Martha Gimbel on the Impact of AI and the Trade War on Labor Markets

Martha Gimbel is the executive director and co-founder of the Budget Lab at Yale. In Martha’s first appearance on the show, she discusses the missing BLS job market data, the consequences of losing two months of labor market data, the impact of AI on the labor market in the short and long term, why it is hard to determine which job sectors AI will impact first, why people will keep learning foreign languages, the future impact tariffs will have on the economy, why US treasuries might get left for the hometown guy in a Hallmark Christmas movie, and much more. 

Subscribe to David's new Substack: Macroeconomic Policy Nexus

Read the full episode transcript:

This episode was recorded on November 19th, 2025

Note: While transcripts are lightly edited, they are not rigorously proofed for accuracy. If you notice an error, please reach out to [email protected]. 

David Beckworth: Welcome to Macro Musings, where each week we pull back the curtain and take a closer look at the most important macroeconomic issues of the past, present, and future. I am your host, David Beckworth, a senior research fellow with the Mercatus Center at George Mason University, and I’m glad you decided to join us. 

Our guest today is Martha Gimbel. Martha is the executive director and co-founder of the Budget Lab at Yale. Martha joins us today to discuss AI’s impact on the labor market, the trade war, and the fiscal outlook for the US economy. Martha, welcome to the program.

Martha Gimbel: Thank you so much for having me.

Beckworth: Thanks for joining us. This appearance is long overdue. I followed your work in the private sector at Indeed, at the CEA, and now you are leading out at the Budget Lab at Yale. A lot of interesting work. Of course, as the trade war has unfolded, you guys have put out very timely estimates. In fact, it was funny, you would put out an estimate of an impact of the latest tariffs, and the next day Trump would change it, and so you guys had to go back and recalculate. That must have been taxing.

Gimbel: It is taxing for my staff. I don’t want to pretend that I’m the person running the models. A wonderful economist right now named John Ricco, is currently the holder of our tariff model. Everyone, please say thank you and send cookies to John Ricco, who is constantly having to reevaluate things.

Beckworth: Yes, but you’re running models on tariffs, on the fiscal outlook. You looked at the One Big, Beautiful Bill that went through. Recently, you’ve worked on AI. I want to get to that, really exciting work on AI because we hear a lot of either really optimistic stories or ominous stories. We’re all going to be structurally unemployed at some point. Robots will be taking our place. Before we do that, tell us about the Budget Lab.

The Budget Lab at Yale

Gimbel: Yes. The Budget Lab was launched in April of 2024, which feels both so long ago and so recent. It was started by me, Danny Yagen, who’s a professor at Berkeley, and Natasha Sarin, who’s a professor at Yale Law School. The three of us were working in the Biden administration together. Danny was at OMB. Natasha was at Treasury. I was at CEA. Since this is a podcast, I hope I can say, unoffensively, for the economics nerds, I will say that many of you who are listening immediately went, “Ah-ha, troika.” For those of you who do not know what troika is, troika is the process by which the administration comes up with its budget estimates. That involves Treasury, OMB, and CEA.

I think all of us were just really, really struck both by what a difference the budget estimates made in what policies made it through and what didn’t, and also how hard it was to get rapid estimates of things as you were trying to figure out policy choices. We started the Budget Lab to do a couple of things. One was to provide that rigorous but fast analysis. It was also to look over a longer time period. A lot of our analysis looks over 30 years instead of 10, which is the usual budget window. That’s partly because there are a lot of policies that look very different in the out years than they do closer in.

As an example, our original analysis of the TCJA that we did when we launched, we found that in the short run, extending the TCJA would lift economic growth, but it was at year 10 that the GDP level would be below if you extended it because of the drag on debt. The drag from debt takes time, and so it wouldn’t show up until year 10. If you didn’t analyze the out years, you wouldn’t see it. Similarly, there are other policies where we do think of them as being very long run. For instance, any investments in children. I used to say that no five-year-old has ever paid for themselves in the 10-year budget window. It was pointed out to me that is incorrect. Taylor Swift had paid for herself in the budget window by the time she was 15. Now I say that no 15-year-old, except for Taylor Swift, has ever paid for herself. 

The other thing was that there are reasons that we as society do certain economic policies, and it’s not necessarily that they’re going to drive economic growth. As an example, we sent arms to Ukraine after Putin invaded Ukraine. Adding arms to Ukraine at various points in 2022, I believe that’s right, added 20 basis points onto quarterly annualized GDP. That isn’t why we did it. 

I think it’s really important to be able to think about what is the outcome that you’re trying to get, what is the cost, what are the long-term implications, and how do you weigh all of those things together.

Beckworth: Now, do you find it very similar on the outside doing this think tank work? You took these questions you had inside the administration, you go outside, you work on them, you feel there’s a demand for that, and policymakers are responding to it?

Gimbel: Yes, we’ve been shocked at how much demand there has been. We aren’t able to fulfill every request that comes our way. There’s a lot of demand for modeling and a lot of demand to really understand what are the impacts of policies that policymakers are proposing. That comes from policymakers also who are trying to figure out how to design policies. They don’t want to put something out there that then turns out to not be effective or that they could do it in a better way or it’s just a waste of money. They want to figure that out as well.

