Podcast: AI’s impact on the labor market (with Brian Albrecht)

On this week’s episode of the podcast, I am joined by Brian Albrecht, the Chief Economist at the International Center for Law and Economics. We explore the complex intersection of generative AI and labor economics, alongside a deep dive into the regulatory and economic implications of personalized pricing models. The podcast is tethered to two articles that Brian recently published: You are not a horse, from his Substack, Economic Forces; and Eliminate ‘personal pricing’ and you risk harming consumers, which he published as an op-ed in the Financial Times.

Among other things, we discuss:

• Whether humans will face the same economic obsolescence as horses after the introduction of the internal combustion engine
• How the redistribution of consumer savings from automated services creates new employment opportunities in diverse service sectors
• If the historical stability of labor’s share of income can survive the rapid advancement of generative artificial intelligence
• What long-term productivity gains from previous technological revolutions reveal about the potential economic trajectory of modern AI
• Why government policy responses to automation might inadvertently hinder productivity by imposing restrictive taxes on data and computation
• How personalized pricing models actually improve market efficiency by reducing the deadweight loss associated with traditional discounting methods
• When the public’s perception of fairness in pricing conflicts with the economic reality of supply and demand for scarce resources

Thanks to the sponsors of this week’s episode of the Mobile Dev Memo podcast:

• ⁠⁠INCRMNTAL⁠⁠⁠. True attribution measures incrementality, always on.
• ⁠Branch⁠. Branch is an AI-powered MMP, connecting every paid, owned, and organic touchpoint so growth teams can see exactly where to put their dollars to bring users in the door and keep them coming back

Interested in sponsoring the Mobile Dev Memo podcast? Contact Mobile Dev Memo advertising.

The Mobile Dev Memo podcast is available on:

• Spotify
• YouTube
• Apple Podcasts

Transcript

Eric Seufert: Welcome back to the Mobile Dev Memo podcast. I am your host, Eric Seufert, and I am joined today by Brian Albrecht. Brian, welcome back to the podcast.

Brian Albrecht: Happy to be here again.

ES: We last saw each other in Rome in March. How have you been since then?

BA: Good. It has been a little hectic with a move, but there is a lot going on in the world and plenty of fun things to talk about.

ES: We probably have more than one episode’s worth of content today. The impetus for bringing you back was a blog post you wrote called “You Are Not a Horse.” You also wrote a fascinating op-ed in the Financial Times recently called “Eliminate personalized pricing and you risk harming consumers.”

These two topics essentially have nothing to do with each other, but I want to try to talk through both. Before we get to that, could you please reintroduce yourself to the audience?

BA: I am Brian Albrecht, the Chief Economist at the International Center for Law and Economics. I am a wide-ranging economist, which is why we are going to cover a few different things here. If people want to read more from me, I have a newsletter called Economic Forces on Substack. I work on all sorts of things, from antitrust to price discrimination. General economics is my beat.

ES: I feel like you have gained a lot of notoriety on Twitter recently over AI stuff. Do you perceive that to be true?

BA: It is hard to tell how much is just that everyone is writing about AI, but I am leaning into it more than most. I am trying to write a lot on it because I think a lot of the discourse around AI could use some basic economics. That has been my shtick for as long as I have been writing. When we come to a new thing, whether it is COVID or tariffs, I always try to ground our discussions in basic economics.

I think that has been missing in a lot of the AI discourse, which is either tech-heavy and focused on technical details, or trying to be grounded in economics but being too clever by half instead of starting where most economists would start with things like supply and demand.

ES: I really appreciate that about your writing. I follow a lot of economists on Twitter and they often jump immediately into a very technical analysis. It is helpful to start from first principles because the more complex and convoluted the argument, the less substance there often is to it. It seems like an easy way of hand-waving away an opponent by saying they didn’t study enough economics.

BA: That is definitely a debate strategy. Even within technical economics, there are people who want to throw the kitchen sink at a problem and think about all these bells and whistles together. An old way to think about it is whether your model has a million moving parts or narrows down to one or two. Both serve different purposes.

In the public discourse, I think there is a real benefit to starting these conversations in the basics. Especially for something like AI, where we can take for granted that a lot of things are going to change. It is not enough to just say the world is going to change; let’s try to ground the analysis in something concrete.

ES: Let’s start there. You wrote a blog post, “You Are Not a Horse,” which responds to the familiar argument that tractors replaced horses, and AI can replace human labor, therefore humans will eventually suffer the same economic fate as horses. You argue that this analogy skips several important steps between automation and eliminating human labor demand across the economy. Walk us through that argument and why it fails.

BA: Let’s start with how you would get to that conclusion. If an AI could do everything that I could do, including interacting with the physical world through robotics, then it could do everything I could do and there would be nothing left for me. I do not think anyone thinks we are remotely close to that in general. Economists call that perfect substitutes.

The question is: in the world where we are a little bit different, in the same way that the car and the horse are a little bit different, what are the ways of thinking about how humans and AIs are different? Where can they replace us, and where can they not?

One thing I stressed at the end of the piece is taking a step back and thinking about the economy as a whole. One thing that is interesting about humans is the huge diversity on the supply side and the demand side. We can do all sorts of things, and we like all sorts of things. Those forces are different than if we are just looking at ATM tellers as all workers. Thinking about what happens to ATM tellers or telephone operators is not the same thing as workers overall.

What I tried to do in that piece is build out the logic starting from one very narrow task. Let’s take type-setting as an example. That is something I do in my job; I type words out. What happens when that can be replaced? There is still verification and other steps. The most important thing to think about is the consumer side.

