Ioana Marinescu: Why Workers Are Underpaid, and What We Could Do About It

Ben Zweig: What would you earn if the labor market were perfectly competitive? According to very interesting research from Ioana Marinescu, it's a lot more than you're making now. Ioana is a professor of economics at Penn whose work on labor market competition and basic income is reshaping how we think about worker power. Here's our talk. Welcome to the Economics of Work. I'm excited to welcome Ioana Marinescu, professor at Penn and someone who is always pushing the boundaries of labor economics and doing interesting topical work. So Ioana, welcome to the show.

Ioana Marinescu: Thank you so much. It's great to be here.

00:30 What would a perfectly competitive labor market pay?

Ben Zweig: Yeah, so let me ask you, let's say we had a perfectly competitive labor market. How would that look different than the labor market we have today?

Ioana Marinescu: So traditionally when we think about perfectly competitive labor market, what's really important for people to understand is that in such a market, people would be paid their, what we call marginal productivity, but essentially the amount that they contribute to their company's bottom line. And in the real world, typically people are paid less than that. And one key reason is that there is a lack of competition in the labor market that is, you don't have infinitely many employers who are vying to hire you who are bidding up your wage, but rather most people only have a few choices and that limits their ability to earn what their true worth is, so to speak.

Ben Zweig: Yeah, so how do we know that? We hypothesized that there's some market concentration among firms, but not market concentration among workers. But how do we know that is very different from where we are today and how different? Do we have a sense of magnitude?

Ioana Marinescu: Yeah, so all of the things being fixed, workers might make between 20 and 50% more if they were paid according to their current productivity. Of course, I just want to outline for the listeners that there is a heroic assumption because there's many other things that would need to change and so on, but it sort of gives you the upper bound that you think that this could contribute to holding wages down.

02:16 Rent sharing and the role of unions

Ben Zweig: That seems very, very high. I mean, so if we were to think that this like 20 to 50% premium from competition, like how much of that is really what we'd get from competition versus what we'd get from kind of like a more even split of economic rents? Or do you think that's a part of it?

Ioana Marinescu: So in this, there are different ways of looking at this problem, but essentially, in the perfectly competitive market, of course, there are no rents to be shared very importantly. And therefore, everybody, including the owners of capital is paid exactly what they're contributing. And in that sense, there are no rents to be had. And so, again, it's perfect competition, there's no rents. And so then there is a question of when there are rents, like how does the rent sharing work? And there's actually active research on that. And therefore, some of these measurements that I'm talking about here might look different if in fact, for example, employers might be willing or sharing some of the rents, for example, because there's unions. And actually, that's one way that this could happen. And we know from research, from empirical research that in markets that are less competitive for workers, if there are unions, that doesn't diminish their wages as much, the lack of competition has less of a negative impact if unions are there because presumably they're able to act as a countervailing power and get a share of some of the rents. And how exactly some of these things happen is not fully understood. And there's active research in terms of the fine print of the details of the models. But we do know empirically from several, in fact, many studies, there's different ways of looking at it, but that generally higher competition tends to increase wages and that it can increase them fairly substantially. As I told you, this would be the maximum that you could think of potentially having if you went to full competition.

Ben Zweig: Yeah, and also presumably it would lower the benefit of unions in a way. Like if unions are really meant to counteract anti-competitiveness with more anti-competitiveness, then maybe there's some like rebalancing that way. But if there were more competition, then the benefit of unions would be lower.

Ioana Marinescu: Right, exactly. Although it really depends a little bit on how we see exactly the role of unions operating and what happens to competition in the product market. So there's a number of other dimensions. Like for example, hypothetically, you could imagine that again, very stark unrealistic situation, but that labor markets are perfectly competitive. So people would be paid their marginal productivity, but maybe product markets aren't competitive. So firms are making a ton of profit and now unions might be able to demand some rent sharing to get a little bit of that, above and beyond what the workers are contributing, for example. So of course that's a very stark scenario, but just to understand that there's other elements, of course, of like unions in the firm.

Ben Zweig: Right, and you can make the case that they have a role to play and just like messaging to leadership and stuff like that, less, more like coordination benefit.

Ioana Marinescu: Absolutely, let me clarify. When we talk about monopsony power, or this issue on competition and the fact that really employers have some power to pay you less than your full economic worth, it's not just in terms of the pay, but it's sort of the whole package. That also means that employers have the ability to offer you terms and conditions of work that are maybe not exactly what you would prefer, because hey, what are you gonna do if there aren't very good alternative options, right? And so what the, and that if you had very good options, you could bargain certain terms that are individually set, but it could be that certain aspects by given the kind of thing that it is have to be collectively set. You can't individually, let's say it's a firm, everybody has to be there at the same time. It's very important that people are in the office at the same time. Well, that has to be a company wide policy, let's say. You can't describe just for one worker, well, I'm working on Sunday, but you are, okay. If that's the case, oh, the one worker, even if they have a lot of competition, they can't really negotiate that term because it's part of what the firm does. But the union as representing all the workers potentially could be like, look, let's think through possibilities. How about this? We think that the workers collectively prefer this other thing. And like we're gonna try to find some way of organizing ourselves that is more responsive to what workers would prefer. And that's especially valid for the sorts of things that are very hard to customize. So that even if the worker had a lot of power to negotiate, it's just not the sort of thing that you negotiate on an individual basis.