Missing Government Data

Beckworth: That leads me to my first big question. That is, how do you do this work given the constrained data that you face or the limited data? We’ve not had employment data. There’ll be some coming up now. We’re recording this here in November. We’ve missed a lot of important data. The real side of the economy has been missing. We’ve had some CPI data, but we’re missing the real side. It’s an issue because the Fed’s got to make a decision about cutting rates or not. The Trump administration’s got to make decisions as well. We don’t know if we’re driving into a recession, a slowdown, or maybe it’s just a soft landing.

Gimbel: I should say the budgetary work we do, the type of data we use is not reliant on that data. We do do a fair amount of work thinking about the state of the economy and where things are going. For that work, like everyone else, we’ve been flying blind. I feel terrible for the Fed. I would not want to be making a decision in December with the limited amount of data that we have right now. You mentioned that I used to work for Indeed. I’m very familiar with private sector data, which is what we have had to rely on for the last couple of months.

Private sector data is great, and it has many, many advantages, but it is not a replacement for public sector data. It’s not comprehensive in the same way. It really reflects what is happening on companies’ individual websites. You have to think about the broader economy-wide context when you are looking at that data and interpreting that data. We’ve been having to rely on the data that we can get from companies, but it’s just not as good and doesn’t present as clear a picture as the official government data that we get. We really don’t know what is going on.

Is the labor market slowing down? Is it slowing down in a worrying way? Is it falling off a cliff? Is inflation worrisome? We have terrible private sector data for inflation. We have somewhat better on the labor market, but we really don’t have that much on the price side. You saw the White House being reliant on data from DoorDash about how much people were paying I think it was for breakfast on DoorDash. I’m not sure that’s the best way to gauge inflation in the country. It’s a weird moment because it is such a pivotal time for the economy, and we just don’t know.

Beckworth: It’s always hard to forecast turning points in the best of times when you have complete data. When you don’t have any data or limited data, and you’re trying to forecast a turning point, it’s really tough. Now, there’s been some attempts. You mentioned private sector data, but also some of the Federal Reserve regional banks have come out with their high-frequency labor indicators. But you’re saying that is no substitute for the real deal.

Gimbel: It’s just not. First of all, a lot of the private sector data is benchmarked against BLS data. Now, in the short run, that’s probably fine, but it does mean the longer they operate without data, the less they’re able to make adjustments on the backend. 

Let us take Indeed, which is where I used to work, so I know it the best. Indeed has a large share of the job openings and job search activity in the United States. It’s not 100%. There’s selection effects in who posts on Indeed, who posts on other websites. You have to think about that really carefully when you are talking about what are the impacts that you can draw out about the overall economy from what a specific company is seeing. You also have to think about how these different datasets are crafted. 

One thing that got a lot of attention over the last couple of months is what’s referred to as the Challenger data, which is about layoffs. They compile layoff announcements, and they say how many layoffs there have been over the last month, and they ascribe a reason as how the layoff announcements characterize the layoff there. There’s a couple of issues there. Not all companies make layoff announcements. There’s selection effects in who announces. The hardware store down the street isn’t issuing an announcement that they’ve laid people off. Verizon is. It’s going to skew toward larger companies that have an incentive to make an announcement and try to characterize what’s going on with their business. 

Then similarly, when you’re talking about why the layoffs are happening, companies have an incentive to describe what is happening at their company in a certain way. For instance, according to the Challenger data, AI is eight times as responsible for layoffs this year as tariffs. I don’t think any economist thinks that that is correct. If you think about the incentives there, are you going to ascribe it to tariffs, which the administration has not been happy about people ascribing economic pain to tariffs? Are you going to say to investors, “Oh, we’re so forward-looking. We’re making big productivity-saving investments for the company of the future.” When you’re working with private sector data, you have to be really careful to think about what is the incentive, what is people’s behavior, and how is that reflected in the data.

Beckworth: Not only do we have missing public sector data, but the private sector data in a politicized environment is going to be very questionable if they’re trying to motivate or protect themselves from outside forces.

Gimbel: It’s not even necessarily a politicized environment. Just in general, if you’re a CEO, you’re not going to make a layoff announcement and say, “Sorry, guys, I’m a bad CEO. I overhired—

Beckworth: Fair point.

Gimbel: —over the last couple of years. That’s my bad. Everybody keep investing in the stock. I promise I’ll do better in the future.” Of course, you’re not going to say that. That may be what happened, but you’re not going to say that.

Beckworth: Even in the best of times.

Gimbel: Even in the best of times.

Beckworth: All right. One last question on the data. Let’s assume we get back to the world where we do have BLS data fully, GDP data fully. Are there things that you would like to see it do better, like a bigger sample size, more staff? If you could wave a magic data wand for a day, what would you do?

Gimbel: Oh, so many things. Yes, the statistical agencies need more money. They need more staff. We need a bigger sample size. They have been playing around with new techniques around how to capture data. I’m interested to see if they had resources, what they could play around with there. I also would just love if we had enough money to take a step back and think about what are the statistical products that we are missing. The United States has a statistical system that even now, after it has been cut for a very long time, is the envy of the world.

When I was at Indeed, I managed an international team, and I made my team insane. Because I was always saying to them, like, “Oh, I have this idea. You can see this thing happening in the French labor market or the German labor market.” My economist in another country would always have to say to me, “Right, Martha, just as a reminder, you can do that with your beautiful US labor market data. We cannot do that because our countries—” even countries like France, Germany, Australia, did not have the statistical infrastructure that we have. We really have a crown jewel here in the United States.

