly an interesting sort of cocktail right there, and what 152 00:12:20,200 --> 00:12:24,279 a recipe for you know, monopoly power and all that stuff. So could 153 00:12:24,279 --> 00:12:30,120 you talk to us a little bit about potential pitfalls of licensing because of the 154 00:12:30,519 --> 00:12:35,399 licensing regulatory approach, because I can actually see that in a new space kind 155 00:12:35,480 --> 00:12:41,720 of being explained to members of Congress senators in a way that makes sense to 156 00:12:41,759 --> 00:12:46,440 them. When something really scary comes up, they have the nightmare of social 157 00:12:46,480 --> 00:12:50,480 media as they see it, sort of hanging looming over their heads. Bloomenthal, 158 00:12:50,519 --> 00:12:56,240 I think said that directly yesterday made it very explicit, the comparison between 159 00:12:56,639 --> 00:13:00,600 you know, being too late to regulate social media because they think stole the 160 00:13:00,639 --> 00:13:05,639 twenty sixteen election blah blah blah, all this other stuff democracy. So that 161 00:13:05,720 --> 00:13:09,159 in mind, I could just see licensing being something that members of Congress a 162 00:13:09,559 --> 00:13:16,279 sort of fall for what potential pitfalls could come of that approach. Yeah, 163 00:13:16,039 --> 00:13:24,279 you know, licensing is a a just a pit of cronyism waiting to happen. 164 00:13:24,399 --> 00:13:30,279 Right. For the most part, licensing regimes historically have been one of 165 00:13:30,320 --> 00:13:37,639 the best ways for incumbents at the top of the game to stay there without 166 00:13:37,720 --> 00:13:41,039 having to compete in the marketplace, if they can raise the barriers to entry 167 00:13:41,039 --> 00:13:45,120 in a way that aren't just you know, economic. We talk a lot 168 00:13:45,120 --> 00:13:46,559 about this, and you know, Capitol Hill it was. It was a 169 00:13:46,559 --> 00:13:52,039 bit strange because Capitol Hill has been super worried about the ability of some of 170 00:13:52,039 --> 00:13:58,320 the large tech companies to raise barriers to entry to competition. The best way 171 00:13:58,360 --> 00:14:01,600 to raise a barrier to competition is to make it illegal to compete unless you 172 00:14:01,679 --> 00:14:07,639 pass you know, some set of certifications and and uh and so I think 173 00:14:07,720 --> 00:14:16,919 that the licensing regimes have every risk of doing becoming a having an anti competitive 174 00:14:16,960 --> 00:14:20,159 effect. And in a space that is as you know, you have so 175 00:14:20,200 --> 00:14:24,759 many people jumping into this space in a way that I think Congress finds concerning 176 00:14:24,879 --> 00:14:28,200 in many ways. Right, Like, we can't control this industry because there's 177 00:14:28,200 --> 00:14:31,960 so many people. Anybody with a computer and a you know, a nice 178 00:14:31,440 --> 00:14:37,200 chunky graphics processing unit can build one of these models. Now, and like, 179 00:14:37,240 --> 00:14:41,399 what do we do about that? Um? You know, that's that's 180 00:14:41,399 --> 00:14:45,720 a vibrate ecosystem of competition. And an easy way to shut that down would 181 00:14:45,720 --> 00:14:50,559 be a licensing regime and and constrain that to you know, just the big 182 00:14:50,600 --> 00:14:52,679 players you can afford to comply. And as you point it out earlier, 183 00:14:52,720 --> 00:14:56,960 this is a playbook that's you know, it's surprising that it's happening this early 184 00:14:58,120 --> 00:15:03,000 in this particular base, but it is a playbook that is pretty common even 185 00:15:03,519 --> 00:15:07,279 and well known by some of the big players. You know, Microsoft is 186 00:15:07,279 --> 00:15:13,440 a big backer of open Ai and Microsoft has played this game in the past, 187 00:15:13,679 --> 00:15:20,279 and so I think they are quite aware of the potential of regulatory approaches 188 00:15:20,320 --> 00:15:26,759 to raise barriers to entry. I'm not I don't know for a fact that 189 00:15:26,799 --> 00:15:30,480 Sam Altman has that this is what