Ben Zweig: Right, right, it makes sense. I mean, the whole compensation, the whole like total rewards package is like a singular decision. That's not like a market determined decision.

07:58 Google perks and the limits of monopsony power

Ben Zweig: Yeah, I think that makes a lot of sense. I mean, I heard a story from, I don't know if it's like apocryphal or a real story, but from Hal Varian who is chief economist at Google, where he said when Google was in the early days competing for engineers with Apple and whoever we know the story that there was collusion around wages for entry-level engineers. And Steve Jobs was the big bully here and basically didn't want anyone competing for wages like basically price setting wages. And Hal Varian basically attributes the explosion in amenities from Google to a way to circumvent this kind of cartel, if you will. So that I guess is an illustration of like the limits to collusion where you can just like switch to non-salary related forms of compensation and give everyone like free haircuts or whatever they did. So yeah, I guess I'm curious to hear your thoughts. I'm like, where does monopsony power kind of hit its limits?

Ioana Marinescu: Right, so obviously the power of employers isn't unlimited because it's really around, it's a continuum. So what I want people to understand is that you have to think about what do I mean with continuum? I'm a worker and I'm looking at my range of possibilities. And the thing is like, they might be my favorite firm. Maybe my current firm, if I'm very lucky, is my favorite firm. Then on the sides, they're comparable but slightly worse from my perspective. And more and more things are further away from my favorite. But there's still jobs out there. It's just that it might be that they are worse as a package than what I currently have. And that still severely limits employers' power because if essentially if they are trying to shortcut you too much, shortchange you too much, then you can go to this other job that maybe isn't as desirable in other ways, but now that they quote unquote, shortchange you enough, you're just gonna leap. So that's sort of the, especially the sort of limits in most situations, workers do have other options. It's just that they're not as good. And so that gives employers some power but it's not usually complete, right?

10:25 Who holds the power, workers or firms?

Ben Zweig: But does it give them more relative power? Like that to me sounds like a kind of symmetric phenomenon. Like just how workers have some rank preference of firms. Firms presumably have that for workers. And the more we have these kind of like differentiated jobs where the extent to which you want a good fit is really important, that kind of works both ways. Like I would think of that phenomenon as like kind of taking like a thick market and making it thinner where you have lots of search and match and then you have a very limited amount of search and match. And that is like a friction full market. It's a bad it results in bad matches. But I wouldn't think of that as something necessarily like shifting the relative balance between firms and workers.

Ioana Marinescu: So typically there are multiple applicants per job. So and typically workers get only one offer like at any point in time. Like the offers the match, like a worker's... Like a variation, right? And a lot of workers are searching while unemployed, which means if they're unemployed, that's the only offer. So of course they can decline and keep searching, but that's pretty costly. So what I'm trying to say is that for many workers, it could be different than I'll get there in some markets. But for many workers, there are just, there's a lot of competition essentially on the workers, typically. And perhaps the argument is less on the other side. And that's especially true in markets where the roles are such that a lot of people can potentially do the job. And maybe much isn't that important. I mean, of course there's always so much quality, but it may be less, more or less important. If we're getting to highly specialized jobs, like maybe the kinds of Google engineers that we were just discussing before, that might start to become a little bit more relevant, especially if at the point in time when the market is really hot, as in everybody's trying to hire and there's not enough talent. Like maybe we're seeing right now with some of the AI.

Ben Zweig: Right, making a hundred million dollars or whatever it is.

Ioana Marinescu: So but even then it's like a balance. The question is, and that worker side rarity is gonna go the other way, but the question is on net, which way it goes and what is the case is that I think the average, for the average worker on net, the power is on the side of the employee. But again, there could be circumstances where that's different. And obviously again, it's like still, if you're a better worker, even if it's not a super specialized market, you're usually in a better position just because you're more better fit, more productive, et cetera, to negotiate.

13:16 Is talent hoarding anti-competitive?

Ben Zweig: Yeah, so what do you think of, what do you make of talent hoarding? Is that anti-competitive? Like how should we think about like this phenomenon of talent hoarding that was like in the news and stuff? Like how does that fit within this framework of kind of like relative worker and firm power?

Ioana Marinescu: I think I know what you mean by talent hoarding, but just define it for me. Just so we're on the same page.

Ben Zweig: Yeah, the thing is like, I don't really know how to define it, but I'll try. Because I mean, the way I've understood it is that firms will kind of over hire and relative to what they need. And there's like two hypotheses, two hypotheses as I understand it of why they might hoard talent. One, which sounds kind of fine to me, is more like just like being ready for needing to accelerate production if there's a new hot trendy thing. And that just kind of seems like keeping a bench, which to me feels just like good workforce planning or like responsible workforce planning. But then the other hypothesis is kind of that they don't want their competitors having top talent and would rather pay someone to do nothing than kind of let others compete. And even within that explanation, there's a question like, okay does Google not want Netflix having good engineers or does Google not want these talented engineers starting the next Google? So it might be a barrier to entry, might just be a like direct competitor phenomenon, but I don't know, is that a real thing, do you think?