All of that being said, there are questions about the labor market, like technology diffusion and things like that that would be really useful to take a step back and think about what questions should we be asking now to be able to track this moving forward. Census is doing some of this. They have a survey called BTOS [Business Trends and Outlook Survey] that’s been very useful here. 

I’ve been thinking a lot about the Industrial Revolution recently. It’s crazy to realize that we talk about weavers losing their job in the first Industrial Revolution. We have no sense from a statistical standpoint what that looked like. I haven’t seen someone come up with an unemployment rate for weavers in England in 1822 kind of thing. Which is kind of insane to realize. This is this huge moment in labor markets that we talk about, and we’re not able to really quantify it in the way that we can now, which makes sense, right? They didn’t have this giant survey-taking infrastructure at the time. As we’re experiencing technological and labor market change now, it would be useful to take a step back and think about, “Okay, what do we want to be measuring now so we can see what happens moving forward?”

Artificial Intelligence

Beckworth: That is a great segue into your work on artificial intelligence. I’ll just briefly say before we get to that, I listened to a podcast with Bill Beach. He used to be a BLS commissioner, and he used to be actually a colleague of mine here at Mercatus before he did that. He said, “Look, we need to actually have enough money, fund enough money to the BLS so that we continue the current survey and run an experimental one on the side to see if it’s robust enough.” You can’t just say, “Hey, let’s stop this one, adopt a new one.” It takes resources and time and care.

That’s what you’re saying. We need to be thinking and testing. For all those people listening who have some influence over this possibility, please make do. Let’s talk about your work on AI. You’ve had several interesting articles. You have one entitled “Evaluating the Impact of the AI on Labor Market: Current State of Affairs,” “AI and Demographic Changes,” another one. This is so fascinating because there are very different opinions on this. Everything from doomsayers out there to optimists. Even some of the people who run AI firms are very apocalyptic in some of their statements.

I heard this earlier. During the late 1990s when we had the tech boom, people who developed industry, people who used industry were all very optimistic. “Oh, this is going to change the world. It did.” Today, we hear, “Yes, it’s going to change the world, but it may change it in a bad way.” The people who actually are running it, OpenAI, some of their own leaders have questions. A practical question here now, which is what you’re alluding to earlier, is it affecting the labor market? How would we know?

Gimbel: I do think it is astonishing if you think about the hype and the way that people talk about AI. I am not an AI skeptic. I’m not someone who doesn’t think it will affect how we work, anything like that. If you think about how long we have been experiencing generative AI and these advancements in AI that people think can really change how we do work, it is only just now three years. That is just not very long for technological change to impact the labor market, particularly for something like AI, which in and of itself isn’t going to impact the labor market. It’s going to take people having to figure out how to use it for their businesses for it to impact the labor market.

I’m cribbing a line from Erik Brynjolfsson here. Electricity didn’t change the labor market. What people did with electricity changed the labor market. Yes, you have this technology that can do really amazing things. I was saying to someone, “I’m really annoyed. I asked Claude to plot French 10-year bonds, whatever the French 10-year bond is called, versus 10-year Treasuries. It took forever, and I could do it faster.” I’m complaining about that. It could do it five minutes faster than I could, and I could do something else instead. It is actually insane what the technology can currently do. It takes time for businesses to figure out how to use it. It takes time for workers to figure out how to use it. It takes time to figure out what the pros and cons are, where the risks are. 

It would be highly unusual in the history of technological change for AI to have substantially impacted the labor market this quickly. That is a very long lead up to say that I wrote this paper with Molly Kinder of the Brookings Institution and my colleagues Josh Kendall and Maddie Lee, who are with me at Budget Lab. We did a couple of things. One is we just looked at how fast the occupational mix was changing. If occupations are changing more rapidly than they have in the past, that could be due to any shock, but that could reflect AI. 

Then we also looked at whether or not unemployment in more exposed occupations, when I say exposed, I mean exposed to AI, is lasting longer, is higher, whether or not employment has changed. For occupations that have more usage of AI, has something changed? It’s just flatlines. You’re really not seeing that much happening since the release of ChatGPT. The thing that I find a little bit funny about this is I think the study has, in some circles, gotten me branded as an AI skeptic, which is not what I am. I am a technology-paced adaptation skeptic.

Beckworth: There you go.

Gimbel: It takes time for people to learn how to use these technologies, but that doesn’t mean they won’t have an effect. We plan to keep updating the measures that we looked at every month once we get the government data back. Please, can I have my data back now? I do expect that we’ll see some changes over time, but it will take time.

Beckworth: To paraphrase one well-known commentator, the facts don’t care about your AI feelings. That’s what you’re good at. You’re good at, “Just show me the facts. What do they say?” To summarize what you just said is there really isn’t much support for these wild claims that, at least for now, there’s been a big disruption in labor markets. 

Now, I want to go to a specific example because you talk about this in your paper, the young adults, college graduates. You hear a lot of stories in the news. I’m a father of a senior in college. I got a sophomore in college. I’m thinking about this, “Well, is my son going to move back in with me at home?” Actually, that’s a whole other issue on housing for another podcast, another day. There is the issue like, “Man, is he going to be structurally unemployed because all of his skill set now is done by AI?” What can we think about college graduates and AI?

