Plain English with Derek Thompson - The Job Market for Young Grads Is Flashing Red

Episode Date: May 6, 2025

Last month, the unemployment rate for recent college grads surged to nearly 6 percent. Compared to the overall economy's jobless rate, the unemployment rate for recent grads is higher now than in any month on record, going back at least four decades. Business school grads are struggling, too. Last year, The Wall Street Journal reported that elite MBA programs saw that their most recent classes had "worse job-placement outcomes last year than any other in recent memory." What’s going on? Today’s guest is David Deming, an economist at Harvard who studies education and the future of work. We walk through some plain-Jane theories about the weakening labor market for young college grads. But then, as you’ll hear, the conversation expands to consider the coming storm of AI—and how artificial intelligence is changing education and the workforce for young people. To read more on this topic, check out Derek's Atlantic column from last week. If you have questions, observations, or ideas for future episodes, email us at [e-mail]. Host: Derek Thompson Guest: David Deming Producer: Devon Baroldi Learn more about your ad choices. Visit podcastchoices.com/adchoices

All right, my birdie buddies, my car saving pals. My Eagle Enthusiast, it's Joe House here. Major season is finally upon us. The Masters, the PGA Championship, the U.S. Open, the Open Championship, and Fairway Rowan is here to break down all of the storylines. Offer a little help on those betting cards for every single major this golf season. Join me and our incomparable accomplice, our tour boots on the ground, Nathan Hubbard, as we guide you from Augusta all the way to Northern Ireland.

Royal Port Rush, away we go. Today, something very strange and potentially very worrying is happening to the job market for young people. Last month, the unemployment rate for recent college grads surged to nearly 6% compared to the overall economy, the unemployment rate for recent grads is higher now than any month on record going back at least four decades. Even newly minted MBAs from elite programs are struggling to find work. Here's the Wall Street Journal from last year. Quote, 23% of job-seeking Harvard MBAs who graduated last spring were still looking for work three months after leaving campus. Harvard isn't the only elite business school

where recent grads seem to be stumbling on their way into the job market more than a dozen top-tier MBA programs, including those at the University of Pennsylvania's Wharton School, Stanford's Graduate School of Business, and New York University's Stern School of Business had worse job placement outcomes last year than any other in recent memory.

End quote. Meanwhile, law school applications are surging. And while that might not initially sound like a particularly scary statistic, it is an ominous echo of 2007 and 2008 when thousands of young people used graduate school to bunker down during the Great Financial Crisis. So what's going on here? Today's guest is David Deming, an economist at Harvard who studies education and the future of work. First, we walk through a few plain-Jane theories for the weakening labor market for recent college grads,

But then as you'll hear, the conversation takes a turn. And we begin to consider another possibility. The possibility that we are at the beginning of an age when artificial intelligence will begin to eat into labor market opportunities for young people who are competing with tools like chat GPT to read and write and synthesize and research and make reports and answer questions. And so the rest of our conversation, hinges around this coming storm of AI, how artificial intelligence may be changing not only education, but also the workforce for young people, and how they and their parents and their teachers should respond.

I'm Derek Thompson. This is Plain English. David Debbing, welcome to the show. Great to be back with you, Derek. Thank you for having me. So the task before us today is to solve an economic mystery. according to the New York Fed, the labor market for young college graduates has, quote, deteriorated noticeably in the last three months. David, what do you think is the most plain vanilla explanation for why this is happening? Yeah, so I think there's two explanations that are plain vanilla. One of them, I would say, has some credence. I would say if you push me, this is the most likely explanation, and one which I think a lot

of people say, but isn't true. So the first explanation is where we are in the business cycle and where we've been over the past few years is dragging down employment for new college graduates much more than for other kinds of workers. And that's because if you think about why a company would hire a new college grad in the first place, it's because they think about young people as investments. You know, like you hire somebody, they don't really know how to do anything relevant on the job right now, but they're smart. They have a general toolkit. You know, they can write pretty well. They can read and synthesize information. They can communicate. And so you figure, you know, I'll train them up and in a few years, they'll be worth it, worth the investment. And so when times get really uncertain, as they have been over the past few months, businesses pull back on capital

investment. They are also going to pull back on investment in young college graduates because they think about them like capital. Like I hire a, you know, a lump of clay molded into something useful, a young college grad, and those are just risks and time horizons that I'm not willing to invest in right now. And so I think I think there really is something to that, Derek. There's been a longer-term trend in, you know, deteriorating conditions for young college grads that doesn't explain all of the last few months. But I think there's actually been a lot of economic uncertainty in the U.S. and around the world, going back to the pandemic and even a bit earlier, lots of things going on,