Ioana Marinescu: So I think it really depends on, I think it's very important to understand how they're hoarding the talent, quote unquote. And so I think that the, hearing it is like, there is some reason other than your productive capability while they're keeping you on. I mean, the extreme, you're doing nothing. I mean, that's really extreme, but like they have some strategic reason to keep you on. So if they're paying you for that service, of you staying there, and if somebody else were paying you more, they'll be like, okay, you go, that's one thing. And I think it might have some adverse in practice effects on competition. For example, if that stops you from going to a competitor that might start a competing company or just maybe a smaller competitor that could develop, but that might be difficult to say that it's direct impediment, because they're paying you and they're not stopping you from going. So if the other guys feel like there's enough value in buying you out they'll do. But now it gets a trickier when, for example, you have a non-compete, some sort of a fee you would have to pay your employer if you're gonna leave. Then that's a different issue a little bit, because that means that even if the competitor were willing to pay you as much and more, you would face a hurdle in order to be able to exit. And there you could say that's really trying to put sand in the gears of workers' mobility with potentially adverse effects on the other companies' development. And so I think that's really important to consider. There's a paper that looks at this in the case of non-competes and sort of tries to model that and shows that for executives, they are often fairly well compensated for what it would seem like for signing a non-compete. So like, I think they're getting paid more for accepting this non-compete, and it's probably roughly reasonable what they're getting paid for that. However, that hurts the other firms that are trying to compete. So if you think about it from a social perspective, it would seem like, at least according to that paper, that it would be more efficient socially to not allow this non-competes, even though the worker themselves are getting compensated, but from a social perspective, it would be better not to allow them.

Ben Zweig: Totally. I mean, that kind of feels like how it works in finance, where if someone works at a hedge fund, they get paid to not work for a year or two. And the worker is fine. They're on garden leave and they go teach their kids volleyball or whatever they do. But it's there's so much productive capacity, just like sitting idle. It seems crazy.

Ioana Marinescu: Right. So that's something that then the law has to figure out which ones of those might we be able to address. And you could also either apply existing laws in some cases, and in some cases, you might need to legislate to clarify what is not allowed, or why we don't want it to be allowed for various reasons of the sorts that we discussed.

18:15 Why the non-compete ban fell apart

Ben Zweig: Yeah. I mean it was so exciting when non-competes seemed like they were going to be not allowed. Well, why do you think that has reversed?

Ioana Marinescu: So I think the argument that has been made, as I understand it, I'm not a lawyer, but in very broad brush--

Ben Zweig: Well, you were part of the DOJ, so more of a lawyer than I am, you know.

Ioana Marinescu: Yeah, sure. But the courts argued that banning them outright, which wasn't exactly so, but basically, as the FTC did, was a little bit a bridge too far in terms of the FTC substituting themselves for the legislature in imposing like a really big change in regulation, essentially. And so that's sort of so-called the major questions doctrine, as I understand it. If it's a really important, like really big thing the agency can substitute themselves for the legislature. Some law would have been to be passed to ban them completely. That's very different from saying that companies couldn't be sued because we still have the antitrust law. And in the case that you could show that this specific non-compete is a so-called unreasonable restraint of trade, as in the terms are really onerous relative to what the firm could reasonably claim is a reasonable business stand that they're taking, then they could be sued and that could be struck. But you can't do it-- or at least the current state of affairs is that you can't do that in a blanket way. You have to go like sue in particular cases for particular non-competes and prove that this situation is such that it can be addressed legally. And also some states have banned them too. So I'm saying instead, in situations where they're not banned, you can still get sued. But it's a process versus just having the straight up ban.

Ben Zweig: Yeah, that happened to us recently. We wanted to hire someone from a competitor. And our kind of read on the non-compete was that it wasn't enforceable. It was overly broad. It was kind of ridiculous. And then they had sent a letter saying, you shouldn't be doing this. And then we were like, well, let's fight it because we don't think they have a good to stand on. And so we were going to fight it, firm against firm. But they were suing the worker. So the worker kind of got chickened out. And he can't bear these costs as a human person. So basically, the whole thing kind of didn't go anywhere. And we ended up not hiring this person. And I don't know, he just seems like a victim. And yet nothing went to court. Nothing happened. So it's just the threat of suing a person that seems so powerful that no one really wants to touch it. It seems like a really bad equilibrium.

Ioana Marinescu: Totally. There's research by Evan Starr and co-authors on this topic. It's so-called the in terrorem effect. I think basically it's terrorizing workers from doing things that, in principle, they shouldn't be allowed to. Because as we said, maybe this is clearly not applicable because it's really excessive, et cetera. But the worker isn't necessarily very sure about what would happen or the uncertainty of bearing the legal costs and so on. So as a result, in some cases, the firm could get away with doing something that isn't even-- if it went to full trial, would probably not be upheld. So that's a problem. And that's why perhaps if that's the direction that the public wants to go, we should have legislation on this. I'm not saying we should. That's not for me to decide. But if people feel strongly about this, one way to solve it would be to pass a law clearly saying, this is not allowed. For most people, there could be some exceptions, but basically similar to what the FTC did.