Gimbel: We did look at recent college graduates and whether or not the jobs they’re doing look more different in the past than slightly older college graduates. You have seen a gap emerging there since the release of ChatGPT. It’s very small, and the data we’re looking at is noisy. The other thing to keep in mind is that ChatGPT was released almost perfectly in the middle of a Fed hiking cycle. People have a lot of feelings about AI. One of my feelings about AI is that Sam Altman can never be forgiven for releasing ChatGPT at literally the worst moment for labor economists trying to figure out the impact of AI on the labor markets.

If I had to guess what’s happening with younger workers, my guess is that it’s much more about weakness in the labor market overall than it is about AI specifically. However, there are two papers. One is from some authors at Stanford. One is from some authors at Harvard. Both of them, in different ways, look at young workers that are either in occupations or at firms that are more exposed to AI, have more usage of AI. They find that after ChatGPT, there is a decline in employment for very, very young workers in a very narrow band, so basically just recent college graduates in very exposed occupations.

Now, there are some questions about these papers, and I’m not saying anything the authors themselves would not tell you, which is that the decline happens almost immediately after the release of ChatGPT. That just seems a little weird. Going back to my previous point, employers didn’t know what to do with this technology in December of 2022. It’s very plausible that part of what we’re picking up there is economic effects. Now, is that interacting with AI in some way? If you’re hiring a bunch of coders and the economy is weak, do you think, “Man, maybe AI will affect this and the economy is weak anyway, so I’m going to hold off more than I would have anyway?”

It’s a little bit unclear, and we need to be studying this more. If I had to list my worries for young workers right now, I would put the overall health of the labor market ahead of any AI-related disruption, which is not to say there’s none.

Beckworth: It’s more about the business cycle than it is about some structural change going on right now for the college graduate.

Gimbel: That is my personal guess. I do also want to emphasize in all of this that we’re talking about right now. I am not saying what’s going to happen in four, five, 10 years. When you think about effects in the labor market, it’s always hard to beat the macroeconomic cycle. 

Beckworth: No, for sure. It’s very powerful, and that’s why I like to follow the Fed. It has a huge inordinate effect, not just domestically but internationally. What do you see in terms of industries? Maybe this is anecdotal data you draw upon, but when I read news stories, I hear, for example, about Bloomberg now has an AI-generated newsletter that comes out. Some writers might lose their job or coders. Where do you see, I guess, the evidence pointing in terms of the initial disruption out of the gate?

Gimbel: I think we are really, really bad at predicting the effects of technological disruption on the labor market and also how fast it’s going to happen. Just as a weird example, let’s go back to electricity. If you had said to people who were there at the beginning of the invention of electricity—which is a little bit of a weird statement because electricity exists in nature, but we’ll just say invention of electricity—there’s going to be a job that emerges from this where you are going to stand in a metal tube, and you are going to take other people many stories in the air and then take them back down again. Within a couple of decades after you do that job, that job’s going to be automated away by the use of something called buttons such that people can take themselves many stories in the air and back down again. Everyone would have looked at you like you were insane. To be clear, I’m talking about elevator operators in case people had not guessed. 

We are not good at figuring out what is going to happen. Adam Ozimek of the Economic Innovation Group, which has done some really good work on this under Nathan Goldschlag, and I would encourage everyone to read their work, made the point on Twitter the other day that we have technology that can automatically play pianos, but people still hire piano players all the time.

I think it is really important not to be overly certain about what we know here. I do worry somewhat about people who do particularly rote white-collar work, but that doesn’t mean that maybe what happens to them is that they then skill up and do higher-skill white-collar work. We just don’t know that. Nathan Goldschlag, again, of the Economic Innovation Group, had a great line about this on a podcast, which is, everyone thinks that they’re paid to be smart, and very few people are actually paid to be smart. We’re paid for the fact that we can be great interpersonally or that we can direct staff, things like that. We all like to think that we’re paid for being brilliant. I, for one, am sure that I am paid to be brilliant. I’m sure you are as well, but that’s probably not actually a lot of our value in the labor market.

Beckworth: It’s not inconceivable that if you fed all of our audio files and all the transcripts into an AI that you could create a voice that sounded like mine and come up with lots of conversations, we’re going to have a whole stream of podcasts from Macro Musings by AI. Even I need to have some sense of humility, what’s in store. Yes, we need to be careful, but let me speculate.

Gimbel: Go nuts.

Beckworth: You’re the cautious, sober person here.  I'll be the speculative podcast host. Just little examples here. Recently on X, Mike Konczal of the Economic Security Project, past podcast guest—he delves into a lot of issues, but macro included—he actually showed on X that he ran this DSGE through Gemini, which is free on Google. He literally gave these instructions: Hey, run a fairly robust model like the Smets–Wouters model. It’s not a real simple New Keynesian. It’s pretty elaborate. It has some extra bells and whistles. Do use standard parameters and check this shock, this shock, and this shock. It took 30 seconds.

Gimbel: That’s amazing.

Beckworth: That’s pretty wild. It makes me think about, “Well, someone who spends a lot of their time to specialize in DSGE models, what should they be thinking about?” They need to maybe ride the wave versus being swept over by it? What do you do, right, if that’s your specialty, your bread and butter?