he has in mind. If you look 190 00:15:30,519 --> 00:15:35,440 at the sort of charter of open Ai, it has always contemplated as a 191 00:15:35,559 --> 00:15:43,799 sort of cooperative frame with government in this space, and so he may not 192 00:15:43,879 --> 00:15:48,919 be thinking about this particular effect as the motivation for a licensing regime. But 193 00:15:50,120 --> 00:15:54,000 I do think that history tells us that that is the likely effect. That 194 00:15:54,080 --> 00:15:58,559 a licensing regime will be one that raises barriers to entry, that makes the 195 00:15:58,600 --> 00:16:06,759 space less competitive, and that preferences those who are already have, who are 196 00:16:06,759 --> 00:16:11,360 already at the top of the pile. And that's a real concern for you 197 00:16:11,399 --> 00:16:15,559 know, somebody like me who likes a dynamic, innovative space, even if 198 00:16:15,600 --> 00:16:22,840 it feels maybe like it's a little bit out of control. The Watchdolt on 199 00:16:22,879 --> 00:16:27,120 Wall Street podcast with Chris Markowski every day. Chris helps unpack the connection between 200 00:16:27,120 --> 00:16:30,399 politics and the economy and how it affects your wallet. If someone told you 201 00:16:30,440 --> 00:16:34,440 the recession started today, would you change anything about your daily life. Has 202 00:16:34,440 --> 00:16:38,039 the big recession monster already arrived, don't be fooled by what you read. 203 00:16:38,159 --> 00:16:42,200 Docs tend to outperform during a recession anyways. Make sure you're getting the most 204 00:16:42,200 --> 00:16:45,480 trustful information, whether it's happening in DC or down on Wall Street, it's 205 00:16:45,480 --> 00:16:48,759 affecting you financially. Be informed. Check out the Watchdot on Wall Street podcast 206 00:16:48,840 --> 00:16:52,000 with Chris Markowski on Apple, Spotify, or wherever you get your podcast. 207 00:16:55,440 --> 00:16:57,879 You're probably one of the best examples of this from very recent histories, Like 208 00:16:57,919 --> 00:17:02,679 a couple of years ago, Facebook now Meta started taking out ads in the 209 00:17:02,720 --> 00:17:07,720 Beltway morning newsletters asking for section two thirty because they know they're the only ones 210 00:17:07,799 --> 00:17:12,039 that can sort of absorb the blows in that space that will cover Sextra two 211 00:17:12,079 --> 00:17:15,759 thirty reform. But that's where it's sort of frustrating, as someone who's like 212 00:17:15,799 --> 00:17:21,480 candidly coming at this from a fairly alarmist perspective on how quickly generative AIM is 213 00:17:21,480 --> 00:17:26,039 moving, is that you know, the sort of libertarian folks on the right 214 00:17:26,079 --> 00:17:30,680 are not, i think, downplaying the potential dangers of AI. And then 215 00:17:30,759 --> 00:17:34,759 at the same time, it's just hard to look at what a potential solution 216 00:17:34,799 --> 00:17:41,960 to those those dangers are if there's basically a smothering of conversation about potential regulation. 217 00:17:41,960 --> 00:17:45,599 I'm not talking about you in particular, but it just gets really hard 218 00:17:45,720 --> 00:17:49,960 to have the conversation about envisioning that when you know it's saying, well, 219 00:17:51,039 --> 00:17:53,640 the regulatory pitfalls are very real, and then everyone else is saying, well, 220 00:17:53,680 --> 00:17:56,799 the dangers of AAR are really real. So I wonder, Neil, 221 00:17:56,839 --> 00:18:00,720 what you think the best way to kind of and you know, this is 222 00:18:00,720 --> 00:18:03,319 probably something that you've you've dealt with in the social media space too, what 223 00:18:03,440 --> 00:18:08,359 the best way to kind of balance those concerns are It's almost impossible in this 224 00:18:08,480 --> 00:18:12,319 media climate and this like partisan climate, But what is the best way for 225 00:18:12,359 --> 00:18:18,480 people to start thinking about this that you avoid hitting that impass well? You 226 00:18:18,519 --> 00:18:22,440 know, I think in some ways