Moody's Talks - Inside Economics - The AI Series: The Value of Labor

Episode Date: August 18, 2026

MIT professor David Autor, one of the world’s leading labor economists, separates AI hype from evidence, explaining which jobs are most likely to be augmented rather than automated, what that means for wages and middle-skill work, and what past technological shifts teach us about the adjustment ahead. While big changes in the labor market are coming quickly, Autor makes the case that they are manageable—and outlines the policies that could ease the transition. 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 Sandy, the chief economist of Moody's Analytics, and I'm joined by my two trusty co-host, Marissa Dina Talley, Chris Doretti. Chris, hi, guys. Hi, Mark. Hi, Mark. Good to see you. Good to see you as well. So, Marissa, I thought you were going to be on a beach or somewhere today. No? You're back at work. I'm back. I was on a beach yesterday, but I'm back. Yeah. Good trip away. It was great. It was wonderful. Yeah. Excellent. Well, I don't know that you guys have been a pretty quiet week for me. I haven't gotten a whole lot of emails or phone. calls or anything. Chris, have you, it's been quiet for you to? I guess that makes sense. Everyone's on vacation, but it's really quiet.

Getting a lot of work done. So, you know, in the last year and a half, this might be the quietest day I've had in here half. I don't know what that means. But to liven things up, we've got a guest, a great guest, David Otter. Hey, David, good to see you. Hey, thanks for inviting me on. Absolutely. David. David is the Daniel and Gail Rubenfeld Professor of Economics at MIT, and you're now newly anointed head of the Department of Economics at MIT. Wow. That's right. I've been beheaded.

Is that what you say? That's what I say, yeah. Well, what an honor. Hey, I was looking at your bio and preparation. I noticed you were raised in Newton Mass. That's right. I'm a local. Yeah, good town.

I had an uncle there. We would visit every summer. Maybe we saw you on the playground or so. I saw you on the playground or something. something. Probably maybe your uncle bullied me. I was, I was a pretty young kid. But that's right outside of Boston, right? You're just a yeah. It's a suburb west of Boston. And then you went to Tufts, which is Boston. That's right. Well, I actually started Columbia and then I dropped down after three semesters. And then I returned to the Boston area, spent a couple of years working. And then I

went to Tufts. And then I moved to the West Coast. And then I came back for graduate study. Oh, I'm sure there's a great backstory there, but we'll move on. But then, and then a Harvard PhD. From the Kennedy School government. So my PhD is not in economics, it's in public policy. And you say that with pride. It's not in economics. I know.

I know. I, well, I, I'm proud of, you know, my education studies, but it's mainly to convey the idea that I'm not a real economist. I just play one at MIT. Oh, yeah, yeah. It's like that holiday and commercial. Do you remember that holiday and commercial? Probably nobody remembers the holiday and commercial.

I slept at a holiday and therefore I can, you know, so whatever it is. But then MIT, and you've been on the faculty of MIT, that your entire academic career has been in MIT, right? That's correct. I've been here since 1999. Very fortunate. And I, you know, I learned, I got a lot of my economics education here, really. I didn't intend to study economics at Kennedy School. and I, so I kind of fell backwards into it, only realized partway through my PhD, I was like,

oh, I should have done a PhD in economics. And so I was hired, you know, hopefully had some potential, but I was very unfinished. So I learned a ton from my, from my colleagues and the students here is it's an incredible learning environment. So, you know, I'm very grateful to do that. You know, that's why I was, you know, willing to be chair of the department also is because I feel like, I owe it to the institution to try to play a leadership role. Got it. Yeah. Well, I saw Larry Katz was your Ph.D. advisor, right?

I mean... That's right. He was on my three advisors. Yeah. That's a huge deal of the Harvard Ed School and then Tom Kane at the Kennedy School. Yeah, very cool. Hey, I'm just taking a wild stab here, but you're a Boston Red Sox fan, right? No.

I actually know nothing about sports. No. Oh, my gosh. Are you kidding me? This is probably the foundation of my academic success is I know the part of the male brain that should be given to sports and the part of the economics brain that should be given to macroeconomics. I don't have either of those things. So I have extra capacity. What a shame.

I mean, Boston's got to be. I know, I know. You know, the Citadel for professional sports. I mean, you got everything. I know. I know. And I get no joy out of it. Boy, and I was going to rib you because I'm from Philly.

and we just got LeBron James on the Sixers and now we're going to beat you guys down, but you don't care. That doesn't work. That's basketball, right? Yes, yes. I've exhausted my knowledge. Yeah, exactly, exactly.

How about Wawa or Duncan Donuts? Yes, Duncan Donuts, yes. Oh, yeah, Duncan Donuts. Do they have Wawa in Boston? I don't even know that. I don't think we have Wawa. No, no. That's awesome runs on Dunkneys, apparently.

It's coming, David. Wawa is Wawa. What is Wawa? Wawa is a convenience store, which that does not do justice to Wawa. I mean, once you have a Wawa, you can't go anywhere. Absolutely, right. Yeah. Oh, cool.

All right. Yeah. Yeah. It's a wonderful place. It's actually headquartered, I think, near where we live in suburban Philadelphia, I believe. But it's now all up and down the East Coast. It's in, I guess not in Boston, though. But once it comes, you'll really enjoy it.

I'm ready for it. You know, 7-Eleven is a cultural institution in Japan. Like they have really distinctive food. People love it. It's like it's it's really treasured. It's weird. Some other country, some other nation was trying to buy 7-11 and the Japanese said no. They didn't want it to leave yet its national home.

Yeah. Well, it resonates with me because I feel the same way about Wawa. I mean, that's right. I'm not seeing Al Jazeera strong. If they took it, if that someone tried to take Wawa away away from me, I'd be pretty upset about the whole thing. But yeah, anyway. But it's so good to have you. And we're going to talk about AI and the labor market,

because you've obviously been doing a lot of really great work there and want to get your insight. But before we go there, since we have you, the job market, what do you make of it? I mean, we got a jobs report for the month of July last Friday. All intents and purposes, it was pretty weak. But if you look at jobs, we're not creating a whole lot. Hiring is very weak, hours worked or week.