22:32 Defining labor markets by occupation and geography

Ben Zweig: Yeah. So one thing I think about in this topic is that we have a pretty good convention for understanding what firms are competitive with each other in product markets. And we have a sense they're in the same industry. They sell similar goods and stuff like that. But we don't really have good ways of understanding who is competitive in labor markets. And those really might be different segmentations. Like, we don't have an industry for labor market competition. Like, is that dangerous to not have that be explicit?

Ioana Marinescu: So actually, we have-- and that's the occupation. And I've done a bunch of work and colleagues have as well to show that typically, the occupation, the six-digit SOC code-- these are occupational codes. There's about 800 such occupations. So it's fairly detailed. So examples are customer representative or registered nurse are some of the biggest ones, for example, just for people to have some idea of what these are. So at that level, by commuting zone-- so that would be like-- I'm in Philadelphia right now. In the Philadelphia commuting zone, it would be something like registered nurses in Philadelphia commuting zone gives you an idea of what the market might look like. And we know that these markets generally tend to be reasonable markets. And by reasonable, I mean, there is no such thing as the market because there's always some porosity. Like I was explaining earlier, the worker-- if the conditions are bad enough, can always spread out and go elsewhere. It's just that it would be worse for them. So there is no strict boundary. But the point is that if you define a market in this way, you can run some economics tests that are broadly used in antitrust litigation to show that this market typically would pass the test of a so-called relevant antitrust market. So it's a reasonable definition. But it doesn't mean that it's the market. And you could, for example, cut it finer by, I don't know, like oncology nurse, depending on what exactly we're trying to talk about. Or you could make it bigger to, I don't know, health care or something like that. But what you have to realize is that there's always a trade-off in the sense that if you define your market bigger, you're more sure that the people are unlikely to go outside of this market. So in that sense, it's sort of conservative. Definitely most of the relevant jobs are in there. On the other hand, you might be including a lot of jobs that aren't really that relevant. And the person will only go there in the situation where they're really, really dire. On the other hand, if you make it smaller, you're more likely to include most of the highly relevant jobs, at least if you do it well. However, it's a bit more likely than that if it's really small, the person might be willing to consider something outside that circle. That's why there is some leeway in how big you want to make that circle. And when we're getting to litigation, then there is a choice whoever is suing can define a market that they would say is a reasonable definition. But again, it's not the market. It has to pass the test, but there's different ways of defining it that multiple ways might pass the test. So that's how you talk about it.

Ben Zweig: Yeah, I understand the desire to define who is competitive in labor space by occupation geography. I don't know if that's-- yeah, personally, I'm a little skeptical of that. Because first of all, you could directly observe who goes where. And then you have some places that don't have similar occupation geographies that might have similar talent market competitions. You might have Goldman Sachs competing with NASA and wanting to source from the same talent pool. And different occupations, but maybe similar skills, maybe-- so let's say that this occupation geography way of assessing competitiveness comes with some noise. I mean, every metric comes with some noise.

Ioana Marinescu: Of course it does.

Ben Zweig: Yeah. So I wonder if that noise is maybe picking up on something that's maybe closer to product competitiveness. If someone has a similar set of occupations, they maybe make similar things. And if we have product competition,

27:19 The sleepy monopolist

Ben Zweig: The product competition effect leaking in, I wonder if-- I mean, let me throw out a theory. And let me see if you think this is plausible. So let's imagine that you have firms that are competing in product market, and they are concentrated. But they don't really compete in labor market. And they get more and more concentrated. And they have a lot of market power. Maybe to the extent that market power creates barriers to entry, maybe it's just not that important for them to hire the best people. So maybe if you're really fiercely competing with others, and you really need to be the best, you need to get the best people. But once you're sort of locked in, you don't really need top people. And you could pay them less. And you could treat them worse. And maybe you're sort of just lowering your caliber of talent. Do you think that is sort of leaking in some way?

Ioana Marinescu: Interesting. So you assume that this wasn't for labor market competition. So that's off the table. It's more around what sort of people am I willing to hire. I think it's plausible that the sleepy monopolist, in the extreme case, maybe they're not making a huge ton of effort to attract the best people relative to more competitive industries, regardless that they don't compete for workers with a competitor who's also in the product market or anything like that. That's possible. But if they get too sleepy, they might run into trouble, because that would impair their productivity. And I know some that we're assuming they have market power and so on, but market power is relative. So usually, it sort of shields you somewhat from adverse effects of being outcompeted. But if you become too sleepy, you could still run into trouble. And so it's kind of a bit of a balance. But that doesn't necessarily mean that it affects how you treat your workers in the sense that by assumption, we assume that we aren't envisioning a change in labor market competition. So in the end, you have to pay these people enough relative to the outside option, because that's what competition is. And since you're not competing with these guys anyway, we kind of put that on the table, these people will be paid whatever the level of general competition in that labor market. You can't really shortchange them just because you don't care. It's more like you're going to potentially hire worse people, because you don't care for putting in the effort or the pay for that matter to hire good people. However, you aren't necessarily shortchanging those people that you're hiring, because the outside competition in the labor market is held constant by our thought experiment. It's not affecting that. Does that make sense?

