Anthropic, Albany, and the AI Backlash

0:00Welcome to the Kato Podcast. I'm Ryan Bourne, Kato's RF and Shaft Chair for the Public Understanding of Economics. There's a first debate at the moment in economic circles at least about how much artificial intelligence is already affecting the job market and workers productivity. There's been a bunch of microeconomic studies of individual firms, industries that seems to show big effects already. Some customer support firms have seen 14% improvements in output per worker after rolling out new AI technologies, consultancies, some PR firms have seen similar gains. But it's not clear yet whether this is passing through to overall economic performance. A lot of the macroeconomic big picture numbers seem difficult to reconcile with the idea we're seeing an AI-fueled productivity takeoff just yet. Even though labor productivity growth, that's the amount that we produce per worker, was pretty healthy last year for the past

1:0312 months, 2.8%, that stronger growth trend actually started in 2022 before many of the generative AI technologies really took off. And looking across industries, it's not clear how far AI is responsible for this labor productivity pick-up anyway. Now perhaps there's another way for us to measure how impactful this new technology will be, and a more simple one. How often it plays a central role in current affairs stories. And I'd humbly suggest that events this past week show that AI is really starting to have a big impact. The federal government of course is now engaged in a legal dispute with one firm, anthropic, over the uses of AI for military purposes. One proposed bill in the New York legislature seemingly wants to disable chat bots from given advice in a range of licensed industries. A new polling this week by NBC suggests that AI is deeply unpopular among the population, with a minus 20% net favorability rating, for context only

2:03two entities pulled worse in that survey. They were the country of Iran, which the US is currently attacking, and the Democratic Party. So to discuss all this, I'm delighted to be joined by Jennifer Harleston, Kato Senior Fellow in Technology Policy. So Jennifer, welcome. Hi Ryan, great to have a conversation with you. So Jen, let's start by talking about the dispute between the government and anthropic. Let's put aside for a second our position and concerns about where this is going. Could you just describe to us in as simple terms as possible what this dispute is actually about? So I think at the heart of it, this dispute is understood of should the government be able to force a private company to change the terms of its product? Anthropic had a contract with the US government providing AI services, and it refused to remove certain safety features that it felt were necessary to ensure that its technology weren't

3:05used for government surveillance or for lethal autonomous weapons. And what we saw was that because it refused to remove these safeguards, the Pentagon not only canceled the contract, but initially they were threatening either to invoke the Defense Production Act, which could have effectively nationalized the company and forced it to make these changes. Or what we have since seen is that they declared it a supply chain risk, which effectively blacklist the company from doing business with the Department of War, as well as with its contractors. And since then what we've seen is that Anthropic has filed litigation about this classification, particularly on first amendment grants. So let's unpack some of that. So as you say, Anthropic was willing to provide their technology to the government, but wanted to maintain those assurances that the government that wouldn't be used for mass, I think they called mass domestic surveillance or fully autonomous weapons, as you said. So

4:08why did Anthropic want those safeguards and restrictions? Why was that so important to them? I think this goes to what they believe is necessary to ensure civil rights and civil liberties are protected. While it's a very exciting time for the development of AI in a lot of sectors and there are certainly beneficial uses of AI in the government, including in the Pentagon. One can imagine how, for example, an AI note taker in a classified meeting can actually provide greater national security by preventing one less person who could potentially leak state secrets from having to be in that meeting. However, because of concerns about how this could potentially be abused by the government and the lack of clarity around access to the private data that the company might have or the way that these tools could be deployed before they feel they're technically technologically ready for certain decisions where there still is an inaccuracy when it comes to something that could be a potentially lethal decision or a

5:10decision to declare war. Anthropic had drawn these red lines. The contract states for all lawful purposes and so you may have also seen in the news that open AI has now entered a contract with the Pentagon. A lot of the question is going to depend on how one interprets that for all lawful purposes. Some would say well this kind of mass surveillance and autonomous weapons are naturally unlawful. Others are concerned that in the lack of clear guidance when it comes to AI policy that there's a legitimate concern that the government could overreach and try and use a technology for dangerous purposes or for purposes that could violate individual civil rights and civil liberties. So I'm sensing from some of the terms used. There's some of the words used like dangerous or whatever that you're very skeptical of the federal government's position. So let's try and unpack this. How libertarian should think about this dispute in isolation before we get to the government's response. So I think we