Gimbel: Well, on the other hand, I come back to a question that you asked me at the beginning, which is, do you get a lot of incoming from policymakers? I said, yes, we get so much incoming, we can’t handle all of it. We have certainly had our staff become more productive due to AI. I personally am giving my junior staff, at least I think, more interesting work than I used to. I asked them this and they reassured me that it was more interesting than it used to be, but much like the CEOs, I think they have an incentive to tell me that, so we’ll have to find out. Because I don’t need them to do certain types of work for me anymore. I can rely on Claude to do that. Instead, I can give them whole projects that I’ve thought through and have them go and do discrete parts of that. It’s made us more productive, which means you can do more work, which means we can take on more requests from policymakers.

Beckworth: Interesting.

Gimbel: Yes, it is a productivity shock. We will see how this plays out in the end, but I think there’s a lot of, “Oh my, God, this means that the machine can do my job, but therefore I am useless as opposed to I can then use the machine to do—”

Beckworth: You’re freed up to do other things.

Gimbel: Exactly. When I think about AI and the risks from AI, I worry about labor market disruption because I always worry about labor market disruption. It is not a thing that we are good at addressing. We do not have brilliant ideas for how to handle labor market disruption. We can’t even get our unemployment insurance system to work particularly well. 

I don’t want to say that I’m not worried about labor market disruption. I am. That is the thing I worry about a lot. When it comes to AI specifically and I think about the risks that really stress me out there, it’s much more on the national security, biohazards, things like that. If the worst thing that comes out of AI is disruption in the labor market, I think that will be great. We can handle that.

Beckworth: Oh, for sure. We want productivity gains. We want real growth because that’s how the world becomes a better place.

Gimbel: If I described my job to my ancestors, they would roll their eyes. You sit all day and you read things and you think about them and then you discuss those with people and people pay you money for this. They would just roll their eyes.

Beckworth: It’s funny you say that because people do that to me now when I go back home outside the DC bubble. They’re like, “You get paid to do this, David?”

Gimbel: Isn’t it great? Can you imagine 150 years ago, you would have been—

Beckworth: Working on the farm.

Gimbel: Exactly.

Beckworth: Physical labor.

Gimbel: I personally think this is so much better than milking cows.

Beckworth: We are incredibly blessed to be this day and age, to be in the world that we’re in, the think tank world, for sure. What you’re doing, what you’ve just described, is you’re riding the AI wave. I have a vision of the Budget Lab on a surfboard, and the wave has AI labeled and you’re on the top of it. Maybe someone’s underneath it. They didn’t quite make the cut. That’s awesome. I see lots of possibilities. When I’m doing research here, I can use agentic AI, “Go find me the papers on this topic.” I don’t have to hire an RA to do this. I have to check it sometimes.

Something else I’ve done, I’ve had opportunities to give talks recently, more and more talks. I’ll make an outline of my talk and I’ll say, “AI, clean this up and make it so I can give it in five minutes or something.” Normally, I would have had maybe someone else give me feedback and I don’t need to do that. When we do policy briefs and other documents, there’s less need for an actual human editor than before.

Gimbel: Totally. If you look at past technological change, we used to have millions of secretaries, executive assistants in this country. We really don’t anymore. I’m trying to remember the number, so I’m going to say a number that I think is right. I would like to ask the listeners to treat this as numbers coming out of AI that has not been checked yet. In the late 1990s, we used to have something like 4 million executive assistants. It’s now 500,000. There was a real decline there. That’s because a lot of the things that many of us might in the past had an executive assistant do, we can do ourselves. We have Doodle, things like that.

Beckworth: Yes, that’s a great example.

Gimbel: It is not the case that that led to mass unemployment; the economy adjusted, and I personally would argue more interesting jobs emerged out of that. The woman who does operations for the Budget Lab, who is amazing, in the past, might have had to do a lot more work that she found a lot less interesting. Now she gets to do at least what I hope she finds much more interesting tasks, because a lot of it has been automated. Per your point, in the past, might we have had three of her? Maybe. Also, because we don’t have three of her, we have two more economists. 

I think it’s really important not to just focus on the doom and gloom. I think the other thing as part of these conversations, it is really, really easy to focus on the jobs that we can see going away, that we can see being replaced by AI. It is much harder to talk about the jobs that we see that could emerge from this because we just can’t even imagine what those are. I go back to the elevator operator. It is hard to imagine a technological future that you haven’t seen yet.

I personally don’t consider this an interesting job, and I wouldn’t want to do it. The fact that something that came out of the 1990s was people getting jobs as influencers is kind of crazy. I don’t think anyone would have predicted that in 1997. We can argue about whether or not that’s been good for society or bad for society. Certainly, many of those people are making money in the labor market.

Beckworth: Yes, I know. For sure. That is the beauty of capitalism, creative destruction. Over the long run, you move up the value chain. You leave behind jobs that were more physically demanding, less mentally interesting, and you get to move up. Maybe it’s not you, maybe it’s your child. The question is, how do you make that transition as peacefully and minimize the pain of getting there?

Gimbel: I do want to just follow up on one thing you said because I don’t want to come off as overly boostery, overly calm about this. Again, we are not good at helping people with transitions. We are not good at helping people through disruption. One of the comments that you made was, it might not be you, it might be your child. That is 100% the case. I am grateful the Industrial Revolution happened. My standard of living is much, much higher. Would I have wanted to be a weaver in 1812? Absolutely not. That was really, really bad for them.