Ben Zweig: Yeah, totally. I mean, I don't even know how we would test this or measure it. But my hypothesis is that if you look at firms that have a lot of market concentration in product space, my guess is that they do have kind of worse employees. Maybe it's a little harsh. But that's just kind of-- that's the vibes that I kind of feel like when you look at the people that work at these monopolists, they seem not as--

Ioana Marinescu: I don't know if somebody wrote a paper on this. That would be a really good thing to investigate. I mean, this would be a little bit descriptive, but still, it's interesting. Yeah, you should go for any listeners. It might be different depending on sort of the dynamics at what stage we're at, because oftentimes, the companies establish a strong market power first by having a great product, a great company, et cetera, potentially also by engaging in anti-competitive conduct on this and together. But often, it's part of the story is that they actually have a great product. And often, in the beginning, they actually hired great people, and that's how they got there. So therefore, let's say that a young market leader, even once they became quite entrenched, might still have good people, because they have a history of doing that. But then eventually, if they stay on for too long, the quality might slide.

Ben Zweig: Yeah, I love this idea of kind of like reputation mining in a way where I think just like anecdotally, as an entrepreneur, I feel like the best people want to work with the best people. And that's a trait of talented people. So if you, let's say, pay a premium to attract the best people, that pays dividends, because then you can get even better people. And then what people get, their total rewards package is their pay plus the ability to kind of work with great people. And that self-fulfills, it self-perpetuates. But then let's say you say, well, now I have all this benefit and people want to work at McKinsey or whatever it is. Then you can pay below market and still get away with it. And then you're kind of like, you've got this reputation asset that you can mine, and it erodes at some point. But I don't know. That seems like a hard thing to optimize for if you're a firm with a good reputation.

Ioana Marinescu: Right. So in this case, if you're adding the reputation in the loop, I would see this as an aspect of the job that is job differentiation. That is, from the point of view of the worker, there is some value that they place on the brand name for some reason. Perhaps because they think that in the future, at the next job, having worked at the company X, that's very prestigious, will increase their ability to find some other job in the future, let's say, or something like that. And if that's the case, then they would accept lower pay because that's a perk of the job. And so that's covered on the differentiated jobs, and that will increase the market power through that channel of that company. If they have a really good name for themselves, that working there for some reason is very appealing just from the brand name.

33:45 How economists and computer scientists see AI

Ben Zweig: Totally, totally. Yeah, I think that's very plausible. So let me shift gears a little bit. I really want to ask you about AI and all the policy recommendations. I mean, you had such an interesting paper. So let me start this way. So you are married to a computer scientist. How do economists and computer scientists differ in how they think about AI, technology? And yeah, how are we different?

Ioana Marinescu: Yeah, so I think right now, obviously, AI is a lot on everyone's mind, especially in some circles, like people who are in computer science adjacent fields, as well as more and more economics. But I think that typically people in AI adjacent fields and economists tend to think really differently about AI and the impact of AI on the economy and labor markets. So very broad brushes. I would say that AI people tend to think that, oh, AI is going to revolutionize everything. It's going to lead to unprecedented growth. And it's going to kill a lot of jobs and maybe even all jobs. And maybe we'll have a singularity, like accelerating growth. Who knows? So that's sort of like a little bit the AI vibe. And then the economics vibe is more like, yes, it's a new technology, a general purpose technology, meaning that it can be applied across many domains. It can even improve our methods of research, thereby further contributing. And so we know that such technologies tend to increase economic growth. But we've had others in the past. It will be something positive, but likely limited. There's limits to everything we've had in the past. ICT, for example, has increased growth for 10 years, roughly up to one percentage point per year. Great. But it didn't change the world. It was very important. It did a lot of things. So therefore, economists tend to be like, it's another one of those. Maybe even a little bit bigger. Who knows? But in the same ballpark of, OK, it's a new technology. We've had others. It's going to change a lot of things. But yeah. So I think that those worldviews are really quite far apart, taken to extremes. And of course, in each camp, there's distribution. And people might find themselves more in the middle. And so in that sense, it was fascinating when I was talking to my husband, who's also a co-author on a paper we wrote about AI. Wow, what's going on here? And our whole project was to think about how you could reconcile these views in the sense of, what do you need to think as an economist, for example, to think that whatever the tech people say is going to happen? So essentially, what are the parameters in your model? What are you assuming could be possible to think that could happen? And conversely, what are economists assuming implicitly where they think that is going to be more limited? So that was the departure point of our project.

Ben Zweig: Yeah, I really like how-- I mean, I think of economics as a way to formalize anything in a way. You have some ideas, some hypothesis about the world. And put that in a model. Solve for it. And I think you really do that very well in this paper and others. Like you take some idea that's in the zeitgeist and put it on paper, see what happens.

Ioana Marinescu: Exactly. That's the point.

37:34 Intelligence saturation and the growth ceiling

Ben Zweig: Yeah. So what happens? I mean, what do you think we can learn from this framework?