and in some ways, the AI discussion 227 00:18:22,519 --> 00:18:26,680 might be a little bit easier than the social media discussion in that the social 228 00:18:26,680 --> 00:18:33,839 media discussion is entirely around speech issues almost entirely around speech right, whereas AI, 229 00:18:34,079 --> 00:18:37,119 like I said earlier, like there's a bunch of areas in which the 230 00:18:37,160 --> 00:18:44,039 harms could be like physical right, like if you have AI medical devices or 231 00:18:44,079 --> 00:18:48,079 if you have AI driven vehicles, you're talking about spaces that could actually, 232 00:18:48,640 --> 00:18:52,400 if they go wrong, they could kill people, and we could talk about 233 00:18:52,440 --> 00:18:56,160 like what the what our risk tolerance should be in spaces like that. You 234 00:18:56,200 --> 00:19:03,680 know, lots of very powerful, useful drugs also can kill people, and 235 00:19:03,759 --> 00:19:10,279 so we need to talk about like how how as a society or as individuals, 236 00:19:10,599 --> 00:19:15,799 what is our risk tolerance for that type of benefit costs, you know, 237 00:19:15,839 --> 00:19:19,640 trade off. But but that's that's easier in some ways than talking about 238 00:19:19,759 --> 00:19:27,440 speech issues, where we have a strong constitutional right plus and limit on what 239 00:19:27,559 --> 00:19:33,599 government can do. Plus um, the harms are much less tangible, although 240 00:19:33,720 --> 00:19:37,599 they can be they can also they can feel severe, but they are often 241 00:19:37,680 --> 00:19:42,960 less tangible when you're talking about speech and so so I think there's a better 242 00:19:44,160 --> 00:19:48,240 I guess what I'm saying. I think there's a better justification for in specific 243 00:19:48,400 --> 00:19:53,640 areas of and applications of AI, having a regulatory regime that looks similar to 244 00:19:53,720 --> 00:19:57,240 like what we might have around drugs, or what we might have around medical 245 00:19:57,279 --> 00:20:03,039 devices, or what we might have around you know, automobiles, and so 246 00:20:04,920 --> 00:20:08,039 to the extent that AI is being applied in those areas, I think we 247 00:20:08,200 --> 00:20:14,559 have frameworks for thinking about the risks balanced trade off. And you know, 248 00:20:15,359 --> 00:20:18,400 again my priors are that in a bunch of those spaces we probably are two 249 00:20:18,480 --> 00:20:22,720 risk averse um, and that we should allow more experimentation. But at least 250 00:20:22,720 --> 00:20:26,079 we have you know, we're talking about the you know, where we might 251 00:20:26,160 --> 00:20:30,559 draw the line on experimentation, you know, around automated vehicles or you know, 252 00:20:30,839 --> 00:20:37,119 medical devices. So I guess my big picture is we should disaggregate. 253 00:20:37,119 --> 00:20:40,519 We should stop talking about AI generally and try to find the one size fits 254 00:20:40,559 --> 00:20:44,559 all solution for all different applications of AI, and we should focus on what 255 00:20:44,599 --> 00:20:48,480 are the specific AI uses that we are worried about, what are the concerns 256 00:20:48,519 --> 00:20:52,000 that are there, What tools do we already have in the toolkit to address 257 00:20:52,039 --> 00:20:56,240 some of those, and if we need to supplement those That's a good idea, 258 00:20:56,400 --> 00:20:59,799 right we should. That's that's where we should be focusing our attention rather 259 00:20:59,839 --> 00:21:03,400 than and trying to like solve the whole big ball of wax of like artificial 260 00:21:03,440 --> 00:21:07,240 intelligence, like how do we make it safe? Right? Um, we 261 00:21:07,279 --> 00:21:11,799 should focus on it. How do we make AI in vehicles safe? How 262 00:21:11,799 --> 00:21:15,000 do we make AI and medical devices safe? How do we make you know, 263 00:21:15,079 --> 00:21:25,319 content generation AI, you know, work the way that people expect it 264 00:21:25,359 --> 00:21:29,200 to work and that the companies make when they make promises about how it