I think making businesses a little more hesitant to hire young college grads. So that's boring explanation number one that I think has some truth to it, a lot of truth to it. Boring explanation number two is that college doesn't pay anymore. And I just think that's not correct. I think if you look at the college wage premium, it is flat over the last few years, but that's mostly because wages are so high, which is a good story, so high for non-college grads. So it's not that college grads are doing worse. It's that people without a degree are doing very well.

The labor market is tighter than it's been in a long time. And that's really helped out workers at the bottom of the wage distribution much more. The other thing is that all college grads, so people who are not just starting out, but in their 30s and 40s, are still doing really, really well. And in general, the college wage premium, meaning the earnings premium you get for having a degree, actually increases, it almost doubles over the course of somebody's career. And so what we're really seeing in the labor market right now

is not bad conditions for college grads overall, just bad conditions for new college grads. So I don't really think that's the story either. I'm going to do one more beat on this question of, is college worth it? Because it's a question that I see in financial news sometimes, in op-eds sometimes. It's like a very popular hot take,

especially from, ironically, college graduate pundits who weirdly like to punch at their own side and suggest that the future will belong to people who skip college. And they're starting to reevaluate the degree to which college will matter at all in a world where electrician salaries

will far surpass, say, software programmers. One more time, just looking at the data that we need, know today, without being able to guess at the data that's available in 2030, is there any evidence suggesting that the value of a college degree has significantly deteriorated in the last few years? It has deteriorated only because it was such a good bargain in the 1990s and 2000s. It's not that it's a bad investment in an absolute sense. It just hasn't been as good as it was.

And that's actually, Derek, because many more people are going to do. college. So roughly, you know, between people ages of 25 to 29, so late 20s, the share of people with a bachelor's degree was 30% in the years right before the Great Recession. And it's now 40%. So we have a third more people getting degrees than we're getting them 20 years ago. And most of that increase has come from people attending the kinds of schools that have expanded a lot, which are usually less selective public universities. And so part of this is just a kind of, we're victims of our own success. We're sending many more people to college. Some of them maybe aren't learning as much as if they go to more selective schools. And that's dragging down the overall

college wage premium. But for any individual who's thinking about whether they should make this decision to go to college, it's still, by and large, a very good decision. It's not risk-free. Not everybody benefits. But if you think about it as like, is this an investment and what is the return on it? You're getting a return of, you know, that far exceeds returns in the stock market, returns on buying a house, returns on starting a business. It's just a good investment for people to make, and it pays off over the course of a lifetime. So that's still true. It was true 20 years ago, and it's still true today. Obviously, we never know what the future holds, but it's still true today. I was talking about this with some other economists as I was putting together this article,

and the word that I used to describe the college wage premium was stagnating. And the guy was not a third-up-a-conversation, but the guy said, yeah, don't say stagnating. A plateau in the Himalayas is not a stagnant point. It's a point of extremely high elevation over sea level. It's just not in the process of becoming Mount Everest. So when you use the word stagnating, it sounds like a pejorative. But when you think about the fact that the college wage premium is more or less stagnating or maybe declining a little bit from a really high level, people should think about the level

and not think about the rate, right? They don't have the opportunity to decide whether to go to college in like the 1980s, 1990s. The choices today, and right now the college wage premium suggests that for the most part, college pays off relative to no college. Just want to make sure that we buttoned that bit up before we move to the spicier interpretation. No, that's exactly right, Derek. And I would just add one more thing, which is that people often say things like, you know, people aren't, you know, there aren't very many jobs that require a college degree. Those are declining. Companies are not requiring them in job postings. And I think that's very misleading because, um, you see a lot of jobs where a college degree isn't formally required is required. But, um, even within all occupations, even if you're a plumber, actually having a college degree, plumbers with college degrees, are a lot more than plumbers without college degrees. And that's because a college degree isn't required to be a plumber. But knowing how to use technology, knowing how to, you know, implement some some ideas of starting a business. You know, most plumbers are independent contractors, and so being able to market yourself and talk to potential clients and organize your finances.