6:14all recognize as libertarians that national defense is a legitimate role of government and the government is saying that these capabilities are vital to that objective. Now we can agree or disagree on that. But I want to try and get out. Would you always say that the company has no right, the company, would you always say that the company has a right to insist on these safeguards? What I'm getting at is is your objection to the government's position here because you don't like the specific uses and thropic was trying to prevent or because you think and thropic should just have the right to pursue whatever conditions it wants and maintain those agreed conditions with the government in negotiation. That's a good question, Ryan. And I think what's really important for libertarians is to understand why this is more than just a typical contract dispute. If the Pentagon had decided that because of these safeguards and thropic no longer fulfills their purposes and canceled the contract, there could be a continued debate about what the appropriate red lines for an AI company to draw as a private

7:17company are. This could have been a kind of typical contract dispute type of scenario. What really makes this situation different is the retaliation. The labeling and thropic as a supply chain threat for its choice to engage in expressive activity is decision on how to train the model and where to set the safeguards is a form of expressive activity the same way that a newspaper deciding what articles to put in its particular paper is a form of expressive activity. And that's what makes this situation so unique and so concerning not only for what it might mean for anthropic, but what it might mean for the future when it comes to the government's ability to try and intervene in decisions by private companies or otherwise have them face what could be a retaliatory measure when it comes to their expressive rights. Yeah, and I think this is a pattern that we've unfortunately seen with this administration using various statutes and economic tools to exact retaliation and companies doing things that they dislike. So once these, well, not negotiations, but once this, once this

8:22agreement kind of broke down, as you say, the conflict escalated and anthropic said that the government threatened to remove it from military systems. And then, of course, it has since designated it as a supply chain risks. So I mean, for listeners unfamiliar with that area of law, can you explain why that designation is just so significant to the company? Well, I think in some ways, one of the reasons it's so significant first is to think about who's not designated as a supply chain risk. This is an American company being designated as a supply chain risk at a time when things like DeepSeek and AI company with ties to China are not currently designated as a supply chain risk. In designating a company as supply chain risk, you say they cannot do business with the Pentagon and that they cannot do business with department of defense contractors. So it effectively blacklist a company from a significant portion of interacting. And we saw this when Microsoft actually

9:25filed a brief in support of anthropic in the ongoing litigation, pointing out that as a part of an of war contractor, this could significantly impact its ability to provide the services it's expected to provide. Despite being a competitor in the in the AI industry, at a time when those services could be particularly crucial to ensuring that there is military readiness. So that beyond the supply chain designation, we've also seen the president Trump in a true social post called to remove anthropic from the entire administrative systems of the US. So we've seen other departments like the state department taking these further actions. And there as of, you know, today we're recording on on March 11th, and I say that business is moving very quickly. Yesterday, there was a was some information that there might even be an executive order that could expand this even further within the government's use of AI

10:27product. So just to spell that out really, really clearly. So say you have a defense contractor that's doing some work for the Pentagon, highlighting them as a supply chain risk, doesn't just mean that that defense contractor can't use anthropic within the services it's providing to the government, but can't use those services at all within its business if it if it engages with the Pentagon. Well, I think that that goes back to where these lines going to be drawn and how do you distinguish when a product is kind of integrated into another product of what can and can't be used. And that's one of the reasons that this is is very difficult. And why we have seen this kind of being portrayed rightfully as a very broad step and a very broad interpretation by the Pentagon, where if you were to try and look at it in the generous light and say that the Pentagon had any kind of reasonable national security claims, are there less restrictive steps it could have taken first beyond this kind of really extreme step of labeling

11:30anthropic as supply chain risk. One can imagine the most simple straightforward answer to that is, well, they could have just cancelled the contract as a less restrictive means. And when we're analyzing the first amendment concerns, that's going to be a key element as well. Not only of is the of a even if the court does not find that the national security claims are pretextual, which there does seem to be, you know, reasonable questions of of is this just a excuse to get this retaliation against the company. You know, were there less restrictive means to resolve those national security concerns if the Pentagon had them? Yeah, I was going to ask you that. Is the fact that OpenAI is willing to provide an alternative source of services? Would the courts take that into consideration or is it kind of irrelevant to this case? I think that's less relevant in this case because again, what we are looking at is the Pentagon's actions towards anthropic as a company and what that meant and was that based on their