works, 265 00:21:29,200 --> 00:21:30,599 it works that way, right, like things like that. No, 266 00:21:30,720 --> 00:21:34,599 this is a really important point and it's um you know, it's one that 267 00:21:34,799 --> 00:21:38,279 I've directed a lot of folks on the sort of new right many different times, 268 00:21:38,319 --> 00:21:41,839 because in so many cases it's exactly what you're describing. We actually already 269 00:21:41,839 --> 00:21:45,200 have laws, and we might not have an agency, but we do have 270 00:21:45,279 --> 00:21:51,039 copyright laws. We do have medical malpractice laws. We do have laws, 271 00:21:51,200 --> 00:21:53,640 you know, just like we have so many, so very many laws, 272 00:21:53,759 --> 00:21:59,119 so many laws. We have all of the laws that can can be applied 273 00:21:59,160 --> 00:22:03,240 in different circumstances. Like it's just a matter of making the lawsuits. It's 274 00:22:03,279 --> 00:22:07,480 just a matter of you know, actually like being proactive and finding places where 275 00:22:07,440 --> 00:22:12,200 there are elapses on that note, those places where we should be sort of 276 00:22:12,400 --> 00:22:18,200 looking to make sure that that AI is being applied safely. I know that 277 00:22:18,200 --> 00:22:22,440 you've talked about election issues. Josh Holly had a confrontation with Sam Altman a 278 00:22:22,480 --> 00:22:26,079 little bit about that, and Altman himself obviously conceded, Yep, I'm pretty 279 00:22:26,079 --> 00:22:32,519 worried about that. What spaces do you think the energy should be focused on 280 00:22:32,720 --> 00:22:38,000 right now when it comes to conceiving safe uses and applications of AI. Well, 281 00:22:38,000 --> 00:22:42,680 on that discussion, I did think like the it was. Altman's response 282 00:22:42,799 --> 00:22:48,319 was pretty careful. As I recall, he said, yes, it's true 283 00:22:48,640 --> 00:22:55,480 that you may be able to generate misinformation on these platforms, but that the 284 00:22:55,519 --> 00:22:59,079 harm all comes from when you distribute it on social media. So he was 285 00:22:59,480 --> 00:23:02,359 sort of like pointing at the previous guys. He's like, it's not us 286 00:23:02,359 --> 00:23:04,000 so much, um, And I think, yeah, something of a point 287 00:23:04,680 --> 00:23:11,279 um there that you could do something new at scale with AI. That's convincing. 288 00:23:11,880 --> 00:23:19,240 And I do think that there are some concerns around election manipulation, but 289 00:23:19,319 --> 00:23:25,039 I think that whole, that whole narrative has been way overplayed. I mean, 290 00:23:25,079 --> 00:23:30,920 it's if you listen a little bit, it's like people are worried that 291 00:23:30,920 --> 00:23:33,960 that other people are dumber than them, right, and that, and that 292 00:23:34,079 --> 00:23:41,559 the average voter is going to somehow get tricked into voting a different way than 293 00:23:41,599 --> 00:23:45,720 they would have if they just get the right combo of targeted advertisements, and 294 00:23:47,799 --> 00:23:52,359 um, you know, if that's what you think about your average human citizen 295 00:23:52,559 --> 00:23:56,319 or your average US citizen. Then um U, that's that's not a very 296 00:23:56,400 --> 00:24:03,440 positive belief about the ability of democracy to make educated decisions. So so the 297 00:24:03,559 --> 00:24:07,880 animating belief for many like journalists and technocrats like that is the one thing that 298 00:24:07,960 --> 00:24:12,880 they now think needs to translate into new regulatory power and censorship and etc. 299 00:24:14,480 --> 00:24:17,400 Well and and and it's and it's crazy to me, Like it's crazy to 300 00:24:17,400 --> 00:24:21,599 me to think that, Oh you think that, Um you know that the 301 00:24:21,640 --> 00:24:26,240 New York Times consistently having its frame on issues is is not going to change 302 00:24:26,279 --> 00:24:30,880 people's minds. But you know, some targeted ads generated by chat GPT, 303 00:24:30,079 --> 00:24:37,559 will I just don't understand that that