Wage growth is decelerating. now below the rate of inflation. But on the other hand, no layoffs, at least looking at U.I claims and the job opening labor turnover survey. And, of course, the unemployment rate is very low at 4.1%. So when you look at all that, how do you think about the labor market? How do you characterize the labor market? Well, there are a lot of different ways to think about it. It's not great. It's not an apocalypse. From the, like from the West Coast Silicon Valley View, they're stunned that it's holding up. They're like, why hasn't the labor market collapsed already?

We've had given way I for three years. Like, why is the one still working? And so I would say it's not great. It's a low, higher, low-fire labor market. The employment, the labor force participation rate is falling. And wage growth is much, much slower than after the pandemic, especially below the median of the income distribution, where, you know, we actually saw a lot of wage compression, you know, a reduction,

a major reduction in U.S. income inequality or earnings inequality after the pandemic. And I think, you know, it's really hard to put your finger on why this is the case, right? You could give a lot of possibilities. And depending on your political leanings and your technological preferences, you could come up, you know, you could say it's AI, it's the Trump administration, et cetera. I think, I don't think the evidence, there's not smoke. smoking gun evidence that says AI has had a huge impact on hiring. Even the most persuasive evidence that we've seen says, you know, in, you know, junior software engineers, you know, firms are

hiring less of them, fewer of them, although they're hiring just as many people in software. But so, but it's not, I don't think we're in any sense seeing the, you know, we're not not seeing any sense of, you know, strong number of AI layoffs, you know, firms do, you know, what, you know, what we call AI washing, right, where they say, oh, they don't want to say, you know, revenues are down. They'll just say, we've realized efficiencies through AI. But, you know, and another issue that I think employers are facing is, it's actually a very tough labor market for hiring because AI has made the signal extraction much problem much harder.

Actually, I have some students who are working on this, and part of the reason firms may be shying away from junior workers, they all look the same. Right? And it's like, wow, it's almost as if that resume was tailor made for my job. Oh, actually it was, yeah, by AI. I didn't know that. Yeah, it's really, it's a very serious, it's a very significant issue, right, that employers are having trouble figuring out what is reality when everyone has, you know,

I mean, there are these websites that, you know, do mass, mass applications for you, right? And mass customized applications, right? You pay for it. But they write the cover letter, they tailor your CV, send it on in. You may never even realize you applied for that job. Huh. So, geez, I thought it was the other way around. I thought the employer had such an advantage here in terms of discriminating because of the, you know, the AI capabilities.

But it's working in the opposite direction. Yeah. I mean, you could say, like prior to generative AI, right? You had all these, you know, application management tools, right? You have some tools that do video interviews, hire you and so on. And you could say in that period, you know, AI, the tools were really working on the, I don't know you want to call it the offensive side, working on the employer side, right, giving them additional leverage and insight.

But now the tools are mudding the water. So in some sense, the arms race has shifted in the other direction. And applicants are, you know, kind of have gotten the upper hand. Now, I don't, it's not a good setting. It's kind of a, it's not good for anyone. Right. It's not really helping applicants, not helping employers. It's kind of a prisoner's dilemma, right? You know, you would like everyone else to not do it and just you do it, right? I'll just use AI and everybody else holds still and all look really good. But when everybody does it simultaneously, it just creates, you know, a haze. You can't, you can't tell, you know, what, I mean, you must, you must get emails all the time from young people. Oh, they said they've read your paper and they wrote your paper and they work their lot. But of course, you don't even know, right? They may very well of. It's possible they've read. your paper. Right.

Right. It's quite possible they haven't. And so yeah, it's very, you know, we used to do, I mean, they're like, it's not even clear what you, how you evaluate. Like, you know, my, you know, my son who was finishing college and he, he would spend time with Claude and like get, dig into something and then write a piece of code and then create a presentation and then you would send it to a firm and say, you know, I'm interested. And, and I said, you know what, that's not going to work.

They're just going to think. that slop. Like, it may very well be very thoughtful and deeply research, but they'll just like, oh my God, 20 pages of this, this must be AI, you know, delete. So it's not clear how you break through. So I do think, I do think that's a problem. Like, let me clear. I don't want to say that's the only challenge. The U.S. I think we have, you know, we have high tariffs. We have huge reductions in labor services to, you know, to health care, to construction, you know, to immigration intensive services. We have whipsawing energy policies, right? We have, you know, multiple wars of choice, have the Federal Reserve Bank under, you know, political attack. Our universities are, like my own

are really suffering under the Trump administration. So there are many, many reasons why businesses might feel cautious or uncertain. And if you're making investments, if you're making, you know, long-term investments in people or an equipment and plant, you know, you may just say, well, we'll just wait this out. Right. until we get some clarity. But that's just a guess. I can't, I can't prove that's the interesting. So you're thinking that there is a meaningful impact of, of this on hiring and therefore on the labor market. You know, again, it's possible, right? I have, you know, Yeah, you're working on.

Sorry, Nick Bloom of Stanford has done a lot of work on uncertainty and the effect it has on business investment in one. And I think the evidence is pretty strong that periods of economic uncertainty have a cost. in terms of they just, how can you make investment decisions? How can you make doing long-term planning if the ground is shifting under your feet? And so it makes firms sort of weight. Yeah, totally. But going back to your point about AI being used by applicants, that's freezing up businesses because they can't distinguish between applicants.

I think, you know, again, I have students are working on this. And that's the evidence that they, you know, we're seeing so far. And, you know, they'll be on the market this. this winner, so they'll tell you more. I don't want to get too far ahead of them. But it also, I think, makes businesses want to hire more established and credentialed people over younger people, right? Because once you have a track record, then we know. Right.

So if you spent a few years at a well-known business and you got promoted, then that's a hard signal, right? We can trust that. But until that time, you know, we just, it's very difficult to, evaluate. There's just to, and you know, you can't, it's not, when you previously you could have had your application management system, like sort through the resumes and look for the, then you say, well, no, I don't really trust me in that information of what do I use. People use LinkedIn, right, that has some hard credentials. They say, well, what university did you go to, right, which is actually, you know, unfortunate because it will also make employers more risk-averse, right, or more cautious.

