Moody's Talks - Inside Economics - The Geography of GenAI

Episode Date: May 9, 2025

Will generative artificial intelligence lead to nirvana or dystopia? Great question, which we don’t exactly answer in this week’s podcast, but we do weigh the most critical downstream effects of the technology based on recent research done by urban economists Frank Levy and Scott Abrahams. We assess how GenAI impacts the benefits of a college degree, the nation’s political dynamics, and which metro area economies will win (think Savannah) and lose (think San Francisco).Guests: Frank Levy, Visitor in the Strategy Group of the Fuqua School of Business, Duke University, and Scott Abrahams, Professor of Economics at Louisiana State UniversityRead Frank and Scott's recent research on Gen AI here: From San Francisco to Savannah? The Downstream Effects of Generative AI (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4874104)Hosts: Mark Zandi – Chief Economist, Moody’s Analytics, Cris deRitis – Deputy Chief Economist, Moody’s Analytics, Marisa DiNatale – Senior Director - Head of Global Forecasting, Moody’s AnalyticsFollow Mark Zandi on 'X', BlueSky or LinkedIn @MarkZandi, Cris deRitis on LinkedIn, and Marisa DiNatale on LinkedIn Questions or Comments, please email us at [e-mail]. We would love to hear from you.  To stay informed and follow the insights of Moody's Analytics economists, visit Economic View. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Welcome to Inside Economics. I'm Mark Zandi, the chief economist of Moody's Analytics, and I'm joined by my two trusty co-host, Marissa Dina Talley and Chris Dorees. Hi, guys. Hey, Mark. And we got a colleague, Adam Kamens. Hey, Adam.

Hey, Mark. And just to tease a bit, just a few minutes, we're going to bring in a couple guests, Scott Abrahams and Frank Levy, the two Akim editions who did a really cool study on AI, artificial intelligence, and it's, as they say, downstream effects. I say fallout, they see, say downstream effects.

And they actually tell us which three metropolitan areas cities are going to benefit the most from AI, the kind of adjustments to AI. So very interesting discussion, and Adam is going to play a big role in that. Before we get there, a couple things happened this week. I thought we should talk about just to consider. one is the Fed meeting. The Fed met on this past week, decided not to do anything. Chris, you want to kind of just give us a quick rundown

and what the market reaction is and whether it had any impact on your thinking of how things were planning out? Yeah, I think market expected the Fed not to do anything, right? So that was not a surprise. Reactions, perhaps more to the statement. The Fed did acknowledge that there's still quite a bit of strength in the economy, some resilience.

So that justified not taking any action at the moment. But they also at the same time indicated, of course, that there's more uncertainty now relative to their last meeting and that they are attentive to the risks of both sides of the mandate, right? So certainly indicating that they're filling the pressure, right? They're trying to figure out whether the price stability or the maximal employment part of the mandate is going to win out here. So clearly some uncertainty there. But again, I think this meeting, not terribly surprising. It's going to be the next meeting that we'll be telling.

We'll have a little more data there in terms of inflation and unemployment. And I think that's the one that will certainly be a driver or decider of the future here. Yeah, it seemed like very consistent with our forecast for the Fed, right? We have a Fed cutting rates for the first time in July, I believe, at the July meeting. And then again in September and then again in December. The idea being that they're going to kind of sit on their hands a bit until they get some clarity around the trade war and economic policy more broadly, and then ultimately figure out that the effects of the trade war are going to be more on growth than inflation and then start

cutting interest rates again slowly in the beginning in July. And I think, correct me if I'm wrong, Chris, but I believe that's consistent with market expectations now. The market has kind of moved towards us in that regard. It has. It has. Yeah. That's pretty much aligned at this point.

So did, did Powell, Chair Powell, Fed Chair Powell say anything in, that you found that you just thought was different than what you expected or not consistent with the way you think? Was it all just kind of straight down? I think it was really, you know, they're very careful in the communication. And I think, I think they, yeah, again, they acknowledge that there is strength. There are some underlying strength here. So, again, no need to panic. We can take our time, wait and see. So that's very consistent with what I thought he would do.

Yeah. You know, I was a little, taking it back is not the right word. I guess it took a little bit of exception to the economy's strong statement. Is it really? I mean, is he just saying that? Because he doesn't want to get, there doesn't want to have the pressure of having to cut it. If he says the economy's weak, then why aren't you cutting interest rates like right now? but is it really strong?

I mean, strong is relative the labor market is still kicking. But even there, is it strong? I mean, it's weakening, right? I mean, hours are down, hiring is off, fewer open positions, everything seems to be weakening except we still haven't seen the layoffs, but

I mean, the GDP number was negative. I mean, I know it overstates the case, but there's a case. No? Again, we got the asterisks around that GDP number. I just wonder if he was as well, gaming things a little bit. I mean, I don't mean that in a pejorative sense, just meaning he's been under a lot of pressure to cut interest rates, obviously. And he comes out and says the economy's,

you know, weak or weak or soft or, you know, anything, but it's doing really well. He's going to come under some pressure. No? I think, I certainly, I think that's a consideration, but, you know, objectively, the economies, I wouldn't say the economy is weak, right? No. It's weakening, I'd say. It's not as strong as it was. Yeah. But I wouldn't say that it's, again, no need to panic at this point. Well, that's different than weakening. I mean, panic is like, no, I'm not saying panic.

All right. Anyway, that's good. Okay. The other big thing is consistent with our forecast, actually, the Trump administration struck some type of arrangement with the U.K., the first, presumably there's more coming. and hopefully there is because we're expecting, particularly with China. Marissa, I haven't been able to, I've been traveling a lot, haven't been able to follow. What is that arrangement?

And I hesitate to use the word deal. I just doesn't feel like a deal to me. Nothing is signed. Yeah. And signing what exactly? So the arrangement that was announced was that the U.S. would exempt the U.K. from the steel and aluminum tariffs and would lower the tariffs on autos, which remember were 25% right for everybody, would lower the tariffs on autos down to 10% for the first 100,000 vehicles

imported. And then apparently there's some exemption for Rolls-Royces, which I didn't know. Which we'll find out later in the podcast. Yeah, we'll find out later. But yeah, so that's the deal. And in exchange, the use. UK said it would lift tariffs altogether on some imports of U.S. beef and energy on ethanol. And then there's also some concessions on purchasing more aircraft from Boeing and opening up agriculture markets in the UK to more U.S. imports. But that's all very fuzzy. I don't know the details.

