Related Links

• Automation, Learning and Career Dynamics
• Erik Hurst interview in Econ Focus

Transcript

Tim Sablik: My guest today is Erik Hurst, the Roman Family Distinguished Service Professor of Economics and John E. Juke Faculty Fellow at the University of Chicago's Booth School of Business. Erik, welcome to the show.

Erik Hurst: Thank you for having me. This is wonderful.

Sablik: We've talked on the show before about CORE Week, the week-long collaborations between Richmond Fed and academic economists, hosted in Richmond seven times a year. You're here attending and presenting at the August CORE Week, which is happening right now as we record this. I'm grateful you could take a few minutes away from the presentations to come join me in the studio and talk a bit about the research you presented.

Your paper that you talked about covers AI adoption and how it might affect human capital development. There has been a lot of discussions on AI's impact on the labor market. One place we could start is with the pre-AI picture. How have economists thought about the way workers accumulate skills and knowledge over their careers?

Hurst: The effects of technology on the labor market is a fundamental issue in economics that we've been studying for half a century, if not longer. Most economic growth that we've experienced comes from technological change. The reason that our wages are higher today than they were 50 years ago, than they were 150 years ago, is because new technologies are coming along, making workers more productive — those technologies like electricity or the highway system or the steam engine.

One thing I've been interested in is how do those technologies that come along affect workers in the short run relative to the long run, because sometimes these technologies can be disruptive. If we go back 150 years in the U.S., about 80 percent of men were farmers or worked in the agriculture sector. Today, that's 3 percent. Along the way, a robot came along. We just happen to call the robot a tractor.

Sablik: Some of the tractors these days are robots.

Hurst: Exactly.

But initially, you know, it comes along and it displaced all those workers from the agriculture sector. And so, we've been thinking about how those workers in agriculture got reabsorbed in other sectors. [It's] the same with the early parts of the Industrial Revolution. We had artists and craftsmen, and then you have factory assembly lines, etc. displace those artisans and those artisans had to be reabsorbed.

When we start thinking about technology's effect on the labor market, I break it into four questions that we need to think about. One: is the technology a complement or a substitute with a worker? These are fancy words that say, does the technology make us more productive or does it displace us? If it's a complement with you, that's economic growth. That makes us all better off. That's how we've gotten the standard of living increasing over time. If it displaces us, that could be problematic.

But then you ask the second question: What sectors are growing at the time those workers are displaced, and do those workers have the skills to be absorbed by those sectors? So, if agriculture is declining and manufacturing is growing, how hard is it to take an agricultural worker and make him a manufacturing worker. And, is that different than when manufacturing is declining and then lawyers are increasing? It might be harder to take a manufacturing worker and make him a lawyer than it would be for an agricultural worker to make him a manufacturer. So that's the second question: how easy is it if you are displaced to be absorbed by another sector?

The third question is: Is there any barriers then for those workers getting the skills to move to those new sectors? If there are some barriers, maybe there's a role for the government to come in and help get workers' skills in order to take a manufacturing worker and help them become a lawyer if that was [what] the economy needed.

The last thing [is] how fast it occurs. The faster these technologies get implemented, the more disruptive they could be [and] the harder the economy is to adjust to things that happen fast rather than things that happen slow.

So, when I think about AI, I'm always going to break it into those four things. Is it a complement or substitute with workers? If it's a substitute with workers, where are those workers going to be reallocated? Is there any barrier to that reallocation? And, how fast is it happening?

Sablik: How do you feel that AI is similar to large technological changes in the past?

Hurst: You're hearing a lot of doom sometimes from people saying AI is going to displace people in certain sectors. We've gone through that before. To some extent, the same economics we did with other technological change is going to be something we're going to ask with AI. If those workers get displaced, where are they going to be reabsorbed to? What are going to be the growing sectors in the economy that workers are going to be needed? How fast is it going to occur? And, is there any barriers to the reallocation? Same thing.

AI, though, I do believe is distinct in two ways relative to these other technological changes. The first is that it does substitute with cognitive skills as opposed to manual skills. Robots hit manufacturing workers. The tractor hit agricultural workers. AI might hit new students who are going into the law profession. So, how does that affect incentives to accumulate human capital? That's what I find is an interesting component to AI relative to these other technologies.

The second part is AI might be moving a little bit faster than the tractor did. So, the speed might be something a little bit more of interest.

Sablik: This question of how it might affect human capital development is exactly the focus of the paper that you presented at the conference. What's the case for AI disrupting human capital development?

Hurst: There's going to be a pessimistic view and an optimistic view. I'm going to start with the pessimistic, then we'll come …

Sablik: We could have done it the other way. But we'll leave people with the optimistic view.

Hurst: Exactly. I agree with that.

When we think about the tractor, the tractor displaced the agricultural worker at every stage of their career. You didn't need young agricultural workers. You didn't need old agricultural workers. It just killed the agriculture employment overall. What people are talking about with AI is it might be a substitute for you at the things you do when you're young, like entry-level positions, but might be a compliment for you at older stages of your career.