Those are all things you learn in college. You don't only learn them in college. But a college degree is still just a very general purpose toolkit that helps you in a lot of things, even if it's not, strictly speaking, a requirement. So I just think people are off base when they talk about it this way. I'm glad that we check the boxes of plain vanilla explanations under the general principle that when you hear galloping, you should think horses, not zebras. But the theory that I'm most interested in investigating with you is the possibility that rising unemployment for recent college graduates today might have something to do with the effect of generative AI in the labor force.

Or that continued rising unemployment among recent college graduates might speak to the fact that companies are using AI to do the work of young. recent college graduates. And the reason that I think this is a worthy question to evaluate is that when I think about what AI, like ChatGPT and Claude, is very good at doing. It's very good at reading, summarizing, writing reports that get some things wrong, going away and clicking around the internet, and then coming back and presenting some kind of written document. It's good at doing all of this very, very quickly and relatively cheaply. And when I think about the skills that are typically demanded of young college graduates that say law firms, marketing firms, consulting firms, tech companies, I think, man, it's a lot of reading and summarizing and writing

reports to get something's wrong and going away and clicking on the internet, et cetera, et cetera. Am I crazy for thinking that at a first principles level, generative AI is in fact quite competitive with some of the skills that are typically demanded of college graduates? at that entry-level white-collar position? I think that's right. I don't, with the caveat that it's too early to have actually seen that happen in the labor market yet,

because, you know, ChatGVT is only two years old, more or less, two-and-a-half years old, and lots of companies haven't quite made the changes that you would need to see something like this, you know, affect the unemployment rate. But I think from first principles, you're right that the kinds of things that we ask entry-level college grads to do

in white-collar office work are the kinds of things that you can do more easily with generative AI. And one indicator of that is that generative AI use is rampant on college campuses and in classes, you know, because the kinds of work, you write a 10-page memo, write a three-page memo, you know, that communicates some clear thing or make a clear argument using data and evidence. Those are kinds of things that we teach people in colleges. And, you know, if you talk to college professors or college administrators, you can see that people are struggling with how to deal with, you know, incredibly widespread usage of generative AI in classroom assignments and having

to change your assignments and trying, I think, mostly futilely trying to make your classrooms AI proof, that's exactly consistent with what you're saying, which is that this is kind of what we're teaching in college, and all of a sudden we've got a technology that's kind of on tap that can do it instead, and it wouldn't be surprising at all, to see that. Certainly, if nothing else, affect what companies are asking new college graduates to do, even if they still hire them, they might be asking them to do different things because we now have this technology that can do it instead more cheaply. Before I ask you about how you use generative AI and how you see it used within academia,

I do want to press forward this idea that generative AI might be a bit player, at least, in driving up the unemployment rate for recent college grads. For this to be true, for my hypothesis here to be true, you would need pretty clear evidence that tools like chat GPT were being implemented. and taken up by the labor force at a rate that's basically competitive with every previous technology in like the history of modern technology. Is that in fact true? Do we see a genuinely sharp rate of uptake among tools like Chachibit? We do. So in a recent paper, my co-authors and I conducted a nationally representative survey of generative AI usage. And we conducted the

survey in a way to make it comparable to some surveys that the current population survey had done about computer usage and internet usage in the 90s and the early 2000s. And so we kind of put it on the same scale and tried to ask, is the adoption of generative AI on the same scale as these other two technologies that turned out in retrospect to be quite fundamental to our modern way of life? And the answer is basically, yes, the adoption of AI, something like 35 to 40 percent of people say they use AI at least once a month, let's say, and about a quarter of workers say they used generative AI at work at least once in the last week. And that's a quite high rate of usage for a technology that's only been around two, two and a half, three years. And if you compare that to PC usage, what you find is

that generative AI usage is a little bit higher than PC usage at that same period of time. So we date the beginning of the PC to the release of the IBM personal computer. That was the first computer to sell more than a million units. We say that's the beginning of PCs. And the beginning of generative of AI is Chatsy BT in November of 2022. So we just kind of compare the rate of adoption for those two things. And adoption of AI is a little bit faster other than adoption of AI, sorry, of computers at work, but it's actually way faster outside of work. And that's because, you know, it's not an expensive piece of hardware.