And I don't think anyone knows the details around that. And this isn't inked in pen. this is something that was announced yesterday by Starmer and Trump in a joint appearance. And I guess it's – oh, sorry, go ahead. And just one more thing, that the kind of 10 percent baseline tariff on everything else stays in place. And the UK has said they're still trying to get the Trump administration to bring that lower. So there's some speculation of, is this 10 percent just going to be a floor for everybody? Does this indicate that, you know, you could work around the edges of things, but there's still

going to be a 10% tariff on most other goods imported. And I think that's really unclear. Right, right. Okay. Well, I guess, I mean, it's good news in the sense that it looks like the administration's taking an off ramp here, kind of sort of what we're hoping for expecting in our baseline forecast, which is a non-recent, a weak-inning economy, but not a able to. recessionary economy. And for that to happen, we got to see more of these trade deals,

arrangements soon. And most importantly, with China, right? I mean, that's got to be. Yeah. I mean, the thing about the deal with the UK is we actually run a trade surplus with the UK. So it's not like this was some huge win that upends the balance of trade here, right? This morning I saw, and I didn't, I don't know if there's more detail on this, but on the Wall Street Journal, it said Trump made a statement about reducing tariffs on China, perhaps down to 80% from the 145 that's prevailing right now. So, and Treasury Secretary Scott Besson is going to Switzerland this weekend to start to talk with Chinese trade officials. So that number was thrown out there. I don't know, I don't know how concrete that is, but that's still also extremely high.

And we got some data from China this morning that shows that trade with the U.S. fell pretty dramatically, right, about 25%. So we're already starting to see sort of this month of April or something. Was it April? It must have been April. Yeah. We're already starting to really see in the hard data the impact of this kind of freeze and trade between the U.S. and China.

Yeah. I mean, I think at the end of the day, what really matters is what we do with China. Yeah. Yeah. I mean, in our baseline, we're assuming that those tariffs come down, not all the, so if he goes, if the administration goes from 145 to 80, that would be consistent with our expectation. And we kind of sort of avoid recession. But that has to happen between now, like in the next few weeks, you know, sometime around Memorial. Yeah, maybe it happens this weekend.

Yeah. Yeah. Maybe. If you're going on that, then it's a problem. Yeah. It has to go below 80, right? 80 is still way too high, right? Yeah, but I think that, you know, that's enough to take the pressure off and signal that

more is coming, that that should be sufficient. It kind of in my baseline worldview, that would be consistent with we can avoid a recession. But if they don't, they can't cut it to 80 soon by, you know, in the next couple three weeks, I think. Yeah, yeah. Although I'd say if it goes. to 80 and stays there, right? That's still recession. Still a problem.

Still a problem. Okay. Still a problem. I get the impression. I get the impression this 80 figure was a number that was just literally thrown out into the ether with not a lot of anything backing it up. It's sort of like 145. By two, you know.

There's a lot of numbers being thrown out. Yeah. And I think 80. Everything you divide by two, just divide by two, you know. Yeah. But then President Trump, said whatever Scott wants. 80, but whatever he wants, whatever he decides.

Yeah. He's delegating. Delegating. Yeah. Okay. Before we move on to our guests and AI, artificial intelligence, Adam, I don't know if you follow this data. This goes into the camp of, you know, I feels like to me the economy is, let me say it again, weakening, it is weakening.

our house price data. We construct repeat sales house price indices for all the areas across the country, metropolitan areas, and monthly, and we got the data for the month of, Christmas was in April? No, March.

It must have March. March. March. And it showed a decline, an actual decline. Now, I don't want to, I don't want to put too much weight on the decline, just like the GDP number,

because there might be measurement issues. It might get revised. We might get more transactions. you know, that kind of thing. But it does feel like that's just another sign of weakening. Do you look at that data at all, the house price data? I do look at it. I haven't scrutinized it that closely yet, so I'm a little.

And I looked at Vera Beach. I looked at Vera Beach very carefully. But is that? I think it held up. I think it held up. Congratulations. But most of Florida did not. But do you take Scott or, I mean, Adam or Chris, do you guys take.

put any weight on that? I mean, does that make you nervous at all that we're starting to see some additional weakness in house prices? Adam? I think, I mean, Chris, you're the expert on that. I mean, I would, I would feed it like I'm reading. All right, so I'll just talk to anything.

You just bumped it. What the heck? You're the expert. No. Okay, go ahead, Chris. Chris is actually the expert. Chris is actually the expert, but go ahead. I'd say it's expected, right?

We'd expect it certainly slowing in house prices, right? Just in terms of the fundamentals, we've been talking about this a while, just prices versus incomes versus rents. There's an imbalance here. We need some correction here, whether it's an actual sharp decline or, you know, moderation and we let incomes, give incomes a chance to catch up, right? I'm not terribly surprised there.

Again, like you, it was down 0.3% month of month, but these things are a little bit volatile, so I wouldn't read too much. read too much into it, but it's certainly consistent with other house price indices that show weakening, right? We had the median house price down. Again, we have other house price indices by other providers as well that clearly show weakening and definitely states like a Florida where you have other house price costs or housing costs playing a role as well. I think that's going to continue to weigh on house price growth forward here. Yeah, the other thing I point out is if prices are, say they didn't incline, but basically

National House prices are flat, right? And they've kind of been sort of flat since the Fed jacked up rates back in 2022. So it's not, for the past three years, more or less gone sideways here. Maybe with a slight upward tilt. But it does suggest that, you know, there are parts of the country where you're seeing price declines, right? Yeah, absolutely. Right? Southeast, parts of the mountain west.

Austin. Austin, Posterchild. Yeah. So what's the forecast for the next year for just national, how? prices. Basically flat. Basically flat, I think. Yeah, pretty flat. It's slightly positive. Yeah. But yeah, there's a lot of uncertainty. What's that? Real declines. After inflation. Real declines, definitely. Yeah. Nominal, slight positive. And then regionally, yes, there's quite a bit of variation. So some areas are going to continue to slide, like Florida. Right. Okay. All right.

We also got the Equifax data for the month of April. That's based on all the credit files. And that also showed some weakness. I mean, it does feel like this resumption of student loan payments is having an impact, right? Not only in the student loan borrowers, but it's impacting, you know, the ability to pay on other debt. So you can kind of feel delinquency rates on credit cards and auto, even first mortgage, starting to move up a little bit. And they're high already pretty high. So another sign of, I'd say, weakening would be the way I would describe it.

Chris, anything there? Yeah, definitely. We saw retail credit cards. Those are credit cards that are, you know, not Visa MasterCard, but from stores that we saw a sizable increase there. So that's, that tends to be a bit of a canary in the coal mine. So it does certainly see, seen as though you got more and more households under some financial pressure. And I agree. The student loan restart, that's having impact.

And I think we'll continue to see. We may not have realized the full impact just yet now that borrowers are made aware of collections and having to rejigger their finances to repay those student loans. You could see more of an impact on spending and credit growth going forward. All right. Okay. All right. Well, this is a bit abbreviated because we want to get to the main event and our guests

talk about artificial intelligence. And with that, let's bring in Scott Abrahams and Frank Levy. And let's welcome our guests, Scott Abrahams and Frank Levy. Hi, guys. How are you? Good. Good, Scott. You don't sound convincing, Frank. I mean, you know. Frank's a Knicks fan, so he's always waiting for, you know, the other shoe to fall.