They'll put more weight on the established credentials. Well, you know, I know you went to Brown, you know, they don't let idiots in. You must be O'Bay, right? And that's great for the Brown kids, but that's not great for the many, many, many other capable kids who didn't go to Brown. Yeah, yeah, got it. Hey, I might be pushing too much here

on the most recent data, but are you been following this decline in labor force participation? There's since the end of last year and even since the beginning of this year, if you're worried about the population controls in January, we've seen a pretty sharp decline. And I'm sure a fair share of its measurement, but it feels like there's a lot more, which each passing month and each month we get another decline and it's broadening out across demographic.

It feels like something else is going on here. Yeah, I'm aware of it, but I don't know the cause. I mean, I haven't, and I don't know, you know, obviously we have an aging population, so there would be some natural rate decline. But I don't know if that, what percentage of the phenomenon can be attributed to that. Got it. Hey, before we move on, Chris, let me and Marisselma, let me bring you guys in. Anything you want to ask, David, about the labor market before we move on to just AI more explicitly? Chris?

No, I think the theory really fits the data well, right? Recent college graduates have much higher unemployment rate, right? They're doing better than workers without college degrees. That's right. That's right. But just by historical comparisons, they're not doing as much better as college graduates typically do. Yeah. Versa, anything you want to weigh in on here?

No? You want to move on? No, no, I can't wait to get more into the AI. Okay. Hey, so let's talk about AI. And before we kind of dive into the economics of it all, how do you use AI, David? What is it, is it like part of your everyday workflow? I mean, it is.

Okay. Yeah, yeah. I mean, I don't do as much software code writing as I used to, right? So it doesn't like, it's not as transformative for me as it is for my pre-doctoral research fellows or for my grad students, but I use it all the time for ideation. I use it for, you know, proofreading. I use it for, you know, kind of extracting information readily. So it's definitely like a co-pilot for many things I do. And, you know, I use it a lot. I do a lot of writing. And I like writing, but I'm a very slow writer. I can't. That's why I don't, I've never written a book.

Or not really. I have my names on some books, but I wouldn't even say that I wrote them. But the, like one's a report that got turned into a book. But it doesn't speed me up as a writer, but it does help me refine what I do. So like I write a paragraph and then I say, okay, Claude, you know, get all the air out of this thing, like compress it. And then I have a conversation with it. And so I do feel like, and it catches like, you know, in a hundred page paper you're writing, right? You're going to have inconsistencies. You're going to have things that don't line up from what and it'll catch that.

So I feel like it improves the quality of what I'm doing. And then when I'm thinking about a new idea, I will, you know, interact with AI. And often it'll bring up novel sources that I'm not aware of or, you know, just it's a great conversation partner for that. Those are the main ways. I mean, again, it would be much more intensive if I were writing code. Do you, we're having this debate internally about how. much we should use AI for our analysis and writing. I mean, when you get something written, let's say from a student or a colleague,

and it feels like it's AI driven, how do you think about that? Is that, is that, does annoy you? Yeah, yeah, yeah. Yeah, I know. Yes. I mean, I think there's a, I think we all, like, you know, you read something and you, you start looking for AI tells right away. Like, is this them or is this, you know, is. And, you know, again, it's fine to use AI to refine what you do.

But if your mind is not in that content, then that's not, certainly I don't want to, you know, I have less credibility for me. And I don't, certainly if it's a student, I don't want to evaluate what their AI did. Right. I mean, I tell students to use AI to improve the writing. I say it often. But writing is hard. And, you know, we have a lot of international students as well, English not their first language, right? So, but I think the ideas, I want to see them coming from those people.

And I've not yet seen a case of where, you know, AI has produced a brilliant new economic insight. I think it's great for advancing ideas quickly. But it's not, at the moment at least, it's not asking the questions. So so far you think AI is kind of enhancing your expertise. It's not undermining it in any way. No, I don't think so. But let me be clear that, you know, I think it's very important to understand that the, how the technology is used and for what can have, like, very different effects, even the same tool, right?

So I, you know, give the example of, you know, I've had the same assistant for more than two decades. And she's a terrific writer. She actually has her underground degree in English from Brown. And she does a lot of editing and careful proofreading and so on. And, and, you know, now on the week. weekends, I fire up Claude. And Claude does those things with me. And that's super productive for me. It's helpful. I can do instant turn around. And so for me, this is like just a force multiplier. Now, this won't affect her. But the next person who takes her job, right, you know, if she ever retires, will not need the same skill set. Right. So we're both exposed to this technology. It's doing the same task, right? But for me, it's basically doing a supporting task, right, that allows me to, specialize even more in my comparative advantage, which, you know, hopefully is ideation, is research and so on and then. But for the person in that role, that's one of the

most expert things that she does. And it is being commodified in that sense. So I don't want to, so it's important to say, to recognize that the technology has different implications for different jobs. And it's not how we're both exposed. It wouldn't be, it wouldn't be sufficient to say, oh, you're both exposed to claw. sure, but in completely different ways. Right. So, and I think that's what one should be asking, not will it, you know, displace whole jobs. In most cases, it will not.

But, you know, whether it will, if you think about the bundle of things you do, will the technology take the sort of the busy work and speed it up so you can focus on what you're really good at? Or will it take the most expert thing you do, the thing that really makes you valuable and turn into it. into commodity so anyone can do it, you know, them and their agent. And then that, you know, you may still be working, but you're not your laborism is valuable in that case, right? That's specialized expertise that you possess, also in everyone possesses. Right. So, so kind of expanding on that from a macro kind of perspective, you know, there's obviously a lot of dystopic views that have been expressed, largely not from economists, but mostly from, folks in the AI world, although that's changing too.

I just recently saw Sam Altman and he admitted that he got it wrong about the job destruction, at least up to this point in time. This point in time, yeah. But generally the AI leadership, the technologists are pretty pessimistic about what all this means for jobs going forward. The economists are more, what's the right word? Bullheaded. Is that what you said? Levelheaded.