So it's intentionally made very accessible by a bunch of companies. So maybe that's why, but still it just shows that, yeah, it's on the same scale of something that 20 years later in the case of computers was basically ubiquitous in our lives. And so that does suggest that AI is eventually, you know, within a couple of decades, going to be everywhere. You have the statistic in that paper. In 1984, 25% of workers reported using computers for their job compared to 27% of workers who report using generative AI for their job in 2024, which suggests in a way that we are today where computers were, personal computers were, exactly 40 years ago. And that can cut both ways. one way you can cut is that we are at the beginning of a truly transformative technological revolution.

Another implication that argues against my hypothesis is that look around. The computer did not suddenly disemployed tens of millions of people. The unemployment rate is about 4%. It's been extremely low for the last few years. And in many cases, you've seen this phenomenon that economists sometimes talk about is sometimes called Jevon's paradox, that when you drive down the cost of something, rather than replace demand, it actually grows demand. So that, for example, it was very, very difficult, I imagine, in the 1960s and 1970s to, like, make and manage spreadsheets. And then we made the making and the managing of spreadsheets with the invention of Excel very easy. And someone in my position in the 1980s or 1990s could have said, oh, well, now that Excel makes spreadsheets easier,

we're going to need like 10% of the spreadsheet workforce that we needed 10 years ago. and the future is just going to be like very, very few people being very productive with Excel. Instead, it's the opposite. Everybody is constantly using Excel in the white-collar workforce. So rather than disemployed tens of millions of people, it actually just outfitted tens of millions of people with Microsoft Excel for the better or the worst, depending on your particular opinions about the soul-sucking utility of Microsoft Excel. So to what extent does the story that I just told apply? fly to generative AI. Like, is this an Excel kind of technology? Or is this something else? Like, like the relationship between the tractor and the horse, where, like, the invention of the

internal combustion engine and the tractor didn't make horses more productive. It just entirely disemployed horses on the farm. So how do you think about the introduction of Genitive AI right now? Excel or horses? Yeah, it's a lot to unpack there, Derek. And it's a great question. I think you're asking exactly the right questions, at least from my perspective. So on the Excel thing, I think it's worth thinking about how that, you know, what did Excel replace? So Excel, you know, as you mentioned with Jevin's paradox, there's a lot more demand for spreadsheet analysis because we have this cheap tool. But it also replaced some, or at least diminished the importance of some categories of jobs. Like you used to hire an accountant who would write things down in a physical

ledger book. And so being able to do quick calculations accurately, you know, was like really valuable. And so these like white collar clerical back office jobs did suffer. You saw all their jobs like, you know, the way banks were fundamentally reorganized because of optical character recognition and spreadsheets. So instead of having somebody, you know, write down the checks and do all the ledgers and figure out whether it was right, you instead hired people who were more focused on customer support, more kind of teamwork. And then if you think about, for example, how computers and digital technology changed

manufacturing, you went from hiring a bunch of people to, like, physically cut a valve for a manufacturing use to programming a computer to do it by road. And then it became about trying to organize the shop floor and optimize production in a way that increased throughput and customizing jobs. So it became a much more analytical job and team-oriented job and much less of a job-oriented around physical labor. So the computer really did, and digital technology more generally, it really did change the labor market fundamentally, even if it didn't, you know, make us turn us into horses in your analogy. So I think that's likely to happen with AI. So I think AI will be tremendously disruptive in the labor market. even if it's not the kind of disruption that leads to us all being unemployed.

I personally, you know, I don't see any evidence yet. And you never know, like you can never predict the future. But this would be the first time in history. We've had a lot of really transformative technologies like electricity and steam power and computers. And none of them have reduced net employment. So maybe AI is different. But I kind of think it won't be. I think it'll be normal in that sense.