Oh, I didn't know that. How did you become a Knicks fan, Frank? As I grew up around New York and I... Oh, that'll do it. That'll do it. You know, I'm a Knicks fan because of Villanova. They're all from... I actually, you know, hearing that the new Pope went to Villanova,

Yeah. That account of the two wins in Boston right there. It was a little. Oh, I didn't connect the dots. You, you urban economists, that's how you do things. You see one thing going on over here, another thing going over here. You connect them. Yeah, that's right.

And I should say, Frank is an urban economist. You were at MIT for many years. And, you know, Frank, we've known each other for a long time, but I was trying to think, I don't know how, why do we know each other? How does that happen? Do you recall? No. No.

Sorry about that, but I really don't recall. You were doing some similar kind of study to the one you did here on generative AI. Obviously what we're going to talk about here. I mean, the stuff I really wrote about was on living standards and wage trends and things like that. I did a lot of doing that, and I think we may have connected through that. Yeah, I just remember having email conversations with you over the year. with you over the years. And so when did you stop teaching at MIT? It was not too long ago, right? It was 2012. It's a while ago. Oh, really? Okay. Okay. I'm getting everything wrong here.

Yeah. And did you retire from teaching when you left MIT or did you go on to teach somewhere else? No, I really retired from teaching. We moved down to do. Durham, in part because of grandchildren. And so I hang out at the business school at Duke. I go in a couple days a week and drink their coffee, but I don't teach anymore. Okay, got it, got it. And Scott, you're at LSU, you're a professor at LSU, Louisiana State. Yes, that's right.

I actually did my Ph.D. at Duke, which is how Frank and I got connected. I was going to ask, I was wondering, you seem like a motley crew. I wouldn't figure. Yeah, that's the missing link because we were both in Durham. Oh, very cool. Very cool. And are you a Knicks fan too? No, Bowles fan, which was great growing up and has not been much to be proud of in the last couple decades.

Yeah, so you were a Michael Jordan fan then. Oh, yeah. I had the poster of him dunking the six pack of Coke. Yeah, yeah. Those were good years. Yeah, very good years. Well, thanks for joining us. And you guys wrote a really great paper.

Let's see if I can get the title right. It's titled From San Francisco to Savannah, The Downstream Effects of Generative AI. Did I get that right? Right. That's right. We just changed it. So you've got the correct one.

I got the correct. Oh, what was it before? Well, before we had, could Savannah be the next San Jose? I mean, I don't want to jump ahead of things. but why we changed it was because that seemed to imply that we thought Savannah might become a tech hub in the way that San Jose was a tech hub. But really, I mean, as we'll talk about, we were thinking more about movement of people across places, not anything about investment in AI technology.

So we thought if we changed it, we might steer people more towards where we wanted them to go. Yeah. And in preparation for this, I saw you guys, I read the paper. It was a great paper. and we're going to go into depth here. And it's really about trying to understand what the fallout is, as you say, the downstream effects is of AI on, you know, how people are going to adapt to the adjustments created by AI.

And one of those is regional. You know, where are people going to live? And thus the paper's title. I was going to ask something else about that. Oh, I was going to say, in preparation, I, uh, So you got a great write-up in the New York Times. I thought that was pretty cool. Yeah, we did too.

We, you know, we send the paper around some, and some of these people. I've been around for a while, right? So you get to know reporters over time, and I know Steve Lohr, and I sent it to him, and he thought it was an interesting piece. Yeah, it is an interesting piece. Okay, so let's kind of dive in. And the piece begins with kind of an assessment of the adjustments, the downstream effects from the so-called manufacturing shock.

You know, in some sense, we're feeling that fall out right now in the trade war, it seems to me, that because of the rapid globalization of the economy, particularly when China entered under the scene after coming into the World Trade Organization in 2021, that had a huge impact or shock on the nation's manufacturing base, wiped out a lot of manufacturing activity and jobs and all kinds of adjustments occurred from that. And that's where you begin the paper and try to get a sense of what the fallout was there as a way to try to gauge what's going to happen with AI. Do you want to, I don't know, Frank, Scott, I don't know who wants to go first, but would you guys like to talk a bit about your assessment of the fallout of the manufacturing shock and how you're thinking about that?

Scott, you want to start? Sure, maybe I'll even go backwards. Okay. Just because I was looking up the manufacturing numbers. And while there's a lot, you know, millions fewer people are doing manufacturing in the U.S., manufacturing. output is way up. You know, if you index, say, 2017, I think is the index, is 100. In 1980, we were in the high 40s, around 50. So in real terms, you know, the U.S. is manufacturing more than ever. If you look it up, it's like 16 to 20 percent of global manufacturing output.

So we're actually a manufacturing powerhouse. We've just found a way to do much more of it with a lot fewer people, which is, I mean, maybe not something that you'd get if, you're not. you're reading the news, but it's relevant because, you know, what we're talking about is peak manufacturing employment happened in about 1979. You know, that was a, there were a lot of factors that we talk about it in the paper, you know, energy booms and things that really favored people working in specific areas in manufacturing industries. And then suddenly starting in the 80s, we had recessions and other movements structurally so that we were changing our economy from people making manufacturing goods to people providing services.

And there were people who were specialized in manufacturing in specific areas that the labor demand sort of very quickly shifted away from what they were doing. And so the whole economy is doing much, much better. And even the manufacturing sector, in a sense, is doing much, much better. But pockets of people in very specific places, which many of which we now call the rust belt, there was a shift of demand away from what they had been doing in those places, which...

That's an interesting point, just to reinforce it, you're saying, you know, manufacturing in terms of the actual output, the value of the widgets that we're producing, that has continued to grow and increase. I think as a share of total output, total GDP, it's been relatively constant, you know, 10%ish of, the economy, but you're saying it's growing, but what obviously hasn't is jobs, and the number of jobs have actually declined. Productivity has improved, and it's really that loss of jobs that

you consider to be the manufacturing shock, what you're focused on. Right, and we don't even have fewer total jobs. We've just seen a shift in what the jobs are. Right. It's not, in the economy, there's, of course, way more jobs today than there were. Then, you know, we sort of moved up the the value chain of manufacturing you know manufacturing today is is a fairly high-skilled job relative to what it was then you know they're not going to just let anybody touch a five million dollar machine you've got to you know be trained and know what you're doing you know if you watch this old house whenever they go to one of their production factories you know it's this crazy high-tech thing now where you yeah so you just have to have different training and skills even to do

what on the surface sounds like similar work manufacturing yeah so I thought you identified what was pretty cool about the paper was you identified three different ways that, or downstream effects, adjustments that occurred. Frank, do you want to kind of walk us through those? Yeah, I mean, just stepping back for a second, the point was to say, when you lose these jobs, that's not the end of the story. There's a lot that goes on beyond. Right.

So at that time, there were three things that happened. One was that there was a big push to get more education because you needed education to get into the jobs that were opening up. The second thing was just a lot of migration. The first migrations out of the pressed areas were college graduates moving out, and then the third thing, of course, was the big change in political outlook, that areas that had been traditionally democratic, by the end of that shock period, you began seeing a lot shift to the Republicans.