12:35expressive activity. You know, there's also a separate case kind of under the specifics of the law that's in the DC court district court. The main one that's going to probably get a lot attention is the one in the Northern District of California. That's the one. Cato joins several other organizations and filing an Amicus brief in and that really kind of asks these very important and pertinent questions about the First Amendment that, you know, regardless of how you feel about AI, you mentioned at the top that AI can, is unpopular in some ways, you should be paying very close attention because what this ends up saying about the government and First Amendment rights is going to be very important, not only in the AI context, but beyond. And from what I've read in Anthropics case, you know, the Sue in the government on the basis both of some concerns about due process in the government making this designation,

13:39but also the bigger idea that government cannot use like its procurement leverage to just punish protected speech or whatnot. Which of those, I mean, I guess, you know, you want to include as much as possible to give to make your case stronger. But how do you think, I mean, it's difficult to predict what the courts will say about these things, but which of those arguments do you think the court will be more likely to judge the case on? I mean, I think that there's going to be a variety of ways we can see this go. While we have seen the courts typically be very deferential to the state when it comes to claims of national security, I think in this particular case, there seems to be such clear evidence that this is about something separate than an actual security risk that I would hope the court would really consider the First Amendment concerns. That being said, there could be something that's more narrow focusing on the breadth of the application of this law,

14:40focusing on, as you mentioned, the due process concerns. But I would hope that for the fact that this could be a future repeating scenario, not only with AI, but beyond that we would see the court really address that First Amendment concern. I also think it's important to think about this in the global context. This is the American government requiring an American AI company to potentially go away from the values it feels are necessary to ensure its technology is used appropriately in a free society. What sort of precedent does this set for countries that may be less free, that would love to pressure American companies to change their technological safeguards to allow it to be deployed for purposes that very much could be used to restrain liberty? Yeah, I think that's a great point. In a point that we talked about a lot when it comes to tech issues more broadly, let's move on to New York. New York's considering a chatbot bill which,

15:42as I understand it, has been interpreted to prohibit a proprietor from permitting a chatbot to provide, and I want to get the language right, substantive response information or advice will take any action that if done by a natural person would amount to unauthorized professional practice, particularly in the medical and psychological space. Now as I understand it in the past few days, the sponsor of that bill is now insisting, this is just about impersonation when you're impersonating, using chat facilities to impersonate somebody a professional authority. It's not a general ban on asking chatbots questions, but the way that the bill is written, I imagine that that doesn't do much to assuage your concerns. No, and I think this also goes to, while AI is new and exciting in many ways, it also reflects just a further ability to access information. So as Kevin Frazier, who is one of our

16:45keto colleagues, wrote about this bill, you could almost imagine this bizarre scenario of, does this also apply to libraries that might have books on how to fix your car and you're not a licensed mechanic? Is this going to apply to YouTube videos that give you step-by-step instructions for something that's a licensed profession? If you really think about what's at the heart of this law, it's very easy to see that it could be a parent to our basic First Amendment standards, and that one of the really exciting things about AI is actually the democratizing aspect of it. The fact that you can get some levels of information that you didn't otherwise have access to in an era where the only option was to go to what's often a very costly licensed professional. Maybe that helps you determine whether or not you need to take that next step of doing that, or maybe it helps serve some of these service gaps that we know exists between people who

17:47need service and can afford it, or in some of these professions where there's far more demand than there are service providers. And I think there's an important economics point here, which Kevin actually gets at in his very good piece, which is good economists always ask the question compared to what, right? So people are always going to want some medical advice, mechanical advice on their car. And in the real world, the alternative to using chatbot technologies is actually just googling a question, perhaps going on to a reddit thread, or whatever, where the advice might often be much lower quality than the stuff that you're actually getting out of the chatbot. So it's important not to presume that the alternative world to people asking these technologies is always going to a licensed professional. Yeah, I saw on X a medical doctor post, you know, that they were concerned about this because what seems to not be realized is that the alternative to people going to chat GPT is not often them going to their primary care physician. It may be them going to Google, or to less informed

18:53sites, or sites that only are able to provide one piece of information as opposed to an AI that can help synthesize multiple sources. Yeah, obviously, you know, we have to be careful with the use of this. But I think that's the same with all sorts of information. You alluded to something that obviously a chatbot's output is effectively speech. So do you think such broad liability rules aimed at substantive advice do run into serious first amendment concerns? You seem to elude to that in your first answer, but could you spell that out for us? Yeah, so when we're talking about AI and the first amendment, I think there are several ways it comes into play. One is the development of AI, the choices about what sources to train on, what data to train on, what algorithms to run the process through, or expressive activity of the developers of AI in the same way that we've seen content moderation in social media,