But it'll be tremendously disruptive in a sense that it's going to change a lot of what skills are valuable. and what work is more valuable and less valuable and for certain people will be will displace their jobs. But, I mean, it's kind of scary but also exciting at the same time. So I'm in the camp of thinking it's going to be a big deal, but not in a sense that we're all going to be unemployed,

but it's going to really reshuffle the winners and losers of the economy, so to speak. Let me tell you one thing that I'm worried about with generative AI. The horse, Excel, these technologies replaced certain kinds of tasks, or they accelerate, or made more efficient certain kinds of tasks.

With generative AI, I see this remarkable ability to accelerate or substitute for an entire suite of tasks, a bundle of tasks that is heavily concentrated in one part of the economy. That is the lowest rung of the corporate ladder at white-collar firms, right? that if I, for example, have a certain need for research for my work at the Atlantic or my work on the podcast. And there's a world in which I might demand or ask for, I mean, demand in the economic

sense, not demand and like me being rude and saying, you must work for me sense, a very demanding guy. I might ask for a personal assistant. But I have to say, deep research, which is a function by Chachapit is just so extraordinary at whipping up bespoke 10,000 word research documents in five minutes that I have to imagine that people like me are going to think I don't need to hire that extra person who's just going to do little research projects for me. I'm going to rely on deep research. And some of the deep research essays will be a little bit worse than a great research assistant. Some of them will be a little bit better than a deep, than a deep, than a deep research. a research assistant that's human, but they'll certainly be faster. And when I multiply that across

the economy, the fear that I have is that it's like we're lobbing off this piece of the bottom of the corporate ladder that might make it harder for certain types of workers to get their foot on that ladder, to get inside those companies where they can start to work their way up into jobs that are more demanding of sort of the full human suite of skills and also, by the way, higher paying. Are you also concerned that there's something sort of uniquely spooky about AI in terms of its ability to replace this entire suite of jobs that's so disproportionately concentrated at that lower part of the corporate ladder? I am concerned about what you mentioned, which is how do we kind of, if we're replacing entry-level white-color work with AI in the medium run, how do we train the

next generation of people who are going to be business leaders, decision makers, how are they going to learn to do the job if they're not learning the ropes at the bottom of the ladder, as you say? So I'm super concerned about that. However, I'm not sure it's really different than other technologies. If you think about farming, okay, there are still farmers in the U.S. There are many fewer of them, but actually it's a quite high-paid, high-skill occupation. So you're making really high-stakes decisions about how to use your land optimally, whether to use different technologies to optimize, you know, crop rotation and like, what market should you sell in and how should you hedge, you know, against the risk that you'll have a bad weather season in the futures market.

And there's all kinds of things you do that you basically replace the lower rung of farming, the manual labor of actually tilling and plowing the fields. And what you've got left is the higher order things that are left in the job. And so I do think, that's why I think it'll be disruptive because we'll see that what firms expect from white color office workers is just going to be ratcheted up very quickly. So it's like this kind of useful but relatively shallow function that deep research performs at the end of the day, which is just scanning more or less the content of the internet to synthesize a bunch of stuff and put it together for you. It feels really deep because that's what we've been teaching people. But in some sense, it's just taking everything

that's out there and putting a loss on it. And I know, I say that because I use deep research a lot. And when I use it, if I ask it to write a paper about something I know a lot about, so as a labor economist, or if I ask it to write about soft skills or higher education or whatever, it's undergraduate term paper good, but it's not really good. It's not as good as me, not because I'm amazing, but because I've spent my life thinking about these issues and I've read the papers as reading and I have some nuance. It's some sense, I have knowledge about that that is implicit, that's outside, that's not, that doesn't exist on the internet, how certain people talked about it, what the debates were. And all of that knowledge that I've embedded that I've encoded in my mind that's

not on the internet, I have access to and chat GPB doesn't. And so I still have an advantage over the, over the machine in the things I know really well. But the things I don't know well, like if we were going to do a podcast about, you know, the history of farming or something, I would use deep research and I would rely on it because it probably is going to get the details wrong, but it's still going to be better than my existing knowledge, which is pretty shallow. And so what I think you'll want to do in the future is you'll want to become deeply expert in something beyond whatever knowledge exists on the internet. You want to go talk to people.