So all those three, change in political outlook, getting more education and migration, those are the three big adjustments. Yeah, you had one really, a bunch of interesting stuff in the paper, but the one thing I found pretty cool, but you made the point that unionization and churchgoing kind of declined, the traditional church going,

kind of declined during that period. And in its place, you saw, these mega churches and I think he said a lot of national rifle association members, the kind of the change and the kind of that facilitated or propelled this shift from being more democratic, Democrat base to more Republican base. I thought that was a pretty cool observation. That was, I think that was sort of half the story. And the other half was just the ineffectiveness of, you know, congressional Democrats or anybody else to provide any kind of relief for the job suddenly disappearing. You see a lot of that here in North Carolina, because at the time

that you were talking about before, when China came into the World Trade Organization and so on, the whole story was the bridge to the 21st century, just get skills, and everything will be okay. well, that's a, that was a, didn't work so well. Right, right. Adam, you follow the, I'm bringing Adam Kamens into the conversation, our colleague who managed our regional economic activities and follows the demographic trends carefully. Would you concur with the, I mean, the observation of,

particularly as regards to the migration flows, we saw significant movements of people from, parts of the country that had a lot of manufacturing jobs to other parts of the country to find other opportunity. Oh, absolutely. And I think we've seen some shifts. And I think we'll probably get into that as we get to sort of, I think, the next part of the analysis. But the fact that there was this surge and urban in migration in the 2000s in the early 2010s, part of that's demographic, right, that you had a large number of 20-somethings. But I mean, I think it does, It dovetails really well with what you guys talked about in the paper about how these white-collar, high-education, high-skill kind of jobs that in cities is what was really driving the economy.

And there was, you know, people were relocating out of, out of the Midwest, out of the so-called Rust Belt. So I would concur. Yeah. Hey, Chris, Marissa, any other anything else you'd point out with regard to the kind of the downstream effects from that manufacturing shock? I mean, Scott and Frank are, you know, focused on the shift in kind of educational attainment towards college. You need to agree to kind of survive in this post-manufacturing world. The migration flows and the shift in kind of the politics of the. folks that were being affected. Chris, anything else to add there? No, I thought those really succinct

analysis there. Okay. Marissa? I mean, I'm very interested in the education piece of it, right? Because I think at that time, as you, Scott and Frank say, it was everyone needs a four-year college degree. Everyone needs a bachelor's degree to adapt to this new economy. And I think, and I think we'll get into this with the paper, the conversation's sort of changing now. that maybe there's not enough focus of education for skilled manufacturing jobs. And we certainly, I think, see that with the available labor force of potential people that could work in manufacturing and the focus in vocational schools and sub-four-year colleges and what we're training people to do now for the future.

So I'm excited to talk about that with you guys. Yeah. Let me ask, you know, in the current context, you know, we're in the middle of this trade war. The administration has raised tariffs aggressively across the board and, you know, obviously creating a lot of adjustment across the globe. Is it your sense that we're still, this is another downstream effect of the manufacturing shock, that this is a result of that, Scott?

That's certainly a narrative that you'll see. I know there was a paper by David Otter and some co-authors looking at the China shock, and they found that areas that had been more affected saw heavier shifts towards the Republican Party. So there are some studies looking at change in political orientation as what we would call a downstream effect. And then another part of the narrative is, even though the economy is doing so much better, there may be this sense of a relative decline in status in place. where income has not grown as fast or it just seems like things that were there no longer are. And in that case, I may have much less stake in the system and much more, you know, I might be more keen on, you know, rapid or widespread change in the system.

And, you know, it's pretty natural to think, oh, in my area we used to have a lot of manufacturing jobs. were good jobs. We don't have a lot of manufacturing jobs anymore. And so maybe if we made manufacturing by other places more expensive, that would somehow, you know, bring us back. You know, it's pretty, you know, it's a natural way of thinking about things that could definitely be a downstream effect. Does that make any sense to you? Well, there's, there's, one, one thing is, the U.S. is relatively closed. So, for instance, if you add up all imports and exports in the U.S. as a share of GDP, we're coming in at 20, 25%.

So for context in Mexico, that's like 88%. In Canada, that's, I think, mid-60s. So one thing is, you know, even if we completely brought back all manufacturing imports and we're able to make them domestic production, that's not going to be a huge change because it's actually somewhere what small relative to everything we do. But two, that's sort of a reverse causality. It's not like trying to go back in time is going to bring things back, I think.

So it makes sense why people would have this perception, but it doesn't make sense that tariffs are going to have the effect that they hope it does. There was one study, I can't think of the authors right now. I can look it up for you later. They were looking at the effect of the 20- We're a podcast. We're not an academic. committee. I wouldn't worry about it. Just far away. Far away. They were looking at the 2018

tariffs because those were more targeted. And they found that in the areas that they were supposed to help, like manufacturing areas, they did not increase employment in any appreciable way, but they did significantly increase support for the Republican Party. So people like them, but they weren't doing what they hoped. Yeah, interesting. So the difference between perception in reality is what you're saying. You can understand why people perceive it. It can be intuitive, but not the reality of it's very different.

It sounds like that's my interpretation of what you're saying. Yeah, I didn't. Frank, do you have a different view on this? What's your perspective? I mean, I think mine's about the same as you're going to God, that this is a political move. It's not going to have a huge effect

on manufacturing or what's made in this country. And in the meantime, there's a lot of disruption in between. And so I don't even, I'm not sure that even the political calculation is correct unless you're just talking about a very narrow slice of the president's base. Right. Okay. All right. Anything else on the manufacturing shock before we move on to AI, artificial intelligence?

Anything else from the group? Anything else? Anything else people want to. Yeah. Well, I noticed that in the tariff, as I understand. standard in the tariff agreement yesterday with Britain, that Rolls-Royces were exempted from that tariff. I miss that. Is that right?

There's no, the president said Rolls-Royces should be, should continue to be made in Britain because they're a really special car, and there's no discussion of making Rolls-Royces here. Yeah. Yeah. You want to maybe mark that down. Yeah, okay. Yeah.

That's intuitive to me. Yes. Yeah. Okay. That's, that's called sarcasm. I think. What's that, Frank? I was just reporting it straight.

I know you were, Frank. I know you were. I know you are. I know you are. I don't, you know. I know you are. Just the facts, ma'am. One more thing.

It's not entirely related to what we're talking about, but a lot of things manufactured in the U.S. use parts that are built overseas. So we also could be making it to do U.S. manufacturing that relies on imported intermediate inputs. So it would only be good for people who are making the thing that everybody needs, you know, from the base, I guess. Yep, great point. Great point.

Okay, let's move on to AI. So let me ask you this before we dive into the downstream effects. Because there's, you know, this is a script to be written. And it's not actually been written. How big a deal is AI? I mean, how big a shock do you think it actually is going to be? And when I think about it, I think about in the context of manufacturing, really about jobs. I mean, can lift productivity and output.