Okay. No, I mean, you know, we've been studying this for a couple, a couple centuries. So, you know, we have some perspective. In a sense, economists, you know, people have said for a long time that technology, machinery will take away all work. And so far that hasn't come to pass. So we have some reason for skepticism. But that's not to say that there's no risks, right? Even when it doesn't take away all of work, it can be extremely.

disruptive. So if you think about the industrial revolution, the power loom, you know, all those early weaving machines, right, they really did have the effect of taking artisanal skill, hard one, valuable, scarce, and eliminating the need for it. All of a sudden, what, you know, artisans had trained their lives to do could be done by machines, indentured children, and unmarried women, working in dirty, dangerous, polluted conditions, you know, at really low the rates of pay. And so, you know, that way, it didn't eliminate all the jobs, but, you know, it took six decades for working class wages in England to start to recover from that, from the first industrial revolution. Eventually, things changed both because of political forces and also because the technology became

more expertise intensive. All of a sudden, you need literacy and numeracy and so on. on. But that was a very difficult transition. So, and so the focus on running out of work is, is, is, is probably not the right thing to be looking at. We should be looking at the value of labor, right? Not how many people are working. You know, in poor countries, there's a lot of work to do, but people earn, you know, people can't afford to need not be working, but they're earning very little. And so a concern should be, you know, what is happening to the, the value of labor and even what is happening to the share of labor and national income. Right. So your sense is that the technologists, the AI leaders are overstating the case from a

macro perspective. Yes, there's going to be a lot of churn over here, you know, because jobs are going to be upended over here, but there's going to be job creation over there. And then of all that, if history is any guide or a reasonable guide, kind of nets out to work. okay. I mean, at least in terms of an aggregate macro, in an aggregate macro perspective. Yeah. Let me make a few points on that. Yeah. One is, I think people are getting way ahead of themselves in saying, you know, because we have this tool, everything will 20% of white collar jobs will be gone in short order. It just doesn't work that fast, right? And that's, you know, why is that for a number of reasons. One is jobs are much more complex than, you know, a prompt

in the result. They are political, they're interpersonal, they involve lots and lots of contextual knowledge, they involve relationships, they involve high stakes, judgments, and so on. And so people's jobs are a bundle of things, not just one thing. And it is, you know, we've never really, we had very few technologies that could just come in and you fire a worker and you hire a machine and you're done, right? Like, that, you know, the progress progress in robotics, why has robotics been, why don't we see robots everywhere? We've had them for decades, right? Because it's difficult to replace, you know, like an assembly line. You don't fire a worker and hire a robot.

You have to re-engineer the assembly line to use a robot, right? Now, this will change as robots become more dexterous, more humanoid and so on. Or even thinking, you know, self-driving cars, right? Ten years ago, they were going to be here any day. We'd all, you know, none of us were driving. And even now, how incredibly painstakingly slow the rollout has been. and that's after billions of investments, right? And, you know, and billions of miles.

And it's still the case that they don't handle edge case as well, right? They're city by city, and then they do crazy stuff, right? The other day, it was in the middle of a fireworks show in San Francisco going to, just drove right over a live firework and caught fire. Something he probably wouldn't do. So, and I think this, so, you know, it usually takes a long time to figure out how to use a technology well and fully integrate it. So it doesn't mean that AI won't have a very large transformative effect on labor markets. It just was never realistic to expect it to happen in a few

years. The second thing I'd add to that is when most people think about how the technology will be used, what they have in mind is it'll automate stuff, right? It'll just, but many technologies that are incredibly important technologies are not automation technologies. They're technologies that you collaborate with that use your expertise. So, you know, like a stethoscope or chainsaw, right? You know, a stethoscope, no use to me, right? Good for a doctor, chainsaw, right? Dangerous for many, helpful to some. They, they're technologies that leverage your knowledge, right? An automation technology means anyone who presses the button gets a good result, right? An automatic transmission, a dishwasher, a toll, total checkout machine, right? You don't need the expertise anymore.

And so it really matters because an automation technology really does, it at least displaces the labor that's involved in that task. A collaboration technology often demands, makes the labor more valuable, right? You can do more with a given amount of expertise. Now ask yourself, what type of technology is AI primarily? Is it an automation technology or is a collaboration technology? Well, if you think about it, it's actually kind of a crappy automation technology. And the reason is automation technologies need to be very predictable, reliable, robust.

They always do the same thing, right? You know, Excel doesn't make mistakes in calculations, right? You know, it's going to give you the same number over and over again, right? Your word processor isn't going to forget how to spell. But AI is actually not that predictable. It's, you know, it's stochastic. It changes from day to day. And if you, the minor details of the prompt can cause like very unpredictable results

the models are evolving. So it's not the kind of thing that you just want to use, let it do its work unsupervised, right? We know actually, actually, it's not even just unpredictable. Sometimes it's malicious. And so it's a technology.

It's incredibly useful. It's great for collaborating. You have knowledge. You're watching what it does. You're interacting with it. It produces results. You filter those. You ask another question, so on.

But the notion of just turning things over to AI and automating them is not that appealing in many cases. So I think, you know, the people have mistaken the idea that AI is powerful and can do human-like things with the idea that it can be autonomous. And I don't think it's a very good technology for that. So, which is, you know, which is positive negative. Like, you know, it would be convenient if there are a lot of things that could just do, you know, all by itself. Or the other thing from the perspective of work, we actually should be glad that it's complementary to people. And it needs specialized knowledge and expertise, and it can enable people to do more and do it more efficiently and do it better.

But it will, in many cases, be doing it collaboratively with workers. So interesting. So just to, and here I might be pushing too hard again, but I'm going to give it a shot. So it sounds like you're almost completely dismissing the dystopic kind of scenarios where this thing generates big-time productivity gains very quickly. You lose a lot of jobs. we can't adjust and, you know, we have unemployment and all kinds of problems. Kind of sort of saying, okay, that's a scenario, but it feels like a pretty low probability scenario in the context of what you said.

So let's go to the- Can I respond to that? Yes. Go ahead. So I- You're the guess. You can do whatever you want. Go ahead.

I don't want to say it's not going to displace any jobs. No, no. It certainly will. And I'm not going to say, and, you know, there are certain types of expertise that it will commodify, right? Like people, you know, used to have medical transcriptness. Those are just going on. I think when you think about call center workers, right,

there are more call center workers in the United States than there are truck drivers. They're mostly women. They're much less well paid than truck drivers, but people are always talking about truck drivers losing their jobs to autonomous vehicles. That's going to take decades. Call center work could go away very quickly, right?

It's that there's no installed base of, you know, billions of dollars of capital that we're going to scrap, right? It's just low, hanging, low, slung buildings and, you know, low rent areas. call center work, I think, we will see substantial reductions in employment in that area in all likelihood. The people who remain will be more specialized. They'll be higher paid. They'll be higher skilled, right?