You'll want to be there in the moment. See things that you can't, they can't just be captured by a tool like deep research. And then you'll want to leverage yourself, leverage that expertise by using the AI to do all the things you're not so expert in and can't afford to be. I want to go one level deeper on that point because I was talking about our conversation and the article that I wrote for the Atlantic with some bosses at the Atlantic. And they were talking about their kids who are currently in high school. And the question that was coming up was, what the hell do I teach my child to value in education if A and A plus term papers can be written by a machine? what do I teach them to value that will help them become fully fledged human beings, not to mention high-earning adults,

in a world where the unit of assessment for intelligence is now essentially synthetic in many cases. You can just dial it right up, and the A-plus paper that used to be the measure of your intelligence is now just the measure of your ability to ask a machine for that essay. So what, David, what should high schools and colleges be teaching now in order to give people's skills that they can actually use in the workforce rather than just essentially lie about their intelligence to teachers who themselves might be lying by creating their curricula on AI as well? Yeah, so let me give a like a practical answer and then what I hope is a deeper answer.

So the practical answer is I think there's a lot of sense in which, you can use AI to produce a term paper, and you can, you know, if I, there are a lot of papers I would assign in the pre-AI world that I just can't assign anymore in my classroom. But maybe what I was assigning the whole time was actually shallower than what I'd like. Rather, you want people to internalize the knowledge of the topic so deeply that they can, they can take two sides of an issue and explain both sides equally well so that you don't actually know what their position is. And so a lot of what we really want people to do is to be able to embody the knowledge and the expertise they've learned in school in order to use.

it in some particular way. And so I think what AI is doing is it's pushing us as educators to do the things that we should be doing anyway, which is engaging in deeper types of learning that allow people to deploy it in ways that we can't foresee that's more flexible. And so you might imagine in a classroom almost returning to an older tradition of oral education, you know, so we focus more on can you make a presentation. And that's kind of AI proof because you can't like type it in and really, you have to actually be able to do it in the moment and defend your position and convince people, persuade others,

And so I'm actually kind of excited, Derek, about where I think that will push us in higher education to focus more on people developing expertise they can use for real-world things. So that's my practical answer that I think it will push us to focus more on, yeah, on oral education and presentation. My deeper answer is that I really think it's going to push the frontiers of people's creativity. So this is an incredible tool. It's intelligence on tap. But intelligence is not the same thing as wisdom or judgment. And those are things that are about, okay, given I have these tools, I have this knowledge, what should I use it for? You know, how should I use this intelligence on tap to make the world a better place, to fulfill my objectives, to make money if that's what you want to do?

And all of those questions are extremely open-ended. If you go on chat to BT and you say, how should I become a millionaire? Like, it's going to give you some very, very stock answers that are not really going to help you in your situation, you know? Or if you ask it, like, how do I improve the world? You know, it's just not going to get you there. But it can help you get there if you know where you want to go or you have some sense of where you want to go. And so I think the future will belong to people who have really, really strong agency over what they want to do and the creativity to use these new tools to accomplish much more, leverage their expertise and leverage their judgments and their creativity and their agency to accomplish a lot. And so I'm kind of excited about that future there because I think that sounds fun and interesting.

And so I think what we want to do in school is get people engaged in that work, like get people excited about using these tools to go deeper. You're arguing that the bull case for AI improving carbon-based human intelligence is that we choose to engage with this technology through wisdom rather than through some instinct to achieve shortcuts all the time or to shortcut our own intelligence. What do we know from the data to tell us whether that's actually how people are using this technology? We don't know that much about it, but what we do know suggests they're not really using it that way. You know, I think a lot of people are using it to either in school, you know, students are using it to turn in work that wasn't their own. Or, I mean, maybe more hopefully to using it as a tutor to ask questions. The worry I have about that even is that it's easy. You know, it's kind of like when you ask a friend, you know, what did you get on this problem?

And they're like, oh, I got it because this, then you kind of feel like you understand it. But the understanding is shallow because he didn't struggle through it yourself. And then I think at work, I think people are basically using it to, not in a shady way, but just using it to represent their own work. So the boss says, you know, write me a memo, have it on my desk in two days, and you use chatypte to write it in two hours, and you know, get off work early, go to the bar or whatever. So I think that's a short run equilibrium, though.