And that, again, feels very likely and intuitive. But how big a deal is this for jobs? I mean, I listen to some folks in this world. and, you know, it feels like this is going to be dystopic for jobs. We're going to wipe out all these jobs, and unemployment is going to be very high. There's going to be a lot of displacement, a lot of adjusting to do. Then you hear other folks say, ah, not so much. You know, this is going to be, you know, another technology, excuse me, and we'll be able to adjust to it reasonably gracefully.

Scott, where do you land in that spectrum? Yeah, I guess we're getting into the forecasting business ourselves a little bit here. Can I just say everyone forecast? You're forecasting. We're all forecasting. Let's just call it what it is. We're forecasting. And if you're not forecasting, what's the use of?

I mean, in my view, knowledge is the ability to forecast accurately with some confidence. Okay. What's our forecast horizon? I think that might. Okay, that's helpful. That's helpful. Okay, my answer to you is pick the one you're most comfortable with. All right. Yeah, I mean, on the apocalypse, Frank likes to divide the views into team utopia and team apocalypse. If you think this is going to wipe out not only jobs, but all of humanity, or if you think this is going to, you know, unleash unfathomed productivity gains. So let's see.

I can give away some of the stats I pulled for the game. Don't do that. Don't do that. We're going to play the game, which is going to be already quite complicated. Then I won't give you the context on where we are right now. Oh, okay. All right. Go ahead.

I don't want to stop you. Well, I got enough that I think I can still hold something back. But it's useful. I haven't understood it. And I should say, Scott, the problem you're facing right now is the problem we all face every week when we play this damn game. because we're all gaming the system, trying to figure out what to tell and what. Like last week, our colleague Dante

was held in a very important piece of information because he just wanted it for the game. I would rather be informative than victorious. There you go. I like that. That's definitely not Zandi-esque, I'm telling you. But go ahead. All right, well, this could be to my detriment.

But based on census surveys of businesses, and this is outdated now, but this was over basically early mid-20204. If you asked the business if they had used some sort of AI tool to produce output in the last two weeks, we're looking at an estimate of about 5% of U.S. firms currently. Okay. That's where we are now. It's not, which is maybe lower than people might have thought.

Different surveys will get different results, but, you know, this is the U.S. government census estimates. So say that again, 5% of... They said, have your firm used AI tools to produce goods or services in the past two weeks. Okay. And their estimate, so that's about a year, year and a half old now, was we're at about 5%. Surveys of workers using AI at some form in work, which came from the Fed, was 20 to 40%. Oh, interesting. Yeah.

How do you square that survey? Well, one was the Fed asking workers, I guess, and one was the managers asking if they had maybe implemented it in a way that it was incorporated in their production process. I see. Like, if I'm writing a report and I go over to chat GPT and I change it, I don't think the firm would think it was using. Right. Got it. Got it. So is that your way of saying this is no big deal?

No, that's my way of saying it's the apocalypse isn't here now. Okay. The forecast horizon is next quarter. Okay, fair enough. Got it, got it. Okay. Now we can get into some of the things we've done in the paper to try to put a number on this. And maybe I'll turn it to Frank for the macro number to try to get a context.

Frank, do you have those? Well, I have a couple of things. You know, one way of putting in a perspective is to kind of look and percentage terms and compare it to the manufacturing shock. Yeah, right. If you look at the sort of the first five or six years from 1979 up to 1985, we lost about three percent of total U.S. employment. The number of manufacturing jobs lost in that period was equivalent to about three percent

of all U.S. employment. Now, in terms of what you might have. imagine for artificial intelligence, there's a really good paper done by an economist at Warden, Daniel Rock, and then three people at Open AI. And they put kind of probability estimates on virtually every occupation about what's the chances of AI being able to take over at least half of this job, if not the whole job, and speed it up in some way, so on so over, to have a big impact. So if you just take the ones that they're most certain about, that ends up with about 4.1 million jobs that are really at sort of very extreme risk of being gobbled up by AI. And that would be about the same percentage of total employment today.

Roughly have been, I mean, our calculations are focused on metropolitan areas. So that's a little different, but it's not a bad approximation. And what you get is about 3, 4% of all employment in metropolitan areas that are really vulnerable to AI. So you're talking about a significant shock. But it sounds like you're saying kind of on par with the manufacturing shock. Right, right. Yeah.

Now, in fairness, AI is more spread out. Right. Manufacturing was really very consolidated in particular geographic areas. but yeah, you're talking about that kind of shot. Also, it's a little different, isn't it? I mean, because the manufacturing, we're just consuming manufacturer product. Here, the AI is going to lift, you know,

productivity across the board. It's, you know, much more substantive in that regard, isn't it? So it's a kind of different. You're not talking about the output side of the shock as much as you're talking. we're just at the employment side of the equation. Okay, you're focused on the employment side. Okay.

Because just like we talked about with manufacturing, right? Output went way up. Right. You know, the average manufacturing worker in the U.S. is more than twice as productive as they were in 1980. So I guess if you want to be a forecaster, this is exactly where you have to take your first stand because all of these estimates of the effect on jobs,

stem from what's usually called exposure. So the statistics that Frank started out with where they came up with about 19% of jobs are exposed. That means basically they asked ChatGPT, do you think you could speed this up by 50% or more? If you imagine a job is a list of tasks, they asked it about every task. And so if you say, could ChatGBTGPT speed up or take over a majority of the tasks that are required to complete this occupation. You know, that's exposure. But then where you have to take your stand is that if I'm highly exposed, does that mean chat UPT is going to, you know, do all the

menial work and I'll be able to, you know, focus on my more productive tasks? Or does that mean they can fire me and just have chat GPT do what I did? And that's where you've got to make your decision about what exposure means if you want to get an estimate of the effect on employment. Right. Okay. I'm not going to articulate this well, but I'm going to try. Just because we're comparing the manufacturing shock with this becoming AI shock. Sure. Sure. It feels like AI also has potential to generate a lot of jobs, both in terms of the production of the AI. As you can see, you know, we're building data centers everywhere. where the tech companies are hiring people to do perform AI functions and build out the AI infrastructure. But the AI also will create a lot of other jobs.

You know, I don't know exactly what they're going to be, but, you know, I'm pretty confident it's going to generate, create all these new products and services that we're not even contemplating, you know, right now like the Internet did, for example. That doesn't. That doesn't. New work. Yeah, new work. So that's very different from the manufacturing shock, isn't it? I mean, it is and it isn't. You could ask the question, is employment today a lot higher than it was in 1980?

And of course, the answer is yes. So even though we lost a bunch of manufacturing jobs, there were a bunch of other things created there too. And so we're not saying necessarily that AI is going to wipe out all employment. we're just talking about how many people are going to get knocked out of the box in the near term and who may have to adjust one way or another. What happens after that? You're exactly right.