They'll be at the top of the stack. Same with language translators. I think we'll have fewer of them in the long run. Those who remain will be doing like legal documents. They'll be doing diplomatic work and so on. But those will be areas where a lot of, you know, good work will be commodified. So I don't want to say that there's none of them. that. But I don't think we're heading for a world of like anything close to complete automation

in most settings. I don't think there are that many jobs that AI can just do autonomously. And I think there will be new work creation. And I think some ways it will make human expertise more valuable. In the best scenario, it will also open opportunity for people. It will enable people to do, you know, high value work, high stakes work, with less formalized training because they will have better supporting tools, right? So people who are doing software or sometimes of health care delivery or sometimes of legal work or kitchen design or even skill repair, the better tools that enable them to be more effective. You know, we have, our economy is dominated by high paid highly educated experts, right? In law, in medicine, in academia, right, in engineering,

in architecture, software. And, you know, those people are, you know, president and company accepted. They're worth a lot. They're worth what they're, you know, they're paid. I'm not saying they don't do good work. I'm not saying they're not valuable, but they're not that productive. They're productivity is and growing and their cost is. And if we could have tools that enabled people without as much formalized, you know, without not anyone can do anything, but, you know, Instead of having 10 years of medical education, you would have five or six, and then better tools that would enable, would give you both, you know, guidance. This is how you could do this procedure, do this procedure, and guardrails that say,

you know, that, you know, give you help ensure, you know, safety, reliability, you know, not making high-stakes mistakes. So I think there's potential for AI to be actually useful, to, or useful to people to enter new opportunities, not just to be pushed out of them. Got it. So let me put some numbers on it, just so I get a clearer sense of where your mind is, is, you know, if I look at average annual productivity growth since World War II, it's about 2% per annum, pretty much on the nose. Yeah. And there's been a couple periods when it's been meaningfully higher that on a consistent basis.

One of that was then around the Internet, Y2K, second half of the 90s early aughts, we were growing through. 3% per annum right uh and uh if you look at the kind of the uh people who do work economists that do work now i'm kind of saying let's put the AI leadership over here there in a different world we're now in the world of economists you get estimates of the productivity gains that are pretty consistent with what happened during the internet period that 10-year period 19 or mid-90s to the er mid-ohs and then you got a whole spectrum you can go the way down to Asimoloogu Asimu. Sorry, I do that every time.

Yeah, that's okay. Uh, he would, his feelings wouldn't be hurt. Yeah, apologize for me because I know you. No, no, look, none of us can actually pronounce his name the way it correctly pronounced in, in, in, certainly not quickly. And he's basically saying, at least last time I saw him opine on this, basically nothing, you know, very little kind of left. Yeah. Uh, so, in, in that, with that kind of a frame, where would you kind of sort of land? closer to zero or closer to one or something different than that?

So I mentioned I'm not a macro economist, right? Yeah, yeah. I'm pushing you in that direction because I want to. I think AI will be productive. Okay. An additive way, not a business as usual way, but kind of like an additive way. In additive way, right? So it's actually, you raise an important point.

To keep maintaining 2% productivity growth, we have to keep innovating. Yeah. It's not like that's just the natural rate of improvement, right? that's like strenuous effort gets you that. So if we didn't have the internet, if we didn't have electrification, if we didn't have, you know,

it's not like, you know, productivity would have slowed down a lot. So it requires a process of continual innovation. And the richer you get and the more advanced you, it's harder and harder to get another 2% out of that, right? Like we're getting a lot of juice out of what we got. But I do think AI will be additive to that. And the reason is because it brings new opportunities

for productivity growth into sectors where we don't have good tools, right? Where we kind of, you know, we don't have, you know, we had computerization. It's great for information processing and following routine, you know, following rules. It's also great in production settings where, you know, you're doing repetitive actions.

But it wasn't good for writing. It wasn't good for diagnosis, right? It wasn't good for research. It wasn't good for, you know, you know, laying out design and visualizing. it wasn't good, or it had, you know, I don't want to say it wasn't good, obviously, the pure design with some of that, but, you know, these are our, we, let me step back and try, try, like, zoom out a little more, right? Our consumption basket has shifted over decades from many more physical goods, like, you know,

food, manufacturer goods, right, larger houses, you know, televisions, the size of, you know, houses, and all that into much more, much more of our money is spent on expensive services, right? Healthcare, education, legal services, you know, design, contracting, and so on. And those things are intensive in educational, highly paid labor, right, which is scarce and expensive and don't have really good tools for getting much more productive. They're really bottlenecked. And, you know, this is Balmels disease, right?

I'm sure you recognize the description. But it's really, it's because. come, you know, it doesn't matter how much more productivity growth we have in automobiles. It can't be that important because we just don't spend that much money on them. Right. So you would much rather get 2% productivity growth in health care, which is, you know, 18% of GDP than automobile that are probably, I don't know, 1% of GDP. So this is a set of tools that are applicable in the places where we're bottlenecked, we're spending more and more money. So I think that's why there's leverage.

Now, of course, there's, you know, there's many downsides. There's disruption. and also, you know, AI will also impose its own costs, right? We'll use a lot of AI, defend ourselves against AI, right? Against, you know, malicious actors, against hacking, against, you know, misinformation, against, you know, all kinds of, you know, new possibilities for mischief. So, but I think there's great potential. And so I'm, you know, again, all of us are speculating.

Well, that's what we do for a living. it's okay. Yeah, I think, I think that, you know, I mean, people were very skeptical of the computer error for a while too, right? You know, my, you know, my beloved colleague, you know, Bob Solo said, you know, you see computers everywhere, but in the proctum numbers. Eventually they showed up, right? And I think AI will show up, and I don't even think it will take as many years. Okay, so if I think about productivity growth kind of baseline since we're over to 2% per annum, additive 0 to 1 percentage point on top of that.

That's kind of the range of economist perspective. It sounds like you're kind of sort of in the middle of that range. Sure. Just to give you a cent. Yeah, yeah. I think a half percent is reasonable. Half a percent is reasonable. Sure.

But that's a lot. Right? That's a really meaningful number. Yeah, it's a big deal. Yeah. I mean, you know, we don't, you know, it is remarkable. You know, and Chad Jones points this out, you know, if you just look over the last century, right, in plot our productivity growth in logarithic space, right?