sort of belief in the good ability to 304 00:24:37,599 --> 00:24:41,519 shape minds. But then like a fear of like a bad ability to shape 305 00:24:41,559 --> 00:24:45,839 minds. Um. But I do think that there are some are there are 306 00:24:45,880 --> 00:24:48,839 some areas where I, you know, I do worry about the effect of 307 00:24:49,119 --> 00:24:56,039 generative models ability to scale up bad behavior. And in particular, I think 308 00:24:56,079 --> 00:25:03,720 cybersecurity is a big one. I mean phishing emails right now. You know, 309 00:25:03,920 --> 00:25:08,400 I think we've all we all have an ability to spot the you know, 310 00:25:08,480 --> 00:25:14,480 the send a thousand dollars to me if you would like, and I'll 311 00:25:14,519 --> 00:25:18,359 send you a million dollars in gold bullying or something like that. Emails that 312 00:25:18,400 --> 00:25:26,000 we get, but when those get much more targeted and realistic looking to a 313 00:25:26,119 --> 00:25:33,240 particular individual, that that that is a problem that I think cybersecurity professionals are 314 00:25:33,279 --> 00:25:40,119 concerned about. Um, you don't solve those necessarily by shutting down generative models. 315 00:25:40,160 --> 00:25:42,440 But the way these generative models all work, they have a layer, 316 00:25:42,599 --> 00:25:48,720 like a safety layer above them that can spot some of these patterns pretty quickly, 317 00:25:48,759 --> 00:25:52,559 especially when they're being generated on mass on mass like they would have to 318 00:25:52,599 --> 00:25:59,000 be in order to you know, to make a difference in the cybersecurity space. 319 00:26:00,920 --> 00:26:03,119 Plus, you know, we have lots of other you know, two 320 00:26:03,200 --> 00:26:07,480 factor authentication and things that we probably should all just be a little bit more 321 00:26:07,519 --> 00:26:11,799 aware of what's coming into our inbox. I do think that in this there 322 00:26:11,799 --> 00:26:18,240 will be a little sort of gap here when between when people learn to adapt 323 00:26:18,240 --> 00:26:22,000 and when the companies learned to adapt to this, and when the you know, 324 00:26:22,039 --> 00:26:26,440 the cyber criminals start using these tools. So so I am a little 325 00:26:26,440 --> 00:26:30,160 worried about that space. I do think it's worth at tension both from us 326 00:26:30,200 --> 00:26:36,200 as users of email and lots of other technology, but also from the companies, 327 00:26:36,200 --> 00:26:40,960 and I think a lot of the generative AI companies are are are quite 328 00:26:40,960 --> 00:26:45,480 concerned about that particular use case as well. And so for generative AI, 329 00:26:45,640 --> 00:26:49,039 that's that's one that I'm particularly worried about. There's lots of other spaces, 330 00:26:49,039 --> 00:26:53,000 like I've already mentioned, the ones that can actually result in the loss of 331 00:26:53,079 --> 00:27:00,880 human life in a much more direct way where I think AI is. Yeah, 332 00:27:00,640 --> 00:27:03,000 we should, we should pay a lot of attention to those those uses, 333 00:27:03,759 --> 00:27:07,279 but those aren't really the ones that are are raised by generative AI at 334 00:27:07,279 --> 00:27:15,519 this point right now, there are pretty serious geopolitical implications for the regulatory question. 335 00:27:15,880 --> 00:27:19,960 Um you know, China I think recently just unleashed more regulation of its 336 00:27:19,960 --> 00:27:27,799 own um generative AI industry and it's development of that industry, and so obviously 337 00:27:29,279 --> 00:27:33,200 there are I agree with you, I think cybersecurity is the space to watch 338 00:27:33,240 --> 00:27:38,279 for here, but there are questions about stifling um, you know, private 339 00:27:38,319 --> 00:27:42,960 innovation in a way that puts us behind other countries and leaves us more vulnerable, 340 00:27:44,000 --> 00:27:45,960 vulnerable in a national security sense. Do you have thoughts on that, 341 00:27:47,079 --> 00:27:52,799 Neil, because that's I haven't really heard