AI is helping me produce my research a lot more than it did before — a complement for me when I'm old. My graduate students — who were me 30 years ago — they're getting displaced because I don't need them anymore. So, the idea is that entry-level positions might be displaced.

The cost is if there's some learning that occurs on the job. If my graduate students — by working for me when they're young — become better and become a better version of me in the future, by taking away that opportunity when they're young might affect the amount of learning that people do throughout their careers. Who is going to be the managers tomorrow if we don't hire entry-level positions today?

Sablik: Who's going to be the next Erik Hurst after you retire?

Hurst: So that's the pessimistic view.

Sablik: Yeah, and so that's pessimistic not only for the individuals but for the economy as a whole.

Hurst: Exactly.

Sablik: So, what is the optimistic view then?

Hurst: Well, the optimistic view goes to the other side, which is we're going to be able to create a whole bunch more of new things from which people can learn from. AI will also create a whole bunch of new types of tasks and those workers could learn from those new tasks. Couple that with the fact that AI might also help us learn faster.

So, think about a law firm. Entry-level lawyers are doing two things: research for the senior partners, but they're also learning how to work with clients and present in front of judges. If AI replaces some of those entry-level research positions, you could still have some junior workers who could focus more of their time on learning how to get clients. So, it's still a way to keep the pipeline potentially going forward.

In my research, I put these things together. I show that over most models, that second optimistic effect is going to dominate over many periods.

Now, I don't know if you ever watched "Battlestar Galactica." In that state of the world, the robots take over. In a world where robots could do everything — create new ideas and do the tasks — we don't need "human" human capital anymore because there's no value. That's way, way, way off in the future, and I think we'll have more problems there than just the human capital component.

Until then, over the medium run, I think we're in a world where the human capital pipeline will remain robust, even if AI does eliminate some of those entry-level positions.

Sablik: You mentioned some of the results from your paper and your approach with your co-authors is [to] build this model to explore it. The trillion-dollar question in the economy right now is trying to figure out what's the impact of AI going to be on the economy, but also on the labor market. As an economist, how do you think about how we test the impact of AI in the labor market? What do we need to find these answers?

Hurst: Technological change takes long, long periods of time to go through. The tractor rolled out over 40, 50 years. When we moved from steam engines to electrical engines, the steam engine was still used, even though electrical engines were coming through and they were dominating for long periods of time.

Sablik: Computers ...

Hurst: Computers are the same exact way. And so, AI has been around for what, a cup of coffee? AI has been affecting the economy for 24 months, 36 months.

I grew up [in an] extremely lower middle-class family in upstate New York. Most of my family did not go to college. For them, AI is just fancy Google. It hasn't hit the restaurants or the nursing sectors or their mechanic shops or anything else. So, I'm not going to expect to see anything in the labor market for five, six, 10 years for us to start getting testable implications of how AI is going to affect the labor market.

And in that five, six years, it's going to be concentrated in small sectors and not displacing tremendous amounts of the economy as a whole. Why? Because these things just take time. Testing my theories or anybody else's theory about AI in the labor market is not going to be done this year, next year, in the next few years. We're going to be a few years off to try to understand the effects of AI on the labor market.

Sablik: Assuming that this will take a long time to play out, if there are policymakers who want to be forward thinking and encourage the optimistic scenario we laid out versus the pessimistic one, what policies do you think might be effective?

Hurst: The thing that most policymakers should be thinking about is how we could regulate AI. What I am worried about in the short run is rogue actors potentially coming along and using AI in ways that are detrimental to society, even if it has no effect on the labor market — coming up with a virus that could be destructive to humanity.

In terms of labor market effects, you just want to be monitoring right now whether [AI is] a complement or a substitute. We don't know yet. And so, you don't want to tax AI because that might slow down productivity growth going through. But you do want to make sure you're monitoring where displacements are going to be occurring, where complements [are] going to be occurring. The Richmond Fed is doing some of that now through some of its surveys. I'm not expecting you're going to see much in the short run, but having the measurement allows the policymakers to make some informed decisions as they go along.

Sablik: I'm sure there's many, but if you had to pick a couple of outstanding questions on this topic, the ones that you'd like to explore next in your research.

Hurst: I have a series of projects in my head that I'm making slow progress on, well before AI ever, ever came along. It's how long do those adjustments take place once workers are displaced? How long does it take for you to move to a growing sector?

That's really important for policy. If workers get displaced and it's going to take a generation or three generations or five generations for that adjustment to take place, I might be in favor of certain types of policies — universal basic income, things like that. If the adjustment is going to take place, five, 10, 15 years, then I might want to do things like subsidize transitions to those sectors if there's a friction, helping them get the skills they need to move to the new sectors. That might not even be college, but just you know vocational training of various sorts.

And so, understanding better the transition dynamics of human capital acquisition, mobility of workers to these shocks, I think is a first order question for policy.