I think, as I said before, we don’t have perfect economic data about that. As best as we can tell, many of them never recovered from that. It took time, perhaps even to their grandchildren, for them to really see the full benefits of what had happened. I do think we need to be careful, and I include myself in this, about going like, “Oh, well, it’ll be fine. Technological disruption, it’s happened before.” Disruption is hard. We’re really, really, really bad at helping people through it. We’re better than we used to be.

I joke that one of the most famous lines in English literature is about how bad the response to the Industrial Revolution was. Scrooge saying, “Are there no prisons? Are there no poorhouses?” I am hopeful that the response to any AI-related technological disruption will not be, “Are there no prisons? Are there no poorhouses?” That’s a very low bar. I think we should cross a higher bar.

Beckworth: Sure, absolutely. There’s a big role for policy to play, but it’s also incumbent on the individual to, “Look, you’ve got to get ahead of this yourself. You need to be the person who’s flexible, adaptable, who’s looking at how to ride that AI wave that we’re just talking about.” There’s you, and then there’s also a policy response. You may be in a situation where you have no choice, or you’re structurally unemployed, and you fall behind. 

I want to give an example of this, not AI, but you gave the example of secretaries to, I have AI basically my secretary now, my assistant. In fact, I wouldn’t even say secretary, my research assistant. It’s incredible that we’ve gone from word processors to AIs. I think that in itself is an incredible story. Tom Hoenig, who is my colleague here, he used to be the president of the Kansas City Fed. After that, he was the FDIC vice chair. Recently, we were chatting, and he told me a really interesting story about the Kansas City Fed. I’m not sure when this was, in the ’80s or ’90s, but they saw the writing on the wall. They saw that check clearing was going to be going out of business. A big part of what they did, other than setting monetary policy, a big reason for their existence was clearing checks. 

He said, “We’ve got to get ahead of this. We sat down. We came up with strategy. What can we do to make us the next phase of technology?” They ended up slowly phasing out checking, but they thought it ahead of time. They slowly phased people out of those positions. He said, “What happened is the Kansas City Fed now is the biggest clearing house for federal government payments.” They got ahead, they got new IT. Somehow they got their systems before the other regional banks and federal government uses them to clear a lot of social security payments, whatever they’re doing. He said, “You know what, David, we lost all these other jobs, we now replace them with much higher-paying jobs.” He goes, “Much more satisfying jobs, there’s some pain involved. We tried to be slow and careful, and maybe just through attrition.” At the end of the day, you had to shut down check clearing because there literally was no more business for that. Now they’re doing some automated payment work, which is a fascinating story and I think an example of what AI can be.

Gimbel: 100%. You didn’t say this about the story, but one thing that’s interesting to me there is—I was about to say private sector, it is government—but it is an entity figuring out how to do it and retrain. One thing that I worry about is that people often say, “Oh, one thing we’ll have to figure out with AI is how to retrain large parts of the workforce.” The federal government is often not very good at retraining. It’s something we spend a lot of money on, but are not always very effective at. There are individual programs that are good. They’re usually smaller. It’s a hard thing to figure out how to scale. I do worry that people are waving their hands a little bit and they’re like, “Oh, we’ll do some retraining or people will figure out how to become entrepreneurs.” Those are hard things.

I wish we were thinking a little bit harder about how to think about those things in general, which I will say is my big thing about policy in response to AI, which is that we have no idea what this is going to look like. Maybe Silicon Valley is right, and half of white-collar workers won’t have a job in four years. Maybe it’s a different version of that, but we should be trying to think about policy and responses that can stretch and respond to whatever the version of the future looks like, and don’t come in with a preexisting thesis about where that’s exactly going to happen.

It shouldn’t be the job of the government to decide, “Oh, these are the jobs that are going to emerge from AI. We’re training people for those now.” Government’s not good at that. We have to think about policies that are flexible and can adjust and adapt easily to whatever the future might be.

Beckworth: What you’re hinting at is a robust social safety net to get people through the transition. I mentioned that you need something like that, but you also need—again, this incumbent on the individual—but really a third thing drawn on my story from Tom Hoenig in the Kansas City Fed, you need leaders in industry who are also thinking proactively, how can we minimize the transitional pains and phase some people out, bring in new people and minimize the disruptions. 

Let me, again, go back to my speculative self and just throw out some crazy questions. What do you think about AI and learning second languages?

Gimbel: Oh, my. 

Beckworth: I know that’s out there. It’s not really what you guys are doing over there at your think tank. It’s something I’ve thought about a lot because I’ve seen commentators say, “Hey, do we really need to learn these foreign languages if AI can give you instant translation?” I bring up my son again in college. He’s a theology major, political studies major, and he has to learn in German because a lot of the old theology was in German. Would he need to do that? Would someone need to learn foreign language? What are your thoughts?

Gimbel: You are talking to either exactly the wrong person or exactly the right person. I am an economist, but I was a classics major.

Beckworth: Oh.

Gimbel: I studied Latin for 10 years and ancient Greek for eight.

Beckworth: Perfect person, yes.

Gimbel: I can give you a whole spiel on why you should learn Latin and learn ancient Greek. Listeners, hit me up. You should do it. It’s really fun.

Beckworth: My son should hit you up because he has to take ancient Greek.