Percent change. It's a line, right? It's amazing. Yeah. I mean, there are blips like, you know, during the Great Depression during War II, right? There's a little accelerated during the internet era. And, you know, but it's really, it's so, you'd almost think it was a law, but it's not, because most of human history didn't look anything like that, right?

It's all very recent. And to keep it going is an accomplishment. This is, you know, you've got to keep running just to stay in place if you want to keep go at 2%. So you have to believe, so how do you get something big? It has to be something that's very broadly applicable, right? It's got to apply to a lot of stuff. So it has to be a general purpose technology. And it has to do something, uh, I would say it has to do more than just do something a little better. It has to do something new. And that's actually what, what makes AI so interesting and also so hard to break. It, it's not a cheaper, better,

faster version of something we already have, right? It is just not, it's not a better. It's not a spreadsheet, right? You know, you don't want a spreadsheet that hallucinates. It's not a, it's not a better version of something that we're familiar with. It's a really different technology, right? The computers that we're familiar with, they follow rules, you know, they follow rules doggedly. They never improvise. They never deviate. They don't have a new idea, right? They just do what they're told to you, and that's great. That makes them very critical, makes them great automation devices. But they, that means it's not applicable. They're not applicable to many of the things that we do about, you know, about writing, about persuasion, about organizing, about making decisions

with many, you know, kind of, you know, incomplete, imperfect, and, you know, impressionistic variables. And so we didn't have tools for that. And now we have these tools that can do those things, but we don't know how to use them well. And we don't know how when we're misusing them. And we don't know how to integrate that into our production line, into whatever we're doing, because it just doesn't, it's not plug and play. It doesn't drop right into something we're doing. We have to reorganize ourselves to figure out how to use well.

And organizations as well, right? They have to say, well, what do young people do now, right? If we don't have them doing the busy work, what is the work that we have them doing? How do we leverage them effectively? We can't have a corporations that never hire people again. You know, we can't just ride it all down and age into the sunset. So how do, what's the workflow?

How do we train people? How do we mentor people? So I think there's a ton to learn. And we're such in just the early days, right? You know, we really just, you know, we just crawl out of the sea. Yeah, yeah. I mean, I'm very sympathetic to kind of where you're landing.

But if you talk to the, you know, it depends on which AI podcast I'm listening to, determine how depressed I get. But if you listen to those guys, the technologists, you know, what they say, what their retort is, you guys, you, the economists just don't understand the capabilities of this thing. And we're going to talk about recursive learning, you know, this thing's going to start educating itself without us. And it will solve the kind of the weak links and other issues, the frictions that you're talking about pretty much more quickly than you guys think. I know that's not history.

That's not the history of the Internet or the computer or what. wireless or whatever, but this is a different history. Sure. That sounds really right. Look, look, you know, there's no economic law that says, you know, productivity can't grow much faster or that we can't have machines that displace people, right? So we can't be certain.

I think the, I actually think it's very helpful at this point that everyone has taken a breather. I would say over the last few months and sort of, you know, taking a deep breath, looked around and said, wow, you know, the Earth is still spinning on its axis, right? Things have not been nearly as fast as we predicted. And so that should give everyone a little bit of humility about, hey, maybe we don't, maybe we're just don't appreciate, like, how hard this is. Yeah. And how much has to change.

And I, you know, I think, so yeah, I think I, my view, well, I don't, I don't want to overstate my view, but I think for a long time, AI is going to be much more of a collaboration technology than not amazing. Interesting. And if you want to annoy those guys. Yeah, how do I do that? Just a, isn't recursive

self-improvement redundant? Isn't it just self-improvement? Like, it implies. Yeah, good, good point. That's a great point. That's right. So, you know, it's not that smart if you're still saying that.

Yeah, how smart can you really be? Where did you get that from? Yeah, why you're repeating yourself. Yeah, why you're repeating yourself? Oh, that's a good one. I'm definitely going to use it. Hey, I got one quick. quick theory I want to test out on you.

And then maybe we talk a little bit about some of the policy ideas you have, just in case we do see a lot of unemployment. Oh, yeah, we should have. There are policies we should implement that we won't regret, even if we don't have a lot of money. Yeah, exactly. But here's the quick theory. I don't know if this is novel, probably isn't,

but I just, it's my own experience that going back to the Internet, because that's when I started a company and economics from that I sold in movies. And it was the computer that was the technology that allowed me to, And, you know, the fact that we can get off mainframes, we had interstate banking, we could, we could, because I could start a company. But my, my kind of theory is that these technologies like AI don't really take off until new businesses form and optimize around the technology. Because legacy companies, talk about frictions. I mean, it's like, you know, we can tell you all about the frictions because we live in that world. It's hard to do. It's very hard to do. And it's the only way you can start. Novo with that technology, things really, but that by definition means it's a slow process

because it takes time for these businesses to form. Does that sound? No, I think that makes it a time of sense. I mean, you know, what makes new technology is transformative is not the better, cheaper, faster. It's they, they enable some new capability that we didn't have before and it takes a long time to figure out what that is, right? So, you know, my friend Eric Bernjolfs likes to say, like, if you went back to ancient Greece, right, and automated everything they did, right, you wouldn't have a modern city, right? You just have ancient Greece without horses. It wouldn't have it wouldn't have telecommunications. It wouldn't have penicillac. Good point. Like what makes a technology transformer is it enables capabilities, not just automates existing activities. And you and yeah,

so you say this, you know, so the parallel is we need new business that do new things, right? So for, you know, Walmart forever didn't have a viable competitor. And then eventually I had to a competitor in Amazon. But Amazon is not another Walmart, right? It was like, it was a new business model. It's a new way of delivering services and goods, right? That was based around the internet. It was based around, you know, fast delivery. It was a totally different business model. And so, yeah, I think the innovations often come in the form of some, you know,

really novel way that harnesses of doing something that harnesses capabilities that just weren't present earlier. And it takes a long time to figure out what the other. You know, you discover that you have the thing and then you say, well, what is this for? Right? We know, you've heard many times, the story of how electrification. It took so long for it to create productivity growth because people didn't, they just wanted to just take out their steam engines and replace them with electric motors, not realizing, oh, wow, this can, we can have small, precise motors, we don't need to have them on a central

pulley system, just little wires, you know, we can use lighting. So it takes a while to figure out what is the technology really applicable to. And that's good. Like if there were no further progress in AI for the next 20 years, we'd still have two decades of great innovation to do with this tool that we have. Hopefully it will become more energy efficient. And hopefully it won't become more dangerous. Yeah.