Ioana Marinescu: Right. So essentially, I do want to set a very high level that the big difference between economists and AI people is that AI people look at AI and how quickly it scales. And like, woo, it goes up exponentially. And they tend to assume that the whole economy is going to move along. More or less, I mean, everybody realizes that it doesn't have to be exactly one for one, but roughly. And economists realize that's not necessarily the case. In fact, there's a lot of impediments to that being even close to one for one. And that's at a very high level. But it boils down to essentially that there's other things than AI in the economy. And that those interact. And you have to take account of those interactions in order to predict out what the impact of something like AI is going to be on growth or wages. And so what we do in our paper is we say, OK, AI can potentially-- OK, there are two sectors we say. This is a simplification. So as the intelligence sector, you want to think of that as the disembodied intelligence sector where people are people who do remotable jobs that they could do completely behind their computers. And in that economy, in that sector, AI could eventually replace all workers because all their work is on the computer and therefore could be replaced by AI.

Ben Zweig: Is it fair to say this is just exposed versus not exposed?

Ioana Marinescu: Yeah, you could say that. But I think it's useful to think about the in-person because that's going to be a real limiting factor in terms of the ability to, in fact, replace people, in person versus remote. So I'm saying think of the intelligence sector as the fully teleworkable, remoteable jobs. And that's where AI can be deployed and potentially more and more replace people doing that sort of job. And then the rest is everybody else where they have a substantial in-person component. And they work with physical capital in that capacity, which can be buildings, machines, you name it, different things that we use in the physical world to do our job or be supported by in the background. And so now to explain the difference between computer science view if you want and econ view, I want to say that let's say we can expand AI a lot. So then this intelligence sector that uses AI, no matter what happens to workers, is going to grow a lot as we put in more AI. Forget for a moment where workers go and what happens to them. With more AI, the sector grows a lot. So in principle, there's a huge growth upside here. But remember, we still have the physical sector in this economy. And these two have to interact somehow to produce overall output. And so now what we show-- and this is just a basic insight and something from economics-- is it really depends on the substitution between physical and intelligence. That is, how easy is it to replace something that is pure intelligence for something that is physical? As an example, you could think of online teaching versus in-person teaching. I don't know, a meal versus watching Netflix. We're going to go to the restaurant because we're watching Netflix. Things like that. So how substitutable are these activities? And if they are not very substitutable, that is, we call it complementary, then you can add as much AI as you want in the intelligence sector, thereby exploding the amount of intelligence stuff that's being made. The physical side becomes a bottleneck if it stays roughly fixed. And it's very hard to quickly grow the physical side. And there's physical limits to it. So there's a physical side. You can grow your intelligence sector as much as you want. You can show mathematically that if there are complements, the output, the GDP is going to hit a ceiling. Because ultimately, it's being held down by this physical part that you must have the complements. So there's only so far you can go. And so that's that--

Ben Zweig: Is that kind of O-ring in nature?

Ioana Marinescu: Yeah. It's absolutely that exact same reasoning, but it's applied to just kind of thinking about it in terms of physical. And so therefore, if that's the case, we call it intelligent saturation. As you can add a lot of intelligence, but it saturates in terms of its impact on output. So it's not that the intelligence itself can't be improved, but the effect on output saturates. And because wages depend on output, it would also mean that wages would only grow so much, even best case scenario, from the AI. And so that's one critical distinction.

42:57 What is left for knowledge workers?

Ben Zweig: Yeah, this kind of reminds me of the EPOCH framework from Isabella Loaiza and Roberto Rigobon. Where they have this acronym, it stands for empathy, presence, opinion, creativity, hope. These things that are kind of uniquely human. And presence is, I think, the biggest one of those. And you can sort of think about presence as uniquely human, or maybe empathy is uniquely-- whatever it is. But it made me think just personally. I've always thought of my contribution to the economy as kind of being in the intelligence sector. I imagine how you probably contribute to how you think about your own contributions. And if we have this just abundance of intelligence, of the stock of intelligence in the world just explodes, how much do we-- I don't know. I was just introspecting. How much do I bring to the table in terms of presence, empathy, creativity, things that are kind of non-intelligence? And I don't know. It made me uncomfortable. I mean, how should we think about it? How do you think about it?

Ioana Marinescu: Totally. So first, let me say I realize that you might wonder, what is it with this fetishism of in-person versus not in-person? And I forgot to mention the core economics behind it, which is that the cost of robotics and physical capital is continuous to being fairly high. It's declining, but very slowly relative to the cost of AI. And therefore--

Ben Zweig: Well, the cost of tokens is also rising.

Ioana Marinescu: So therefore, the big incentive is if you're going to do something in the business world is to replace people with AI, replacing people with machines is just still very costly. So that's why the simplification, therefore, in the model is that you can't do it at all. Obviously, it's a simplification. But the point is that the action is for you as a business to be able to more and more replace people in the intelligence jobs that's driven by a cost factor.

Ben Zweig: Yeah, it's not that it can't be done at all. It's just that it would not be the machine's comparative advantage.