quite enough conversation on in that area. 342 00:27:52,200 --> 00:27:56,920 Um, although it's it's obviously a major concern, um knowing how much 343 00:27:56,960 --> 00:28:00,880 China has invested into generative AI. So if there were let's say, a 344 00:28:00,880 --> 00:28:07,359 licensing regime to step in that would step into the picture or other, you 345 00:28:07,440 --> 00:28:11,359 know, an agency that starts to regulate generative AI. We don't know exactive 346 00:28:11,400 --> 00:28:14,920 the contours of that would look like. But um, what are the risks 347 00:28:15,319 --> 00:28:21,680 in terms of like our geopolitical standing in terms of competition? Yeah, so, 348 00:28:21,960 --> 00:28:25,400 you know, I wish the hearing on Tuesday had talked a lot more 349 00:28:25,440 --> 00:28:30,240 about this. I didn't even hear a mention of China in the discussion. 350 00:28:30,720 --> 00:28:38,319 The US right now has a enormous advantage in this space. We are set 351 00:28:38,440 --> 00:28:47,240 up in the way that we treat software innovation to focus on the effects at 352 00:28:47,240 --> 00:28:51,480 the end, right, and not on a licensing regime up front. This 353 00:28:51,599 --> 00:28:55,880 is very different than how Europe is planning to do it. They have a 354 00:28:55,960 --> 00:29:03,039 law that's pretty close to being in effect around AI specifically, and then they 355 00:29:03,079 --> 00:29:07,359 have several other laws that already make it very difficult to do some of this 356 00:29:07,519 --> 00:29:11,519 work. And then China, obviously, UM, you know, I've heard 357 00:29:11,559 --> 00:29:18,599 people call it thoughtful regulation, but its goal is not the same as we 358 00:29:18,680 --> 00:29:27,079 would have here. The goal of Chinese AI regulation is to protect the communist 359 00:29:27,160 --> 00:29:34,160 party there and to ensure that you know, the populist does not get out 360 00:29:34,160 --> 00:29:38,400 of hand with these types of models, and that they these they're not too 361 00:29:38,519 --> 00:29:42,559 they're not powerful tools for you know, self expression in a way that would 362 00:29:42,599 --> 00:29:47,599 threaten the leadership in that country. So all of that to say, we 363 00:29:47,680 --> 00:29:52,319 have a we right now have a advantage in our regulatory approach, and that's 364 00:29:52,359 --> 00:29:56,200 shown by the fact that these companies and the innovation is largely happening here. 365 00:29:56,279 --> 00:30:03,480 It would be a huge detriment to us tom seed that ground to China or 366 00:30:03,519 --> 00:30:08,240 even to like copy Europe and and and freeze our regulation in a way or 367 00:30:08,240 --> 00:30:18,359 freeze our innovation in a way that would mean that China's applications would would lap 368 00:30:18,440 --> 00:30:22,039 us. And so I do worry about the lot. There's there's no reason 369 00:30:22,119 --> 00:30:25,680 for us to do that. We have, like he said, we have 370 00:30:25,799 --> 00:30:30,319 the tools to deal with at least the concerns that we have now. We 371 00:30:30,400 --> 00:30:34,680 also have the ability to adapt to that. You know, I keep going 372 00:30:34,720 --> 00:30:38,960 back to the idea that the US is the land of you know, pioneers 373 00:30:40,000 --> 00:30:45,319 and frontiersmen. We are the people who go out and forge into the unknown, 374 00:30:45,519 --> 00:30:51,279 and we deal with the problems as they come up. If we sat 375 00:30:51,359 --> 00:30:55,480 at home and try to imagine all the things that could go wrong and right 376 00:30:55,599 --> 00:30:59,759 the right the rules that keep them from happening, we would never leave the 377 00:30:59,759 --> 00:31:03,839 house. And that's sort of the europe model, um and so so we 378 00:31:03,839 --> 00:31:07,400 should lean into that. We should lean into our strength in dealing with problems 379 00:31:07,440 --> 00:31:11,400 as they come up, but innovating along the way, because because that's gotten 380 00:31:11,519 --> 00:31:14,960 us, you know, the the innovation that we have now. And I 381 