But there's the sort of the capability overhang that we're living with right now. What is possible versus what we're doing is, you know, it's kind of awesome. It suggests there's a lot of potential. So it feels like almost under any scenario dead ahead, there's going to be significant dislocations in the labor market for specific jobs. Yeah. Even if it, in a macro sense, doesn't end up causing unemployment to rise in a meaningful way. Still, you got a lot of stuff going on underneath the hood that people are struggling with. And we can see what kind of damage that can actually do.

you go, all the work you did on China coming into the world, uh, trade organization back in the early 2000s and all the damage that that did to the manufacturing base. And I think we lost five, six million jobs and, you know, kind of a decade, you know, during that period. Yes. No, you wouldn't say, it's not clear that all that was due to the China trade shop, you know, some substantial number was and it happened so fast because of that trade talk. So, you know, I do think it, under, under any scenario, we need to be thinking about, well, what kind of policy

policy should we have in place or how should we change the kind of policies and safety network that we have, safety net that we have in place to potentially address these dislocations.

And you've done a lot of work there as well. I was curious, of all the things you've thought about, what is the single most important step you would take to help address these concerns? Sure. So let me just first reinforce point

you're making that, you know, labor markets, don't adapt that quickly, right? The rate of change of labor market is sort of the rate of refreshment of retirement reentry, right? So people's careers, you know,

let's say the 35 years, that means, you know, you lose, you know, whatever, 2.5% of people at cohort each year. And so it's easy for a business to, you know, shed 2.5% of its workers in a year without letting go of anyone, without firing anyone. And young people enter new things.

But when all of a sudden, you know, you have specialized expertise in manufacturing, right, or as telephone operator, whatever, and that whole opportunity set goes away, you're not, what's available to you next is not as good, right? You're doing the highest paid thing you know how to do. It's specialized as expert. It's scarce. And all of a sudden, it's not needed anymore. It's difficult for adults to make transitions. They're generally either means retraining or taking a big pay cut. And retraining is hard and taking a big pay cut is hard. Psychically hard. It's damaging. And people will say, I'm not going to, I'm worth

than that. I'm not going back for that kind of rate of pay. So one of the policies that I think deserves, you know, tremendous experimentation effort is what's called wage insurance. And this is something actually was done as part of the Trade Adjustment Assistance Act experimentally in the 2010s. So wage insurance is basically addresses the problem that when people lose jobs, they often cannot find an equally high paying alternative because they're not just, it's not just their company or their specific role, but like the whole activity set is just not in demand anymore. And so people will understandably say, well, I'm just going to keep looking until I find something, you know, that's about equally good. And that often isn't available. And then the more time you spend out

the labor force, the harder it is to get back in. So wage insurance says, hey, we get this. Like, we're going to meet you halfway. We will pay half the difference between your old full-time job and your new full-time job for finite a period of time, for finite amount of money. So let's say, you know, $8,000 over up to 24 months. And so just go back to work, take a new job at a lower rate of pay. We'll make a path of difference. And keep looking, right? Keep looking for a better job.

And the evidence is from, you know, this sort of policy experiment in the 2010s that it got people back to work substantially faster, so much so that it paid for itself in terms of reduced unemployment insurance payments and increased tax revenue. And so, you know, what makes wage insurance different from unemployment insurance? Unemployment insurance gives you benefits when you're not working. Wage insurance gives you benefits when you go back to work. Now, we don't know how well this would work at scale, but it deserves experimentation, investment. And there are people working on this and have received large, you know, support to do that, but it's still politically very difficult to pull this off. And one reason I think is a good policy More than just, you know, more than, you know, appeals to me

The evidence is sort of favorable. But also, I think it's politically viable. You know, America is okay with helping people who are working. It's very stinty with people who are not thinking. It's a more, it's, you know, we, the degree to which we moralize about non-employment to the United States is shocking to many Europeans. You know, a lot of people would like you working and they're not, but it's not necessarily moral failing.

So anyway, but you could say it's political. politically viable to say, hey, we're going to spend money to get, help people get back to work, and it's in the long run, it'll have some savings, and it'll help people preserve their dignity and make these transitions. I think it's much more viable than say, let's pay them to retire early, let's pay them longer unemployment insurance benefits. You know, so a work support, it's like the earned income tax credit. Yeah. But for a different rate. So I think that that's, if I had to push one policy. Now, and obviously, look, I'd be in favor of this policy. without AI, right? We've been paying it 10 years ago. This is like what I would call no regrets policy. It's just a good idea. Even if we're disappointed, there's no jobs apocalypse. We'll still be glad to have this policy.

Are there case studies overseas? Do other countries have any form of wage insurance? You know, not that I'm aware. They have what's called short-time work. Yeah, right. We take a reduction in hours, but keep your job, and the government pays the difference. But that's quite different.

We tried that, too. experiment with that as well. A number of states have it. Almost no one loses it. That's right. But I think there, you know, that doesn't help if the firm's going out of business or if that line of work is kind of non-viable. So, you know, I think that wage, if you think there's going to be real change where people are really going to have to swap sectors, occupation, and so on, wage insurance would be a more appropriate policy in that case. Got it. Well, we've taken a lot of your time. Maybe, and I've monopolized, all of your time. Maybe if you don't, if you're okay and you have a few minutes, maybe I can

turn to Chris and Marissa and see if they have any areas they'd like to push on. Marissa, you want to go first? Anything you want to ask, David? Yeah, I have a list, so, but I'll, I knew you would. Pick one. Sorry. And I'm not, this isn't a push. I'm just curious, David, you're working with young people that are going to go into the labor market. What do you tell them? I mean, how do you think we should be thinking about education and preparing people for the next whatever it is, you know, five, ten years of what entry-level work may look like. Yeah. So I think young people are very worried. And, you know, because I do a lot of public speaking, I go to a lot of college campuses. And I talk to a lot of, you know, particularly undergraduate