Ioana Marinescu: Right, exactly. Exactly. So just so we understand why, what is it, why do we need to think about it this way in terms of embodiment? That's where it's coming from. Whereas other concepts like you were mentioning, creativity and empathy, to the degree that they can be delivered virtually, like for avatars personas, making art, et cetera. And if people don't care, right, so we get back to that substitution, they don't care as long as it's creative and empathic and so on. If it's on the computer in virtual reality, it's just as good. That's really important to know, like which one. So that's why I think that the in-person is like really important. Now, going back to your question about how should we feel about ourselves, people like you and I, I think there's two sides to that. The hopeful side is that I think this transformation is going to take some time. And also, some parts of our job might actually be in person. I mean, so personally, I also teach. I go to conferences that tend to be in person. So I think many of these jobs, if you're not going to completely change job, you're going to restructure or change to an adjacent, similar job that is more based on doing in-person activities. And for somebody like me, it would be more interacting with colleagues and students and talking through things, brainstorming, et cetera. And then maybe going back to the computer, but for less time and being a lot enhanced by AI and doing it in that workflow where you would spend more time interacting with people and less time solo at your computer because the computer can largely do that, let's say, I'm imagining autonomously to a large extent. And then you could just go back to base and say, hey, what did you find while I was away? And what should I go do next? So you would rearticulate these jobs to be restructured towards more in-person activity. And that might still be pretty fun because, hey, right now it's virtual. We're doing virtual. But you could imagine maybe it would be even more fun if we did that in person.

Ben Zweig: Probably, yeah.

47:26 Two tiers of worker support

Ben Zweig: Yeah, that sounds right. So you've also done some research on adjustment assistance, UBI, things like that. And I really want to hear your take on what is-- not just what is popular and gaining traction, but also what do we think is-- what should we be preferring as people who want what's best for society and fairness and all that? What is the good solutions in your mind?

Ioana Marinescu: So I think that generally we need two tiers of solutions in terms of-- and here I mean solutions to ensure worker well-being. Just be upfront. That's what I'm thinking about, that AI might replace some jobs which could lead to some-- if we do nothing, some adverse effects for some workers. And how do we address that? Yeah.

Ben Zweig: Well, actually, sorry, is worker the right-- the analysis or just person-- what if someone's not working? Is that like-- are they excluded?

Ioana Marinescu: Let's think about workers for now, because that's what I want to think about in my analysis. I mean, we can-- it's a different question for the non-workers. So let's think about workers, although they will be covered by one of the tiers of my policy plank overall. So one is we need something to address the pains of reallocation. That is, AI, even if it doesn't durably suppress the number of jobs, is going to change the nature of jobs. And therefore, people are going to have to reshuffle, do things differently, change jobs. And we know that often that's hard and can be economically costly. Like, people might go through bouts of unemployment. Searching for a new job. So one set of policies, therefore, is this transition assistance. And one specific policy I'm proposing to consider is the AIAI, the AI adjustment insurance that gives people longer unemployment benefits, retraining, wage insurance to help them with the transition. And you could structure it differently, but I think it's really important to think about, how do we support people as the jobs are being reshuffled and people need to change what they're doing, because the way that we do jobs is changing. So that's one bucket. The other bucket, tier two, is I want to imagine that it's possible that really the number of jobs gets drastically cut. Maybe in the extreme, maybe in some far future, maybe completely. I mean, because of my modeling around physical jobs, I don't think that's the most likely, but who knows, especially in the longer future. And so then, in that case, I want to make sure that since most people get most of the income from work, if that were the case, a lot of people would get no income. So now we need a durable income replacement mechanism. And so therefore, for that, we need to find some way of delivering consistent income to people, even if they're not working. And that would then cover potential also people who weren't working to begin with, but in any case, delivering income to people. And in my own proposal, I propose something like a digital dividend that gives people cash based on the tax, but there's different ways of implementing it. The basic thing it has to do is to replace labor income for people in the case that this becomes durable non-employment. And that's why the second policy, since there's a lot of uncertainty, should be flexible. Should ideally exist, be in place, but be deployed like a parachute if this scenario-- because it's also very expensive to do. But perhaps good case scenario, the AI is fantastic. It produces huge increases in GDP, et cetera. So we have a lot to play with, and we can also redistribute that to people. So that's the second category. And you really need both so that we can feel safe, because I'm all about worker support. We'll support you in the transition, but then in case it comes to that, we have something there too.

51:48 Can a safety net boost entrepreneurship?

Ben Zweig: Yeah. OK, can I tell you what I like about both of those, which may be, I think, underrated benefits. But in the adjustment insurance example, to me that solves kind of like cyclical unemployment to the extent that we think it's attributable to sticky wages. So if people are more open to accepting jobs at lower pay, that seems like we'd have more market clearing in times of recessions, which seems really positive, even outside of technology. So I like that. And then in terms of the digital dividend, or I'm sort of lumping that in with UBI, is that fair to--

Ioana Marinescu: There's different ways of implementing tier two. People also talk about basic capital. So there's different modalities. But I think the core important function is to give people an income, even if they're not.