I think eventually, certainly companies and hopefully educators will wise up and understand that everybody's using this all the time and start to demand more. And that's where I think the potential for, you know, using it in some of the hopeful ways that I outlined could come in time. David, have you done any work on the kinds of workers that are most likely to benefit from generative AI? Because one paradox I kind of observe is that AI seems to me to be most useful for below-average performers, people who struggle at their job that can lean on AI, internists and doctors

that aren't very good at diagnosis, but outfutable. with CHATGBT, they're actually totally A-minus diagnosers of disease. But it also seems to me, and this is what makes it paradoxical, that the instinct to use new technology isn't evenly distributed. It might be the high performers who are most likely to adopt new technology rather than the lower performers. So it has the highest potential to help people who struggle at their job, but it's actually more used by people who are already excellent at their job. Has any of the work that you've done or the surveys that you've done, help to resolve or identify this tension?

So I can give you the practical answer in the data is that it is, the usage of AI as variable. It's kind of what you would expect, white-color office jobs, managers. It's usage is developers and people who write software code, but it's also pretty high in the business world, as you would expect. But I think there's actually something deeper with this that's important, Derek, which is that I think about AI like a teammate. So again, you mentioned staff,

And it is like that. It's like you have on tap a potential entry-level research assistant who can give you facts and synthesis on demand, but maybe doesn't have the best judgment all the time. But you can always ask it for more and it never gets tired. So who's going to benefit most from that? I think it's the people who have unbalanced skill sets, meaning I'm really good at some things, but I have some critical weaknesses, and the AI is going to help me remedy those weaknesses. One obvious example is that its usage is quite high in the developing world for people who are doing

knowledge work, you know, job that's being outsourced from the U.S. to another country, and you need to be able to speak English. And so it's incredible for translating into other languages or for manipulating text to make it sound in certain ways. And so if your deficiency is, I actually have a lot of skills, but I don't speak English, so I can't write this thing very well, or English isn't my native language, and so I struggle with writing. There's incredibly high usage for those things. And so that's what I see AI is. It's like you, it's not going to be as good as the experts in something, but it'll fill in a lot of the gaps. So I think the rewards from AI users are going to go to people who understand that and who, again, leverage their expertise to do what they do best, but use AI to fill in the gaps.

You know, I know this from feedback on every episode that we do on AI. Some people really just hate AI. They really hate JatsyBTBT. They think it's totally bullshit. They think it hallucinates so much as to be fundamentally useless. And I don't want to chalk it up to something like technophobia. I think it's possible that these people are totally into Excel and absolutely into some technologies

and love robotics, but they just think that gender of AI is totally BS. One way that that rubber hits the road at the educational level is that I have to imagine that that's true of some professors as well and even some schools and some departments. I know some of them. Yeah. And so you're going to have, I think, lots of students

coming up in education environments where they're essentially taught that chat GPT is their enemy. But one point that I hear from you is that the winners to the labor force the next few years are very often going to be those who embrace chat chbt as the opposite of an enemy, as a friend, as a research assistant. So are you afraid that the sort of variable uptake of and attitudes toward AI in high school, schools and colleges are going to create a labor force that's like very spiky in terms of its use of and facility with AI.

That's a very real concern. Yes, I think that's true. And I think what you see is that usage of it is lower in jobs and industries where it's not maybe as necessary. Like if you work in customer service, people are not white color customer service, but like if you work in a retail store or something or if you're in personal services, leisure accommodation, things like that, people don't use it as much. Now, I'm less worried about that because I think either it won't be useful or in those jobs, it will become useful as the jobs change.