I mean, if you look at what's gotten created off of the web and everything else, I mean, there's been a lot of other kinds of jobs. On the other hand, you're talking about AI taking over more functions than just what happened with manufacturing. But you've also hit on one of the anchors of our point of departures. here for the paper, which is there's going to be sort of the AI has and have-nots, even among similar workers. Am I a worker whose skills translate into AI helping me in the new work that AI creates?

Or am I a worker who doesn't? And that's exactly, you know, whereas the downstream effect of the manufacturing shock was this widening wage gap, say, between college-educated workers and high school workers, you know, maybe now we'll see the AI college and the not AI college. You know, we might see some exact widening there because of exactly what you're talking about, that AI makes some people, you know, much more productive and reduces labor demand for others. Okay, okay, but so like a kind of a working assumption here is that the displacement, the

adjustment created by AI feels like, at least in terms of jobs, feels kind of sort of like the manufacturing shock that we have. kind of navigated through, which obviously had all kinds of downstream effects. Is that fair to say, roughly speaking? Okay. Yeah, it feels the same in magnitude. And what we're trying to see is, but differs in where and who. And the implications of that. What are the downstream effects? So, if we go back to the manufacturing shock and the downstream effects, you know, you talked about the effects on education, educational attainment, on migration patterns, and on our political kind of dynamics.

So you kind of do the same thing in your paper with regard to AI. Do you want to summarize those results? Scott? Okay. Yeah. Unless Frank wants to go. Yeah. Oh, no. You sure don't go. I mean, it's...

Go ahead, Frank. He's being deferential. Scott is, you know, you're, he's being differential to you. I'm not sure why he's doing that, but he, you know, he's doing that. Well, because he's about to get heartbreak, you know, by game six or so. That's what it is. Yeah, but wasn't he just saying that we got the new Pope, you know, from Villal, you know,

was a Villan- Yeah, the Nick signed him to a 10-day contract. They have divine intervention on their side now, right? It looks to me like divine intervention. Yeah, that's what I'm saying. What you want to do. Tomorrow, tune in, tune in at 3.30 to the game. And if this sort of golden glow, this ends over Madison Square Garden, then, you know that we're really in for something here, I think.

Okay. I will do that. I will definitely do that. So, Frank, you want to talk about the downstream effects. Well, let me just start with education. Yeah. The deal is that a lot of what the current generation of AI does is stuff sort of white-collar work that college graduates do. It's a lot of language manipulation and so on, so.

In the future, there may be stuff going, there may be a lot of attempts now to try and apply this stuff to robotics, in which case there will be a whole different segment of jobs that are going to be effective. But right now, you're doing the kinds of things that college graduates do. That's what GPT and things like that are good for. And so you're coming into a situation where colleges are facing a traffic pressure to go to college, number one. And the costs are high enough that there are a lot of people saying, well, I'm not sure that college is really worth it. And now you have AI coming in, and that's going to put further pressure on what a college degree gets you in terms of pay when you graduate.

So I think the idea that you're going to respond to this by getting a lot more education, that part is not a good analogy with what happened in manufacturing. That's not going to be an avenue of adjustment. Well, you know, it sounds like what you're saying is Scott may be out of a job, is what I'm hearing. Well, if you look at the exposure ratings, The economist is very high on the list. Oh, really? Honestly.

Yes. Oh, no. Maybe that means the exposure ratings are nonsense, which is how I'd like to. Well, can I ask why? Do you know the intuition? You know, what we'd have to do is pull up the list of tasks. But if you imagine, what are the tasks that economists do? You know, data analysis, for instance, okay, you could imagine a chat GPT-like thing might be able to handle.

to handle a lot of that. Or, you know, I mean, or if you listen to NPR something, right now, AI could write the report on the stock market or whatever, right? The dollar was up five points today. So a lot of tasks that economists do are things that you could imagine an algorithm could do, I suppose, is how that... Five points.

When's the last time it was up five points? I don't know. See, that's the problem with AI. It gets it wrong every single time. That's right. Which is exactly the debate is who is AI going to help if we want to. I'll hold up. I'll put a pin in that for now.

But I think while we're here on education, I think it was maybe Marissa who brought this up earlier about vocational schools or something like that. If you imagine education is an investment, after the manufacturing shock, the returns to the investment in college seem to get a lot bigger, whereas the alternative seemed to get a lot worse. and we saw a huge increase in the number of people going to college. Here, if this is right, it seems like the returns to going to college are going to get worse, you know, and maybe relative, so the relative option of not going is going to get better. So we might predict, if anything, a move in the opposite direction. People who are on the margin, you know, of going to college, but it's not really doing a lot for them, might definitely say, might definitely, are more like we might predict them to move in the other direction.

And I think there was an article in the Wall Street Journal yesterday, you know, high school graduates getting 70,000 degree jobs because of, you know, shop class and going straight into the workforce. Yeah. Yeah. Well, I can imagine AI is going to change education too, right? I mean, because I can use AI to be educated. It can be designed. The information can be designed in a way to resonate with that individual. And therefore, much better way to deliver. Yes.

Tudors are very high on the list. Exactly. Right. So I can see that as well. Yeah. The BBC is launching an online course taught by Agatha Christie on tribe writing where they've got an AI. Agatha Christie is going to give the lectures. You know, maybe that's another reason economists could be displaced, right?

What I teach in principles of microeconomics, I mean, they could already go on YouTube and get the same thing, right? So if it's just about delivering the content, you know, in that sense, I'm very replaceable. But they miss your humor. Exactly, right. The inspiration, the intangible. The inspiration. I mean, come on. There's like no way to replace the AI can replace that.

What do you think, Mercer? So when we measured, oh, sorry. Oh, no, no, sorry. Go ahead, Scott. That's what I'm saying. When we measure by tasks, right? It's very clear what's exposed, but there's the intangible elements. Yeah, go ahead.

Yeah, Mercer, are you buying into this? What these guys, what Frank and Scott are saying is that the college education becomes less valuable. I'll get what I need elsewhere. Yeah. I mean, I think unless the point that Scott made earlier, which is, or it has to be a very specific type of college education, right, where people need the skills to code large language models and work specifically with AI, right? So it's going to be a very, and we don't know, we don't know all the kinds of jobs that may be created from AI. So it may, you know, maybe a four-year degree in economics isn't going to help you as much.

20 years from now, but something very specific to AI is what people are going to need. And maybe that's not even something taught in a four-year college. Maybe that's a totally different type of educational system outside of traditional four-year colleges. Yeah, I definitely buy into that. Okay, let's move on to the other adjustment migration. So what do you say there, guys? How's that going to – what's the adjustment going to be like there?

I guess, Scott? Sure. I'll jump in. So we saw with manufacturing the college graduates, you know, tended to congregate in cities and those, you know, were the cities where there were already college graduates, right? Because why do people like to live in cities? Either there's amenities, you know, you like, you know, the stuff that's there. I don't know what it is, axe throwing or, you know, paint and sip, whatever, you know, there's nice things. What was that? Painting, what? We have this around here. It's called paint and sip. Like you go and you get a glass of wine and they teach you how to paint a still life or something like that. Oh, I hadn't heard that before. That's something like Chris would do. He plays botchy ball, stays in his basement.