19:56or website design be held to be expressive activity in the past. When we're thinking about the questions of these outputs, those outputs stem from these design choices. So we're not necessarily talking about the first amendment right of a chatbot. We're very much talking about the first amendment rights of the humans involved in these decisions. That's a really good point. And I also wonder whether this touches on because internet services operate across state lines, you have this sort of New York, New York law, if it became law, seems to me that that would risk being adopted by forcing companies to change that way. There were structural products in ways that it creates a de facto, maybe not national standard, but at least broader standard than just New York. Now the Trump administration is of course continuously signaled its opposition to

20:56a patchwork of AI regulation, some executive orders, but legally, would those executive orders be able to preempt something like this, presumably the New York law makers think that this will be outside of the kind of preemption framework that Trump is talking about. So I think we also have to consider this even without the executive orders and the role that the dormant commerce clouds could play in these type of laws in terms of how they impact things out of state borders. You have on the one hand, in some cases, these questions of basic constitutionality when it comes to issues like the first amendment, but when you're looking at what we're seeing in many states, which are these broader regulatory regimes, they're seeking to regulate AI compute power or AI models and what can and can't be used in them, that involves naturally interstate issues, rather than intra state issues. Now there are certain places where a state could consider

21:56AI policy. So we were talking earlier about the federal government's interactions with data and data privacy and the deployment of AI models. If a state government is considering its own use of AI products in the state government, that's clearly an intra state issue and we may have some states that provide good examples and some states that provide bad examples, but that is an intra state issue. When we're looking at many of these questions that have an impact on the development of AI in general, that is inherently an interstate issue, which means it's going to have to be handled at a federal policy level. So on a practical level, if these kinds of rules do proliferate in a bunch of states, how do you think the AI companies would actually have to respond if these became law in terms of their products? Would it be geo fencing certain things? Would it be overblocking answers, pulling certain services entirely? Just talk us through what you think the consequences

22:59of this type of law might be. In some cases, it could impact development more general, not necessarily with this particular law, but with the development or of AI when we're looking at things that regulate the model or regulate the data use or regulate the compute power as we've seen in other states. In some cases, you may have certain products geo fenced off from a state, but this puts model developers in a really hard situation. They either can't provide the same product in all 50 states. They have to remove information. They may think they think may be helpful to others in general because it could potentially be used by a New Yorker to violate New York law and how they handle that. So depending on the specific law, you end up with usually one of two scenarios. Either the most restrictive state law becomes a de facto federal policy. We see this oftentimes with states like California or New York where because a developer will have to

24:04comply with those laws, it's easier for them to apply it nationwide than to create separate products for large economic markets. Or conversely, in other cases, we see that developers may geo fence off a product from a particular state. This is several years old now, but you may remember a few years ago. There was a program on Google that you could take a selfie and it would match you to what great work of art you most looked like. That product wasn't available in the state of Illinois or in the state of Texas because of their biometric information privacy laws. More recently, we saw Blue Sky pull out of Mississippi at least for some time due to concerns about its social youth online safety regulation again because it was concerned that while it didn't feel that it was in violation that the cost of compliance would not be worth the risk of staying in business in

25:05that particular state. So they pulled out of that state. So again, it's going to depend on the specifics of the law and on the particular company. But particularly for small companies, that can create a real barrier to being able to compete with larger companies. So I think another thing to bear in mind, of course, is that companies have a natural incentive because consumers want it to provide accurate information and to try and mitigate against some harms that could really adversely affect the reputation of the company. If they're giving bad cycle, if the chatbot is giving terrible psychological advice, medical advice, legal advice, there's obviously some gray areas about liability issues, but just reputationally being known as a service that's providing incredibly tough information is a strong market constraint against doing so and for building those safeguards into your product. So I just wanted from a libertarian perspective, what is the role of policy here if anything? Is it just we've already got laws that punish fraud and deception? We don't need anything beyond that. I mean, some of these issues are

26:10quite thorny issues about who has liability for for for certain content that is generated. But what do you think is the big picture kind of libertarian approach to this? So I think the first thing we should do is examine what is the actual harm we're trying to prevent and does that harm already have a policy on the books that either prevents or mitigates or responds to that particular harm. So as you mentioned in cases of fraud, for example, there's often already laws on on the books that can respond to these things. If policymakers want to clarify that by using an AI to commit fraud, you still commit it fraud, that's fine to clarify that in law. I think most courts would probably arrive at that conclusion naturally. But again, if there there needs to be legislative clarity, we can certainly consider that. But that what we want to see is not policy that considers regulating a very general purpose technology that can be beneficial in many cases and benign