But in educational settings, as you mentioned, some professors or some schools or whatever aren't using it as much, I think the problem is that the way education is designed currently, it's the best use of AI is to cheat. Okay, but that should be what this means is like we have this tool. We should change the way we educate. And the real worry, as you say, is that people won't change that and therefore, you know, incentives being what they are. people are going to use AI to substitute for their own effort rather than to change the way they learn. I'm very worried about that. I think partly what people's, some skepticism of AI and any new technology is just some people are like that. But I think a lot of it is that not everyone's used the frontier models. If you just use the free models, they're not nearly as good as the ones that you pay for. And so I think there's quite a lot of just people don't understand how good they've gotten and how they're

going to continue to get better. Like the models are worse now than they ever will be. and they're already pretty useful. And so all you need to do is just project a little bit into the need, not even the distant future, just the near future. And even if you assume a normal rate of improvement, we're going to see these things really having an impact on the economy sooner rather than later. Again, not like replacing everyone's jobs, but just changing which jobs are valuable and what things you should be doing and what things you should rely on an AI for.

And I'm worried people are going to get left behind. David, last question. Are there any studies that have looked at companies that, that have really deeply adopted AI like ChatGBT, BT, to study how it changed workflow, productivity, and production at those companies? Yes, there's a recent study that just came out by Anders Home Lom and Emile Vestergaard,

and they studied this adoption rollout of chatbots in a bunch of companies in Denmark. And they basically found that even though people were using them at higher rates, there was no impact on employment or earnings at these firms, and they actually estimate a precise zero impact. Okay, so it wasn't just they couldn't figure it out. It was that they knew there was no impact. And I think, you know, that's not surprising to me as somebody who studied technology and labor markets,

because what you generally see is it takes time for people to figure out how to use these technologies. And that's one reason why you shouldn't expect, you know, huge impacts overnight. This is going to be more of a slow role. And what you see often in other episodes of technological disruption is that actually the spread of product, the dispersion of productivity across firms increases. So some companies are experimenting and can't quite figure it out and some figure it out faster. And that really, again, like, upsets the established order. You know, some companies become AI native and, you know, really, really improve and some don't. And so that's what that study, I think, shows, first of all, that this is going to take longer.

But second of all, that the real impact is going to be a reshuffling of the economic order rather than some rise or fall of some obvious groups. It's going to be a time of disruption coming up. That makes me think that you listen to Mark Andreessen. and other tech executives talk about how much code at their companies is currently being written by AI. And their estimates or claims are always some astronomical number,

30, 40, 50% of the code being written at meta today is written by AI. Next year we expect it to be 100%. And I kind of don't know what to do with these claims because on the one hand, they're usefully hyperbolic for these executives

because those executives are spending, or Facebook specifically, Meta is spending $60 to $70 billion on the infrastructure needed to build AI to power the AI needs of meta. And you can't possibly ask the market to let you spend $70 billion on infrastructure for a new technology that you simultaneously claim to be currently worthless. Like, you kind of do have to over-promise what that technology is going to deliver if you're going to be investing at it at that scale.

But at the same time, maybe they're telling the truth. Maybe a lot of these leading tech companies that are worth in the hundreds of billions and trillions of dollars are just the early adopters of the technology because they're the builders of the technology. And so maybe they're the frontier getting 20, 30 percent growth and productivity from the tools that they're building for a core function like software programming. But from my perspective, it's just like an analyst from afar. It's very difficult to know exactly how much of this is strategic hyperbole and how much of it is a honest glimpse of the future. Yeah, so I love that. And I think my answer would be it's narrowly correct that the AI is writing a lot of code. But then the error is to make the straightforward inference from that seemingly straightforward inference that therefore you're seeing huge productivity gains.

because you still need a human software developer to check the code. Okay. And sometimes it'll make errors and that's fine. So you need people to read it over and make sure it's right. And then more importantly, you need to decide what you're coding. Right? It's not going to tell you what the next business is going to be. And so in some sense, you're going to write a lot of code, but you might discard a lot of it.

And you're going to be able to rapid cycle innovation of different products, you know, more. But you still need people there making those decisions. And until you figure out how to do that, companies are going to want to keep people around until they figure out, you know, what the new order is for their business. And so I don't think you're seeing 30, 40, 50 percent productivity growth in the short run because companies haven't figured out how to use the AI with enough confidence to start laying a bunch of people off and restructuring their entire businesses. It probably will come eventually.

But I think it'll be much slower and more gradual and, again, much more disruptive in a sense of like winners and losers than some absolute change. David Deming, thank you very much. Thank you, Derek.