Yeah, Bachi. You know that. Bachi ball. Now he paints and sips. I didn't know that. I'm AI proof. So there's something nice about cities that college graduates have revealed that they like. Right.

And there's something nice about, you know, being. historically there's been a wage premium for living in the city, and now it seems that that premium tends to exist more for college graduates relative to non-graduates. But if we combine that with what we've seen that, this is going to be uneven. If we imagine that there are going to be AI winners and losers within college graduates, where is the exposure highest? And in the paper, we measured this, right? We took this general exposure measure by occupation, and then we looked at all of the metropolitan areas in the U.S.,

and we waited that by who works there, and we came up with the areas that are much more likely to get hit with the shock. And those are, you know, big cities where a lot of college graduates are living now. So you'll get San Francisco, you'll get Boston, Austin, Texas, Denver. These are the places. So it's not Youngstown, Ohio, or, you know, Scranton, Pennsylvania. Pennsylvania where we think the hit is going to be highest. It's in these large urban centers. Do you, is there anything more on that you want to say?

Because I want to push back on that. Just, well, you can go first and then now. Okay. So you're saying, and you actually identify a bunch of cities or smaller urban areas that you think are going to benefit because they're less exposed and will benefit from the kind of the in-migration that might, might occur because of the hit to the bigger cities. So Savannah, that's to the title of the paper. Yes, and that's the subtle point is, you know, if you're a college graduate, right, you were

working in San Francisco, but you're not in a job that, you know, benefits from the big adoption of these AI models, right? You still like to live in cities. You still like to live where there's other college graduates, but you need a job and you need to be able to afford, you know, how, So that was what went into our model for forecasting migration patterns is where am I going to look if I have a college degree but the type of job I'm doing is highly exposed. I'm going to look for an area that's still a city, still has college graduates, but has affordable housing, and the jobs there are relatively less directly exposed to these models. So what were some of the other cities that are urban areas that you... where we predicted we might see some flows. And can I just, can I just say, Scott?

This is a forecast, Scott. That's right. Okay, all right. I'm just saying, it's just a forecast. Oh, you can take it's a bank. No, no, right. The forecast is a lot of mid-sized cities at the south-southeast. So I thought you might ask, so I got the list here.

Savannah. I've got Greensville, South Carolina, Chattanooga, Tennessee, Little Rock, Arkansas. So the city. Not Baton Rouge? What the heck? So Baton Rouge showed up on the list in some degree. Like it had low exposure. It had decent college population.

It had relatively affordable housing. But then, and this is where the list splits a little bit is, is there already revealed intangibles? So then I looked at, is there any recent population growth or migration of college graduates here already? Like maybe there's something that's not in our model, but we can see by looking at patterns. So all the places I mentioned, especially in like Georgia, South Carolina, you know, maybe that's where that number five came in my mind. I think Savannah's been averaging like 5% population growth. And if you look at ACS estimates of net migration of people with a bachelor's degree, you know,

that's really heaviest in Tennessee, Georgia, the Carolinas, places like that. Somewhere like Baton Rouge, where it, split is population growth is relatively low. Net migration of college graduates in Louisiana is actually negative recently. So you've got the Rust Belt cities which fit a lot of the criteria except for the revealed population growth. So that's where the forecast is really going to hinge. Is it they're going to reverse? Yeah, people haven't been going there, but then they'll realize Scranton is a great place to go

because it's affordable, urban, has college graduates, is not that exposed. or is the something revealed by people's recent movements that we're not capturing going to be more important? Got it. Hey, Adam, let me turn to you. I know you're consumed their Scott and Frank's paper, and you've done your own bit of research here. Did you agree with that list? I mean, of those cities? So I think the list of cities that are most affected on the negative side, I agree.

I'm coming up with a similar list. So I mean like the San Francisco's of the world? San Francisco, New York. Denver, Boston, Northern Virginia is another one that was high on the list. These places where generally highly educated people cluster and they have these high wage jobs that are exposed to AI, I think, very similar list on that end. On the places, I agree about Savannah and some of these other places that were called out. I had a slightly different list on the areas that might benefit. So what I looked at just for context here, and this is nowhere near as comprehensive as what all of you guys did, but took some data from the Pew Research Center where they're looking at industry level exposure and then kind of spread that into all of our different estimates of employment at the metro area level to see where places might be most vulnerable.

So again, the losers are similar. Some of the winners I thought that showed up, Fayetteville, Arkansas, Tucson, Arizona, Wilmington, North Carolina. Those were fairly high in the list. I think the general, conceptually, I think it's the same idea, right, that people are moving to these areas that have generally experienced in migration that have lower costs where there are some opportunities out there. But maybe the specifics of what exactly those cities are a little bit different. I mean, the other thing I would say with a, this generally looks to me a little bit. I'd be curious to hear if you guys agree with this. The pattern of the last few years kind of coming out of COVID,

where people were moving out of big cities into these, you know, more far-flung places, these cheaper alternatives, or they could work, you know, remotely and maybe choose to go elsewhere. It almost seems like a continuation of that pattern, but sort of the impetus for that movement is different. going forward. I think that's a fair point. I mean, one of the questions is some of this

what we're talking about is really a different kind of movement in the sense of pulling up roots. The stuff that happened during COVID really depended on you're being allowed to work remotely. And to the extent that you're seeing a lot more pressure on let's get back to the office, I think the migration now has to be to a real job on the place you're moving to and not just sort of staying connected because you're living in Boise

and you're working in San Francisco. But there must be a lot of overlap between the kinds of jobs that are exposed to AI and the kinds of jobs you might be able to do remotely. Right. So there's probably some overlap there, which explains kind of why you see this movement to maybe second, tier population centers that are still cities but are much more affordable than New York or San Francisco. Exactly. And I think we're starting to see that even in some of the migration data.

And I wonder when we think about kind of the COVID era worries about the future of the office market in cities and all of that has kind of calmed down a little bit. I wonder if this is almost kind of the next shock to whether it's the office market or just the cities in general and that they're not quite as out of the woods as maybe we thought. That's a good point. Good point. So this isn't fair, but I'm going to do it anyway. So is Savannah at the top of your list of areas that are going to benefit, Scott?

Is that like number one? Let's see. I printed it out in case you asked. So if you sort it by... I would have had it memorized, but... If you sort it by exposure, Dayton, Ohio is at the top. Ohio. But it doesn't have any recent population growth. So second is the Augusta area of Georgia.

Savannah is third on my list. And Adam, what's your top three? All right. I've got Savannah, I would say. Actually, Atlantic City is showing up on this list. I kind of don't find out. I'm not. I'm going to not. So again, I'm saying this is a very dark thing.

All right. Let's move on. Let's move on. kicked out of here now. Okay. The data seem to support Atlantic City. I think what we know about Atlantic City, that's not legit. Who knows?