students, sorry, particularly economics undergraduates, but not exclusively. And a lot of people feel like their future is being kind of like, you know, pulled out from undergraduate. to their feet. Like the thing that they were training to do, now they think a machine will do it, and they won't be needed. And I don't think that's correct by and large, but I certainly understand the anxiety. And I think that anxiety has a real cost, right? Even if, let's say it all turns out fine, they still might, you know, people are, it's not just going to make them unhappy in the short run. They may choose not to invest. They say, well, I don't need to learn that skill set. So I think that's a problem. I do think that there will be opportunities for people to

to do what people are good at, which is to basically, you know, translate between areas of, of, you know, formal knowledge and the actual high stakes personal situations that they're in, you know, whether that's in medicine, if it's law, if you're a contractor, if you're doing skill repair, there's people, you know, work, you learn a formal tool set, but you're making decisions in a much, much more high dimensional environment with other people. And the world is actually becoming an increasingly overwhelming place. The amount of information people are presented with is just, you know, it's limitless, but it's unreliable. And you need people who can make judgments, exercise leadership, convince others, plan, and decide. And so I think there is a lot

of room for that. But, you know, but exactly what that looks like. I don't know. And I would be, I would understand, I do understand the anxiety that people feel. So, you know, I'm optimistic, but I can't be concrete and say, you know, here's the future, go do plastics. Right. Mercer, do you know what he's referring to?

You're too young to know. No? With the plastics comment. That's the graduate. The famous Dustin Hoffman movie, the graduate. They're too young. You know, sons, I have one word for you. Plastics.

And probably right, too. Yeah, that's right. Someone had to go take a look. Now it would be plastic cleanup. Plastic cleanup. Plastic clean up. Yeah, yeah, exactly. Hey, Chris, you got the last question.

All right. So we've spent a lot of time talking about the risks of AI being too successful, meaning to mass job loss. That's certainly a lot of focus, as we've discussed. But I'd submit that a bigger risk is actually that AI falls short, given the demographic picture that we're. we have. I'm wondering if you agree with that thesis or not, right, given the demographics, I'd submit, without some technological change here, growth is going to slow appreciably.

I do think the demographics are something headwind, although I have to say, you know, I recently wrote a paper with Daron Asimoglu, Andrew Scott, and our student, Keelan Burnay. You pronounced his name right. I disclosed. Yeah. No, it's really not. It's like, It's like Oshmobile or something. It's something I can actually say. But Daron doesn't, thankfully, he has bigger things to worry about or to focus on. I got it. Arguing that actually, you know, the historical record is that when there's a decline

birth rates in countries that there's faster technological progress, that there's a lot of adaptation and innovation. And you could even argue AI is part of that. So we, you know, you don't see, what you see is you don't see a decline in GDP per capita when countries start to, when birth rates slow down, country. So I don't know, but I do think, yes, AI, what would be really helpful to us is productivity growth, right? We have aging populations. People are living a long time. They have earned the right to a healthy, comfortable retirement. If we don't learn to become more productive with this relatively small relative to the

size of our retired population workforce that we have, that's going to be a big burden. So if we could basically become more productive in our work and provide better tools and technologies that would support the, you know, the well-being of elderly citizens, that would be fantastic. So I do think AI is, you could argue in many ways, wow, it's a good technology at this time, right, because it can help us in, you know, some of the problems that we face. And I do think, you know, a lot of the challenges we face, you know, climate change, power generation, education, you know, you know, improving education. education, doing agriculture more efficiently and so on and taking care of the elderly, you know, AI could be very useful for that. I should, I should inject one note of pessimism since I've been so optimistic all along. The one area I am most... He's been optimistic. Is that, is that right? Has he been? I didn't say it's Dune, right? Most people who talk about AI,

they're Duneers. That's true. True. They like, yeah, I mean, I think it's shocking. Like, you know, here's this, you know, we've invented fire. We're going to burn. Very level-headed. But the, I am worried about the educational repercussions of AI. It's much too easy for people not to learn. And there is like really compelling evidence now from a paper by David Stromberg and co-authors about China, about just studying 40,000 middle and high schoolers in China starting to use AI for their homework.

And it used to be the more time you spent on homework, the better you did. With AI, it becomes the more, the less time you spend, the better you do, because you're just having a machine do it. And then what happens is people's learning to cheer rates. And this has very high stakes. This is not just on their next test. This is on the Gao Kau, which is the exam for getting into. And people are losing like 20%.

Like, that's the difference between going to Shinghua. university and going to like Guangzhou state right i mean it's like it's a huge cost and it's not that it's not that the i necessarily interfering with their learning it's that they are spending less time because they think they've learned what they haven't and so i am really concerned that ai will you know learning education is effortful if you're not breaking a mental sweat you're not you're not learning and i makes it way too easy to not to not to feel like you don't need to make that effort and so that's my that is actual concern is that it's not that the labor market won't be looking for capable young people is that we won't be producing as many if we don't sort of figure out a way to

maintain you know incentives for learning appropriately in education you know i think that applies to everybody not just kids in school i mean i i i have to fight to sit down and not rely too heavily on i mean it's it's like running i mean you know it's If I don't run, I'm not going to be as healthy. I certainly don't want to run. And, you know, this is the same thing. You've got to work. You got to put energy and effort into it.

And it's just so easy not to. And you got to fight that. Right, right. No, Claude gives you like pages and pages of explanation. You're like, come on, Plod, just give me the answer. Just, you know, never mind all that. Yeah, yeah. If you tell it to do it, it will.

It will do it. Write your answer in half the words you just wrote it. It'll come back and do it. Yeah. So, anyway. So, yeah. So that is a real, I think that is, but especially we got, you know, we are fortunate that we're getting to use AI, but we had to do all the heavy, hard work of, of, you know, of the education we had. And I, I, you know, that was beneficial.

Yeah, absolutely. Well, David, it was really an honor and a pleasure to have you on. Really enjoyed it. Learned a lot. And, you know, very much enjoy all the work that you do. It's just so insightful and very interesting all the time. So from a macroeconomic economist perspective. Keep it coming, please.

It really, really. Well, thank you. Thank you. So much for inviting me on this. It was really, it was a lot of fun to speak. And, you know, we all, we are living in interesting times. And, and that's these times.

They're crazy times. It's beyond interesting. Yeah. Yeah. Crazy. Craziness. Anyway, with that, dear listener, I hope you enjoyed the conversation. We are going to call it a podcast.

Take care now.