Gimbel: It’s really fun. It’s really interesting. Am I going to tell you that there’s some specific economic skill that I got from that? No. I was learning to use my brain. I was learning to think about things. At some point, there are just intellectual activities that we do because they are interesting and they engage our brains, because reading Tacitus in Latin is fun. Reading Homer is fun. I don’t think that we should overly index on, do we have to do this? We get to do this. We get to do certain things.

I will say also, translation is really, really hard. There’s an Italian expression, another relatively less useful language that I’ve studied, that says traduttore traditore, which means the translator is a betrayer. It’s an interesting phrase because, by the way, even just in translating it, you can see why that is true. In Italian, traduttore traditore, it just flows. You feel it.

Beckworth: You feel the passion.

Gimbel: You feel the passion, and it rhymes. It’s great. There’s no way to translate that into English that conveys that. You lose something. I think we should all be learning many different languages.

Beckworth: Fair enough.

Gimbel: I think we should be learning languages that no one speaks anymore. I am a huge proponent of this.

Beckworth: Your hot take on why AI is going to be super awesome, it will free us up to have time to learn all these other languages.

Gimbel: Honestly, 100%. If I am replaced by AI, my plan legitimately is to do a bunch of math and to re-up my Latin and ancient Greek. I’ve said this to people, I’m on the record.

Beckworth: Fantastic. It’s true in terms of other languages, you often have to use a German word to capture some unique frustration or essence, and it doesn’t translate into English.

Gimbel: Or there’s just cultural concepts that don’t translate as well. This is a thing that comes up a lot in Latin love elegy, poetry translation, where just the way that the Romans thought about sexuality and gender is just different than we did, and there are not concepts in our language, per your point.

Beckworth: To capture that.

Gimbel: To capture that. You miss out a lot on it because there are literally no words for it.

Beckworth: Now, I know this is an AI discussion, but one last point on this. Something I’ve dabbled a little bit in is old biblical manuscripts in ancient Greek. You miss nuance. When you read the English translations, there are important nuances of the Greek words that you don’t get unless you—my son, he knows Greek now. He’s been helping me understand this and appreciate this point. Then textual criticism and which manuscripts do you use, and what version? There’s a lot of importance into languages. I will completely admit my question was not well founded, that I asked a few minutes ago about AI and language.

Trade Wars

Let’s transition the time we have left to another topic you guys have worked on, and your models are a part of this. This is the trade wars. We’ve talked about this earlier. You guys would come out with estimates, “Okay, what’s the effective tariff rate, or what’s the effective burden of these tariffs,” and then, of course, Trump would change things overnight. You’d have to go back and start over. Let’s go there because this is one of the things in your staff’s wheelhouse.

One question I think is very interesting, and a lot of our listeners would love to hear, and it’s something that I’m guilty of, and I’m not going to say you are, but it’s this point that a lot of economists at least hinted at or suggested that, “Man, if we take the tariff numbers that disrupted stock markets and bond markets, we’re going to be in a recession, or the economy’s going to tank.”

Guess what? We haven’t completely tanked. Now, maybe we are in the process now going back to the earlier conversation. Maybe we are beginning to have some turning point. We just don’t know it, but how do we respond to the critics who said, “Ah, you said there would be a recession,” or “Ah, you said there would be inflation, where is it, economist?”

Gimbel: First of all, I would point them to Budget Lab’s work, which said that by the end of 2025, growth would be about a half a percentage point lower with the Liberation Day tariffs. We said, slowing growth, rising prices, but not recessionary. I do think it’s a little bit hard because there are many things happening at once. I think if you think back to the policy environment, in general, that we’ve experienced through a lot of this year, particularly in April and May, when there was really the most Sturm und Drang, to use a German phrase, part of what economists were reacting to was the uncertainty of it all.

If you talk to economists, the main reason that many of them were worried about a possible recession was actually the huge increase in uncertainty that you were seeing at that time. They were worried about people not being able to make investment decisions under uncertainty. It was less about the tariffs and more about the way in which the tariffs were being implemented. This is not to say that economists tend to like tariffs. We tend to think that they slow down growth and they raise prices. I will say, I think that you are seeing a slowing down in growth and you’re starting to see prices going up, which is part of the reason why the Fed is a little bit itchy right now.

Beckworth: What do you do?

Gimbel: The other thing I think to keep in mind, and it’s Ernie Tedeschi, who originally built out the Budget Lab’s modeling on this and has been a frequent Macro Musings visitor, has pointed out, which is that if you look at past periods of tariffs, it takes time for the effects to be fully felt. Smoot–Hawley, the effects don’t happen immediately. I think sometimes, when economists talk about the effects of policies, people say to us, “What will happen if X policy is implemented?” We say, “Oh, growth will slow down and prices will go up.” We don’t mean tomorrow.

Beckworth: It takes time.

Gimbel: It takes time. I think that that means that people then go to the grocery store a month later, two months later, and they’re like, “Hmm. All the economists keep yelling about avocados and bananas. Prices have gone up, but it’s not crazy. There’s other things going on. I guess they’re wrong here.” We’re like, no, it takes time. It phases in. There’ll be pauses. There are other factors. They confound. We’re talking about counterfactuals. Everyone just rolls their eyes. I do think this is partly on us just in the way that we talk about it. It is relative to a counterfactual. There are other things that are happening in the economy. Also, that these things happen gradually over time.