Who knows? I don't know. Maybe we'll be surprised, but. Got the boardwalk. A few months ago, it's nicer than I thought. But anyway, Savannah, it's number one. I've got Fayetteville at number two and Tucson at number three. At Wilmington after that.

All right, very cool. Hey, we're running a little short on time. So let's turn to the, political downstream effects of AI. Frank, it feels like you're saying everyone's going to be a Republican. That's what I read. Is that kind of what you're saying? No, I don't think necessarily that's true.

I think. Spoken like a New Yorker who was a Knicks fan. Okay, lived in Boston. Okay. All right. I think that the way we talk about the politics coming out, one of the things that got captured very well, there was a Pew Foundation poll that came out about three, four weeks ago, that polled from their

big public sample, so they're about, I don't know, a thousand people out of, there, five thousand people out of that, and then they're put together a set of AI experts, and they contrasted the responses of the two groups, and the gist of it was, as you would expect, that there was a lot of disagreement. The AI experts said everything was going to be in general, very optimistic, blah, blah, and the public said, we're not so sure. We think things are not necessarily going to be so good. But the place where they agreed was that both groups felt that the government was not going to do enough to regulate AI, that there was a lot more concerned that the government should be in there doing something to be sure that AI doesn't do a lot of damage. And I think that that may be the

axis around which you start seeing political stuff forming because that attitude cut across party lines in terms of the public sample. So as you begin to see jobs being consumed by AI, I think you're going to see people from both parties starting to talk about there should be more regulation. Now, the kind of people that are in the current administration, part of the reason that they signed up was because the promise was reduced regulation. much as possible. I think that may work until you start seeing real examples of jobs being gobbled up by this stuff, and then you're going to start seeing more pressure for government regulation just to slow down the process. And so the party that can take advantage of that, that's the one that's going

to get, you know. Interesting point. Right. Scott, anything to add to that? Just that the key variable that will determine the direction of the forecast is exactly this. What is the government going to do about it. A lot of the downstream effects that we identified from the manufacturing shock were more or less in the absence of any sort of policy intervention to do something about it. That's the first question is, yeah, if we do something about it, then the forecast will be wrong, but adogenously so because people responding to the forecast tried to plan for it. Well, I guess the forecast doesn't necessarily have to be wrong. Those things are going to play out.

It's just that the costs involved in the adjustment will be less pronounced, I guess. Maybe on the political side is where you'd see. Yeah. Close that you could do something about it. Right. Or maybe it doesn't happen as fast, right? Maybe it's drawn out over a longer horizon. Yes, that's true.

And I mean, and you could use examples of things today like word processing that came in so slowly that even though in the long run it destroyed the job of typist, people didn't really notice about it because it was spread over 30, 40 years. I have to say, it feels like there's no regulation now. I mean, we're kind of backed away from trying to regulate AI in any meaningful way, very different than the approach taken by other parts of the world. And I, you know, it's facilitating the rapid growth in AI. I think that's clear.

But if we do have big dislocations, just to interpret what you're saying, it's going to create all these adjustment costs. And whoever, whichever political party is able to identify that, come forward with concrete ways to address it is the one who's going to win. I think so. Yeah. Yeah. Okay. It's not clear. That's Democrat or Republican or somebody else, I guess.

Yeah. Just like that study I mentioned for the 2018 tariffs, people in the areas that got relief were much more happy, were much happier about the Republican Party, even if it didn't actually even change the employment situation. Wow. Interesting. Okay, well, this is great. I mean, I just want to, I was thinking about playing the game,

but I don't think we should play the game. We're kind of running out of time, and it's hard to play the game anyway. And you gave me your statistic anyway, Scott. So I'm just, but I've got, I want to end with, I have a good statistic, but I'm going to do it a little bit differently than we typically do. I'm going to give you the, give you a number in context and then ask you a question. So this is data from a firm, like cast that's kind of a, maybe you know them, Scott.

They're like an aggregator of labor market data around the world. And they've constructed an estimate of the percent of job posting. that are related to AI by just looking at kind of the skills that are being requested in the posting, you know, generally, things around using neural networks, you know, natural language modeling,

you know, that kind of thing. So here's the, here's the number I'm going to give you for in the United Kingdom, the UK, and they do this for most developed economies around the world, uh, 1.1% I think I have that right.

1.1, no, excuse me, 1.3% of the job postings are related to AI, 1.3%. Wow, that seems low. Doesn't it? It seems very low. Let me ask you, what do you think the U.S. is? If the U.K. is 1.3, what's the U.S.? We should have? Go ahead, Frank, you're up. Oh, I'd say 12.

12%. Chris, what do you say? My initial reaction was 15. And Scott? This is overall or just within IT or some other sector? It's, they look at all the postings. They look for, they look for keywords. Any keywords are from robotics to neural nets to, you know,

natural language skills. And they say that's an AI posting. Well, for job ads for economists, where they say they want someone who research as AI. It's already probably one in five. Okay, that 20%? I'll go at 20%. Marissa? I was going to say 20%. Adam. Your UK number threw me. I'm going to go 8%. 1.8%. What? 1.8%. What, okay, here's, this is, you won't get it, but I'm going to ask anyway.

Which country has the highest in the world and what percent is it? The Vatican. Oh, that's a good guess. That's a really good guess. Yeah. Singapore, 3.3%. There's three countries that are higher than U.S. Singapore, Hong Kong, and Luxembourg.

And you think about it, it makes sense. You know, it makes sense. But I, you know, who knows? Maybe it's just the way they're measuring the data. You know, maybe it does feel low, but I don't know. Well, no, but not with the 5%. Good job in there. What's the winner in the U.S. thing? Can you read that all? What do you, what's the winner?

Well, yeah, what's the category with the biggest share of the U.S. postings? I'm not sure I understand. What, what job, what kind of job gets the biggest share of data in the U.S. posting? Oh, I don't know. I'm not sure. I don't know. They were only measuring AI posting. So I'll read it. Job posting is considered an AI job if it requests one. or more AI skills, e.g. natural language processing, neural net, machine learning, or robotics. That's, that's, that's, I should have stuck to my statistic. I told you at the beginning that way fewer firms than we thought are using AI. Yeah, exactly. Yeah, exactly. Anyway, well, this is great. Well, thank you. And thanks for the great paper. And thanks for the forecast. And now you're,

you have a, you're on record. Yeah. We're going to, we're going to do forecast accuracy here. You know, five years from now, we're having you back, and, you know, we're going to count on your accuracy. Right. Double nothing. We'll do the next and six. How about that? Of course, you know, given the way things are going, we might not be here as economists in five years. But because that's, that forecast could be wrong.

And, but Frank, I think you can forecast the next winning. Definitely divine intervention. I think, I think, I think it's in the, it's in the, in the can, as they say. Anyway, well, thanks, guys. I really appreciate it. It was really great conversation. And please keep in touch. And if you change your list, we'd love to know.

And with that, dear listener, we're going to call it a podcast. Take care now.