Ben Zweig: And yeah, I mean, I like that it's not conditional on looking for work, or not having a job, or having low income, because then it doesn't have the same disincentive effects. But another thing I like about it-- and I'm curious to hear your thoughts if you think this is a plausible scenario. But I think of myself as an entrepreneur. And so many entrepreneurs have some form of safety net. And for me, when I quit my job, I took on adjuncting, which is super easy. And it was kind of a form of UBI, in a way. I just got some income to keep the lights on where--

Ioana Marinescu: There's actually evidence that this is a pretty competitive market. So there's lots of options for people who want to do adjuncting.

Ben Zweig: Oh, interesting. Yeah. Yeah. Yeah. Yeah, maybe-- yeah, that's an interesting thought. Anyway, just by the side. Yeah, so for whatever reason, I felt like I had a safety net because of that. And a lot of entrepreneurs kind of come from privileged backgrounds. And I think if people have more of a safety net, they'll probably be more proactive in starting businesses and just taking risks. And that creates more labor demand. So even if there is some disincentive effect to working, that could kind of be balanced out by just being a net positive for labor demand just through increases in risk taking. Do you think that's something that is being thought about? Or do you think that's a--

Ioana Marinescu: So people talk a lot about this one. But I haven't seen any evidence that these kinds of schemes would directly increase entrepreneurship. And I think the reason is that most of the things that have been tried are pretty small. So it's nice to have a little cushion, but it's a little small to take a significant risk. However, one thing that is often neglected-- and I've shown in one of my papers-- is that if you do something like a cash transfer at scale, it increases demand in the product market. And that itself creates jobs to satisfy that demand, especially when it's demand for locally produced things, like in-person stuff back to where we were at before. And so to some extent, actually having that scheme even can create jobs. Again, to the extent that there are still jobs that humans can productively do, such as in-person jobs, then giving people this income-- and I've shown this in the case of Alaska-- seems to have likely stimulated job creation in locally placed jobs where people are spending their money at, like restaurants, things like that.

Ben Zweig: OK. So let me hit you with an instrument. I don't know if it's possible. I haven't looked at the data. But I'm not an academic. So you or any academic is listening, because maybe see if this is possible. But I would think that one way that people get more of a safety net is from inheritance. So maybe if their parents pass away early, that is presumably random and gives people a cushion. And that happens for some people earlier, so it happens for some people later. So yeah, just want to put that out there as maybe an instrument.

Ioana Marinescu: Absolutely. Yeah, totally. Maybe it has been done. It's just that I haven't looked at-- when I've been reviewing the literature, et cetera, I was looking at smaller, more like regular transfers that would look more like UBI. But it's an interesting consideration. And I could imagine that if it's big-ish, and for those who are maybe at the margin of entrepreneurship, that could at least plausibly make a difference.

Ben Zweig: Yeah, unless maybe they don't want to embarrass their parents by starting a company, and then that goes away. So who knows.

57:03 Using AI to improve public policy

Ben Zweig: OK, thank you so much. I just have one final question for you. And I think wrapped in a compliment, I think you really have such a good sense that I think is not automatable for the record of looking at topics that are just really interesting. I think your portfolio of work is very topical and very interesting. So for the benefit of the listener and anyone else, what are you paying attention to? What do you see that's an interesting frontier that we should be paying more attention to?

Ioana Marinescu: That's a good one. Let me see. Obviously, lots of things. It's just that I'm thinking what might be the best--

Ben Zweig: You can rattle off as many as you want.

Ioana Marinescu: The best thing to talk about. I think one of the things that is underappreciated is the ability of AI to do good, and not just for the companies. Obviously, we're locked into this, oh my god, I could suppress jobs, help companies at the expense of workers. But there are many upsides that are outside of that. In particular, using AI to improve public policymaking. I've been teaching a whole course about that. So if you go from gathering data on public opinion, what do people want? What do they think is a problem? To devising an array of possible responses, decision aid to pick that, implementation aid, dashboarding, monitoring. All of that could potentially be made better and more efficient through the use of AI. And it's been done somewhat already for digitalization, but potentially there are additional gains to be made with this new technology. So I'm really excited about the prospect of AI to improve the policy cycle, the policymaking cycle. And I think that this could be especially promising down the line, potentially, for developing countries where sometimes state capacity is weaker. And if deployed appropriately, could be very helpful. And also in less-- in more developing countries, additionally, they have the benefit that fewer people are working in the pure intelligence sphere, and more people are doing in-person work. So there will be possibly less disruption, immediate disruption to their labor market. And there is this upside of being able to adopt the technology to do all sorts of things that perhaps they weren't able to easily do before. So I have a lot of hope and interest in how to apply AI in a beneficial way to yield benefits in terms of public policy, and also global benefits for all countries in the world.

Ben Zweig: OK, very exciting. Well, that's the thing to pay attention to. Thank you so much. Really appreciate you taking the time. Thank you. The Economics of Work is brought to you by Revelio Labs, workforce data for research and benchmarking. If you enjoyed, please remember to rate, review, and subscribe. It'll help us reach other curious listeners. I'm your host, Ben Zweig. Our producer is Cole Wagner. Thanks for listening.