Beckworth: That’s fair. I would push really hard on the fact that when those tariff numbers were released and markets were tanking, eventually Trump pulled back. It wasn’t like a persistent pressure of these rates. He would go back and forth. Or not.

Gimbel: People talk about the TACO trade: Trump always chickens out. I actually think that Trump has done an impressive job of conditioning people to tariff rates that if they had been told they were going to face them at the beginning of the year, they would have totally flipped out. If they’re changing all the time, I don’t want to say an exact number because who knows when this will get published. But if you had said you’ll be facing tariff rates of about 15% at the beginning of the year, people would have cried and rendered their flesh and done the whole thing.

There’s a little bit, I think, of frog, boiling pot of water. I will also say, and Joey Politano has made this point, that a lot of growth in the economy is being driven by AI, as we were talking about earlier. AI is a sector that the administration has specifically carved out from tariffs. What you are seeing at the moment is that the economy is being driven by the main sector that we are not tariffing.

Beckworth: That really puts a new light on our previous discussion about AI, is that if not for AI, we might be in a slowdown, a severe slowdown even, if we didn’t have the one sector that was really thriving.

Gimbel: I think counterfactuals are hard. If we didn’t have AI, the Fed would probably have cut rates more quickly. That might have been helpful to the construction industry.

Beckworth: Fair enough.

Gimbel: This is why everyone hates economists, right?

Beckworth: A lot of multiple steps, yes.

Gimbel: On the one hand, on the other hand. It is the case that a lot of the strength that you’re seeing in the economy right now is coming from a sector that is not tariffed.

Beckworth: Is it not the case that the stock market’s strength is really tied to the AI sector? You take that out and the stock market doesn’t look as good as it does?

Gimbel: It does not, no. Which certainly stresses me out.

Beckworth: I think that was part of the reason some of us were worried in April is because the stock market, it collapsed. Not only that, the Treasury market, that was the other thing that made it really unique because typically they move in opposite directions. Man, the Treasury market went down. In fact, that was what really made Trump pull back. It wasn’t the stock market; it was the Treasury market, which was really interesting to see, what does it take to discipline him? It was the Treasury market. Since the Treasury market seems to be okay, if anything, I’ve been surprised.

In fact, let’s end on this note. Given the incredibly dire fiscal outlook over the long run, we’re talking debt-to-GDP getting really—in fact, what are your forecasts for debt-to-GDP over the next decade?

Gimbel: We don’t forecast debt-to-GDP. We follow CBO.

Beckworth: CBO’s numbers.

Gimbel: It gets higher and higher and higher.

Beckworth: It’s 120%, a very conservative estimate. Some organizations put it even higher than that if you make the One Big Beautiful Bill permanent. It gets really large. What’s shocking to me is the Treasury market doesn’t seem to have internalized this yet, or they have a good reason not to. Maybe they think there’s going to be reform before we get to the cliff.

Gimbel: I think the thing about the Treasury market, because the Budget Lab did some work in 2024 about people in markets underpricing political risk in the United States. The risk of dysfunction and things like that has really gone up in recent years. It’s certainly gone up this year. We felt that markets were underpricing that. I think one reason for that is, for lack of a better phrasing, they don’t really have a better choice than to underprice it. It’s a little bit like, “What are you going to do?”

Beckworth: What’s the alternative? That’s a great point. That’s something I’ve effectively argued for dollarization. There is no other balance sheet big enough to take all the demand for safe assets. It has to be Treasuries.

Gimbel: Europe hasn’t had a huge amount of economic growth, so it’s not like you want to necessarily invest there. China’s complicated. I was talking to someone—

Beckworth: Great point.

Gimbel: —in April who was on Wall Street. I said, “What are you all doing? What new investments are you all making right now?” He said, “Australia and New Zealand, but we’re not that happy about it.” It’s a little bit of, we get to stay the fanciest girl with the ball because there’s no other option. That’s a very risky thing. We’re just relying on the fact that everything is fine. If, at some point, a better option emerges, we could see people disappearing from our market like that.

Beckworth: Those are two, I think, great ways to maybe frame this. One is, there is no good alternative. Secondly, when you have a bank run, they say it happens gradually, then all at once.

Gimbel: Exactly.

Beckworth: Maybe for now, we’ve been given some breathing room, some space, and the hope is that policymakers will get their act together.

Gimbel: I will say it is a Hallmark Christmas movie season, or I guess I should say now Netflix Christmas movie season. Right now, the rest of the world is the heroine who’s stuck in her engagement to a boring big city banker that she doesn’t really like that much.

Beckworth: I love this.

Gimbel: At some point, she may go home and realize that there’s a hardware store owner that’s popped up that she didn’t know about that she didn’t know before. That’s going to be bad for the big city banker. He’s just acting like he’s got nothing to worry about.

Beckworth: Well, on that fantastic analogy, we’ll end there. Our guest has been Martha Gimbel. Martha, thank you so much for coming on the program.

Gimbel: Thank you so much for having me. To everyone, I will say, learn ancient Greek and Latin. It’s really fun.

Beckworth: Macro Musings is produced by the Mercatus Center at George Mason University. Dive deeper into our research at mercatus.org/monetarypolicy. You can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. If you like this podcast, please consider giving us a rating and leaving a review. This helps other thoughtful people like you find the show. Find me on Twitter @DavidBeckworth, and follow the show @Macro_Musings.