00:31:14,960 --> 00:31:18,839 think I think we should we should revel in the fact that the US is 382 00:31:18,319 --> 00:31:22,920 in the lead in this space and we should take advantage of it. It's 383 00:31:23,880 --> 00:31:27,200 it's possibly the only way, and this is a little bit off topic, 384 00:31:27,279 --> 00:31:32,920 maybe it's possibly the only way we're going to outgrow the massive federal debt crisis 385 00:31:32,920 --> 00:31:40,359 that we're facing. So um uh is to you know, take advantage of 386 00:31:40,359 --> 00:31:45,720 our ability to produce good products and services that make people's lives better and that 387 00:31:45,759 --> 00:31:51,839 boost GDP through AI until we then tax all of that revenue and redistribute it 388 00:31:51,880 --> 00:31:56,000 into inefficient government programs. You know, is there is there anything else that 389 00:31:56,039 --> 00:32:00,480 we haven't touched on you think is an important um point or something. This 390 00:32:00,759 --> 00:32:06,119 is pretty critical for people to understand about generative AI or AI more broadly. 391 00:32:07,480 --> 00:32:09,319 There's there's two things I might add in here, and they're they're sort of 392 00:32:09,319 --> 00:32:15,319 related to each other. Um what's really interesting about generative AI is that, 393 00:32:16,440 --> 00:32:22,599 you know, I think it prevents it presents some great opportunities. It has 394 00:32:22,640 --> 00:32:28,559 some threats to interesting groups of people at this point, right, content creators 395 00:32:28,599 --> 00:32:31,640 I think are very nervous. The news media, I think is very nervous 396 00:32:31,680 --> 00:32:37,799 about this technology because it works on you know, their tools of the trade 397 00:32:37,880 --> 00:32:45,839 words and video and audio and images, and I think there's there's a huge 398 00:32:46,000 --> 00:32:50,880 amount of productivity and creativity that's going to be coming out from these products. 399 00:32:51,119 --> 00:32:58,279 But it's going to be very disruptive to the elites who kind of consider themselves 400 00:32:58,319 --> 00:33:05,440 the tastemakers or the agenda centers of the world. And so we saw this 401 00:33:05,519 --> 00:33:07,839 before. Our social media was in a similar way. It was a big 402 00:33:07,880 --> 00:33:14,599 threat at least to the advertising model of some of the traditional media, and 403 00:33:15,680 --> 00:33:20,920 no surprise, traditional media was not a big fan of many of these these 404 00:33:20,960 --> 00:33:27,079 platforms. I think AI is even more personally a sort of threat to maybe 405 00:33:27,079 --> 00:33:30,640 traditional models of journalism, and so I think you can expect to see some 406 00:33:30,279 --> 00:33:37,279 coverage that would come from that frame. I don't think that so many of 407 00:33:37,279 --> 00:33:45,079 these people were often concerned when automation threatened blue collar jobs, but suddenly, 408 00:33:45,480 --> 00:33:49,079 yeah, I think they're concerned now. So it'll be interesting to see how 409 00:33:49,079 --> 00:33:52,519 that plays out. And then the second point I think is around I do 410 00:33:52,599 --> 00:33:58,559 think the AI is sort of the next front in the culture war that we've 411 00:33:58,680 --> 00:34:01,039 been ongoing. I already said, you know, Congress finally figured out like 412 00:34:01,039 --> 00:34:04,759 how to talk about social media in a way it kind of understands it, 413 00:34:05,079 --> 00:34:07,559 and it understands the sort of political I say, figured out that they figured 414 00:34:07,599 --> 00:34:10,400 out how to score political points by talking about social media, right, and 415 00:34:10,440 --> 00:34:15,639 so on both sides. And so I think I think, like like all 416 00:34:15,719 --> 00:34:21,239 things, AI is sort of going to be the next front in this culture 417 00:34:21,280 --> 00:34:23,280 war. It's going to be people arguing about, like, hey, this 418 00:34:23,440 --> 00:34:28,760 misinformation is you know, and we saw some of that in the hearing already. 