27:12in many others for the small amount of potential negative cases. Instead, what we should focus on are what are those particularly problematic and harmful scenarios we agree on and try and provide narrow descriptions of that that can be responded to in a policy framework versus trying to regulate the technology more generally so that we can still have all of those beneficial uses of the technology. So I'm putting you on the spot a bit here and I don't know the altzone. I know that you're following so many different issues that perhaps you haven't looked at it in detail. But I just want to, in New York, why is this sort of regulation emerging now? Is it the incumbent professional self-interest? Has there been extensive lobbying for this? Is it genuine public concern? Or is it simply that AI is cliding with lots of long-standing legal categories, licenses that were never really designed for a world in which this type of technology existed? I just want to, you know, from a political economy perspective, why you think we're seeing this type of proposal right now?

28:13You know, AI is naturally disruptive and that can give rise to uncertainty in various professions and a lot of us, a lot of individuals have uncertainty around AI because of how they're worried it may impact their profession in a way that they haven't necessarily thought about before. So I'm not surprised we're starting to see these kind of concerns in traditionally licensed professions, professions that typically require a great deal of schooling. But I think we have to think carefully about what this may mean for those professions as well. In contrast, for example, to the New York law, you have a sandbox system out in Utah that is looking at, for example, how AI interacts with the mental health, legal, I'm sorry, with the mental health profession and its licensing systems. And are there places where there needs to be clarity about AI products and where they can actually be providing services, whether or not they need to

29:14be used by licensed professionals versus being able to help fill some of those gaps potentially in an area where we know that there are more individuals who need services than there are service providers and that where are those places where we can provide clarity to AI innovators to know that their products are allowed versus shutting down these potential uses of products. Yeah, and I think a great point in Kevin's piece is in certain states, there's a real scarcity here. And hopefully this will provide even more impetus, this debate will provide even more impetus for things like universal license recognition so that actually you can mitigate some of the the scarcity that is generating. You know, a demand for people to use these services as alternatives when actually that in some cases they may well need to see a licensed professional. But let's um, let's move on to the kind of final aspects of this because I think it's related. I do worry that we're going to see way more regulatory proposals on AI just in general developing in the

30:17very near future and that pessimism really comes from the polling that I've been looking into this week. Not only have we seen net favorability for AI in general as minus 22, like very, very low in comparison to most other entities. But Pew's 2025 survey found that 50% of Americans were more concerned than excited about the increased use of AI in daily life versus just 10% who were more excited than concerned. 57% rated AI's societal risks as being high compared with just 25% who said the potential benefits were high. And these outcomes, these survey responses really, really drastically differ from people who are operating and working within the AI industry. So do you think the companies are aware of the scale of kind of distrust and how this environment might lead to bad policy? You know, I think it's always really interesting when we see this polling

31:22on AI and on technology in general because one of the things I always want to ask is, you know, to get a little nerdy of how do the people in the questions define AI? Because most of us have been using AI for far longer than we think about. If you are using talk to text on your phone or auto complete in an email or running a search engine query, that all has used AI for quite some time. So while we oftentimes see this pessimism around AI in this moment, we've actually been seeing a lot of uses of it that we cease to think about as AI. I do think we hear a lot of focus on the potential negatives of AI, whether it's in the media or from policymakers. So I'm not surprised that individuals might have a pessimistic outlook when what they're primarily hearing is about job displacement or energy usage. But the reality is AI is also dramatically improving and saving lives

32:24and we have to think about what's truly at risk in some of those applications. Whether it's the fact that in the medical context as we were talking about with the New York law to some degree, AI is able to help identify cancer, whether it's in the kind of manufacturing context where AI is being able to help come up with new materials or whether it's in cases where AI is able to improve predictions of natural disasters. We have to think about what we might actually be giving up that is quite literally saving lives in the process. Yeah, I think people have a tendency to think very locally. I've been rereading parts of the World of Nations this week. It's 250th anniversary. One of the points that Smith really reiterates is how people think about the things that are very local to their lives. And so I think obviously part of this is worries about job displacement. I think that same Pew surveys found that only 6% of workers thought AI use in the workplace would