419 00:34:28,760 --> 00:34:34,119 This misinformation is going to shape elections or it's going to be woke and 420 00:34:34,159 --> 00:34:40,679 it's gonna you know, suppress conservative ideas, and you know, I think 421 00:34:40,679 --> 00:34:45,119 we should dig into those issues. We shouldn't just let them go. But 422 00:34:45,199 --> 00:34:49,920 also I think we should see them for the sort of next step in what 423 00:34:50,079 --> 00:34:54,840 is a continuing back and forth between the political parties, and that in many 424 00:34:54,840 --> 00:34:59,559 ways it'll be less about the policy at the end of the day and a 425 00:34:59,559 --> 00:35:01,719 little bit more about like who can score political points about this stuff. So, 426 00:35:02,440 --> 00:35:06,639 um, those are just the two things that kind of jump out to 427 00:35:06,719 --> 00:35:12,400 me as far as political dynamics in in DC. It'll be interesting to see 428 00:35:12,440 --> 00:35:15,280 what coverage looks like, how it might be biased, and and you know, 429 00:35:15,599 --> 00:35:19,519 who's going to be the who's going to be able to best and make 430 00:35:19,559 --> 00:35:27,360 the political points from from accusing various AI platforms of shifting the narrative or distorting 431 00:35:27,400 --> 00:35:30,719 the narrative. So, m no, it's more of the same, right 432 00:35:30,840 --> 00:35:35,440 no. And it will be interesting because if you're also sort of in the 433 00:35:35,480 --> 00:35:40,519 space where you're helping to prevent any possible competitors that might be you know, 434 00:35:42,079 --> 00:35:45,519 for just one hypothetical like what Elon Musk is trying to do to Twitter um 435 00:35:45,679 --> 00:35:51,599 or oral rumble like the rumble to the AI YouTube, then you might not 436 00:35:51,639 --> 00:35:55,960 have much of a leg to stand on right right exactly, Neil, go 437 00:35:57,039 --> 00:35:59,880 ahead, go ahead, No, I was just gonna say. Also, 438 00:36:00,119 --> 00:36:01,199 the other thing that jumps out to me is that, you know, the 439 00:36:01,360 --> 00:36:09,679 Biden administration has a lot of different documents that they've come out with around aim 440 00:36:10,039 --> 00:36:15,480 they're calling for accountability, but they never really say to whom. I think 441 00:36:15,000 --> 00:36:22,599 they basically mean accountable to the Biden administration, and so so we should think 442 00:36:22,599 --> 00:36:25,039 really hard about like what that might look like. I think even they should 443 00:36:25,039 --> 00:36:30,000 look I think pretty hard about what it might look like to have AI accountable 444 00:36:30,039 --> 00:36:36,400 to the government agencies when those agencies are in the hands of people that they 445 00:36:36,400 --> 00:36:39,960 don't agree with. So a word of caution, that's a great point. 446 00:36:40,400 --> 00:36:45,519 I'm not sure they have thought about that. Neil Chilson is a Senior Research 447 00:36:45,559 --> 00:36:50,199 Fellow at the Center for Growth and Opportunity obviously also a former de former Chief 448 00:36:50,199 --> 00:36:52,880 Technologists add the FTC and he writes on substack, Neil, where can people 449 00:36:52,920 --> 00:36:59,719 find your substack? So my substack is out of control dot substack dot com 450 00:37:00,039 --> 00:37:06,960 and I am on Twitter at Neil underscored Chilson, and I hope people reach 451 00:37:07,000 --> 00:37:08,639 out. This is a great topic and thanks so much for having me on. 452 00:37:09,000 --> 00:37:13,079 Of course, the book, as we mentioned earlier, is to go 453 00:37:13,119 --> 00:37:15,199 along with the theme of the sub Stack. Is God getting out of control? 454 00:37:15,239 --> 00:37:20,559 Emergent leadership in a complex world. I'm em Elijah Shenski, Culture editor 455 00:37:20,559 --> 00:37:23,639 here at The Federalist. We will be back with more Federalist radio hours soon. 456 00:37:23,840 --> 00:37:35,360 Until that be lovers of freedom and anxious for the fray. I heard 457 00:37:35,440 --> 00:37:45,239 the fame for the reason, and then it faded away.