33:26generate more job opportunities for them in the long run. 32% thought it would lead to fewer opportunities. But the truth isn't a lot of this stuff. We just don't know where this is going to fall out, right? Historically, obviously over time when you see the development of new technologies very slowly, but you do see whole displacements of whole industries. You know, a couple of decades ago, the modal job in many states was secretary and now, you know, secretaries as existed back then just don't really exist as much. You have executive assistants, but a lot of the tasks that secretaries used to undertake are now undertaken by your outlook, calendar, teams, various other things. Now, there are other scenarios, of course, where we've seen what's known as the Jevons paradox, where if you have a new technology that really improves the productivity of something, makes it more efficient, lowers the cost of doing it. You actually get more of it. So, you know, a lot of people worried when ATMs were rolled out, it was going to lead to the displacement of lots

34:26of bank tellers and a lot of banks would close, but actually more bank branches opened because they were providing complementary services to this efficient service. Likewise, in recent weeks, I've seen data showing that, in contrary to people saying that AI and its capabilities was going to wipe out software development, actually software adverts for software jobs are going through the roof at the moment because the lower cost of actually developing software, more and more firms are hiring in-house for more software developers to improve their product. So, we just don't know really where this is going to fall out in individual industries, how much this will change the whole nature of jobs, how much it will complement certain jobs. So, you know, naturally, a lot of people will reach for the pessimistic side because as human beings, we tend to have a pessimism bias, but there are potential for so many different new opportunities here as well. I completely agree, and I think that, you know, as you mentioned,

35:28we've had this kind of disruption around automation before, and I think for many of us, it's a bit disconcerning, but for many adults, this is the first time they've really experienced it in their life. You and I are both millennials, you know, we were young when the internet came about. We went into the workforce knowing how to email, knowing how to use computers, knowing how to rotate a PDF, but at the same time, now we're experiencing our first real technological disruption in some ways, and that naturally leads to this kind of discomfort at times. But I think what we'll find is for many individuals, if you're able to leverage these tools, as we've seen with other technologies in the past, it will actually lead to more beneficial workplace experiences. It won't necessarily be a replacement. We shouldn't think of AI as a replacement. We should think about it as a tool that can help us improve our own use of time and other resources. So what's striking, though, is that job anxiety is obviously part of this, but the PUE work does suggest that people

36:33are also concerned about inaccurate information in personation, data misuse, been used for criminal activities, but also just more diffuse cultural effects like weak human connection, people chatting to chat bots instead of real people. One analysis I saw said the single most common reason people gave for rating AI's risks as high was that it would erode human abilities and connections. So as somebody working on policy issues of this, how do you navigate that? As somebody worried about heavy-handed kind of regulation. What would you say to somebody that came and expressed those concerns to you from a policy perspective? So our response should be education not regulation, in my opinion. Many of those concerns around things like mis and disinformation, we have to think about what it would mean for the government to be the one to be the ultimate arbiter of truth versus allowing all of us to consider information and be able to decide what sources we trust and

37:39distrust. In some cases, it's very hard to, there's not necessarily a generally agreed upon scientific consensus or scholars are debating each other on the specific interpretations or things like that. If we have the government being the one to dictate that this is the only truth, that's a power that could be abused very quickly. Instead, what we should focus on is how do we help individuals understand what AI is? Be able to kind of fact check for themselves all the basic media literacy skills that we've seen in school being re-taught to us as adults in the AI age now. And then additionally, how are their ways that industry can come to standards that reflect the accuracy of the market experience? For example, you hear a lot about labeling AI-generated content or labeling AI-manipulated content. Well, if you think about the fact that we oftentimes see Photoshop used or may see automation tools used to clean up a podcast or audio engineering or

38:46things like that. No, I'll podcast, obviously. I'll podcast. But even in that scenario, that there would be situations where AI is actually lowering the cost of production and it's not changing the medium. It's just helping provide a clarity. Think about a video where someone removes allowed truck in the background and post it. Whether or not that should be considered AI-manipulated is very different than something that is clearly designed to be misleading. And even in the quote-unquote clearly designed to be misleading phrase, we have to also consider what does this mean for things like parody or commentary that may be trying to make a point as a form of art that is using these tools in the process? Well, certainly with all these issues rumbling on the relative demand for your services has been going up a lot. I know you're extremely busy at the moment. So really thank you for taking the time this morning. You can read more of Jennifer's

39:51most recent content. Comment, pieces, op-eds, briefs on Kato Institute page. That's Jennifer Huddleston. You've been listening to me, Ryan Bunn. I'm afraid that's all we've got time for this week. You can log into our podcast channel every Tuesday and Thursday for new content. But thank you for listening.