If I need someone to do my taxes and I decide the AI can do that for cheaper, I have spent less money. What happens with that extra money I have saved? I can spend that money on something else. When you add this up across all sorts of jobs and tasks, the extreme situation in which humans have no role in the economy requires that every dollar I save at the accountant is spent only on more AI-related stuff. That would have to happen continuously until everything I am spending is on AI. That is where you get this extreme horse condition. It is an edge case, and I am trying to figure out when it would actually happen.

ES: That argument necessarily goes to that extreme to have any real impact. If it does not go to that extreme, you get to a middle ground where we just reorganize the economy, which has happened a lot. That is not scary; that is just a technological innovation cycle.

BA: That is the fundamental tension between people worried about jobs going away and how economists think about it. If you look at history, farming is a great analogy. It used to be a huge part of the economy in terms of workers, and now it is very small. People have been worried about this since the 1930s and 1980s, thinking we won’t be able to compete with mechanized machines and will go the way of the horse.

But every time this has happened in the past, people have found new things to spend those savings on. Now that I pay less for a telephone operator, I can buy more healthcare or more services. I run as a hobby, so I might spend extra money on a run coach. As we get richer, we find new things to spend money on. The labor share, meaning how much income goes to workers, has remained basically the same for centuries.

ES: Can we trust any of the data yet? You see charts from the Financial Times saying AI is not having any meaningful impact, and others showing a statistically significant impact at entry-level roles. Data from Indeed shows software engineering job postings have actually increased. Is it too early to tell?

BA: I would say it is quite messy still. Some news outlets reported that IT outsourcing to India has gone up, which seems like something that would be automated. A paper by several economists called “Canaries in the Coal Mine” suggests that entry-level jobs, particularly those more exposed to AI, are affected. That seems to be holding up.

But I think it will be a while before we know what happened to current jobs. It is just such a mess with so many things pulling in different directions. How much is general labor trends versus AI-centric stuff? Implicitly, if we are going to say this is because of AI, we have to have some counterfactual.

For an analogy, there was a lot of discussion about a productivity boom in the 90s related to IT. Some of it showed up in IT producers like Microsoft, and later it showed up in heavy users like accounting firms. That debate was still active a decade later. I would expect the same thing with AI. In five years, we will have more of a sense of what happened in 2024, but it still won’t be crystal clear.

ES: There would be a natural tendency to slow things down if it felt like we were headed into that extreme case, even by the labs themselves. If you follow a trajectory where AI absorbs everything, at some point, the labs train a new use case that wipes out an entire industry. For example, web development. If everything gets absorbed into these models, at some point the labs have to say, “Wait a second, if we keep down this path, there are no customers for us.”

BA: I don’t know if that is the angle I would go. Imagine there are four big companies with models, the equivalents of OpenAI, Anthropic, Google, and xAI. You might have a commons problem where the overall consumers are spread across all four labs. If I drive one company out of business, that doesn’t necessarily affect me because they were actually purchasing from the other guys.

It is sometimes nice to have consumers versus bringing that all in-house. It looks a little bit different than previous discussions of vertical integration, but I don’t think we need to throw out what we have learned about the trade-offs there. There are lots of things gained by vertically integrating, but we don’t have one big vertically integrated firm in the U.S. We have competition because as things get big, other forces take hold.

Models that specialize have pulled ahead on some margins. Within economics, there is a product called Refine which reads academic papers. They have really leaned into their niche. I think they are able to do things that general models are unable to do. That might be short-term, but usually, some specialization makes you a little bit better.

ES: Why does the discourse weight so heavily on the side of negative outcomes? Most people use these tools and are more productive as a result. What are the potential positive outcomes? Maybe that middle ground is the best place to be.

BA: People have written scary stories about the middle. Daron Acemoglu has a paper showing you get a little productivity boost but not enough to grow the pie so extremely that you can do everything. I don’t think that seems super plausible to me. If AI is productive, the pie grows bigger. If it is somewhat productive, the pie still grows bigger. The better the AIs are at doing jobs, the more you are able to do and free up time for other stuff.

Dishwashers and laundry machines freed us up to do other things. Across the world, working hours go down as you get richer. Some of it shows up as easier jobs. My dad was a farmer; I sit and talk on podcasts. Even if we put in the same amount of hours, he worked a lot more than I do. We take those productivity improvements in lots of different ways.

ES: What about the policy response? I think this is one where I don’t quite know where I come down. Over the last hundred years in the U.S., technological improvement has generated riches used to expand social welfare through programs like Social Security, Medicare, and Medicaid. Does every crisis mean the government is great at responding and optimally redistributing gains? No, but there is some policy response.

BA: I reluctantly included that at the end of my piece because it takes us in a different direction, but you could have a policy response that makes things much worse. The initial response to the Global Financial Crisis was perhaps too weak to address the actual risk. The analog here would be if a bunch of people are driven out of jobs and unemployment insurance doesn’t kick in enough.

But there is also the risk of overreacting in the wrong direction. Most of the policy proposals I have seen are about actively preventing productivity enhancements, like taxing or banning data centers or compute. That is not about the response being too weak; it is about actively discouraging progress.

ES: I would categorically oppose anything like that, especially a total cessation of development on AI. You might be able to do that in the U.S., but you can’t do it in China. You shouldn’t want to; you should want to harness the power effectively.

I want to move on to personalized pricing. You had an op-ed in the Financial Times recently titled “Eliminate personalized pricing and you risk harming consumers.” It responds to a wave of state legislation intended to prohibit companies from using consumer data to set different prices for different people. You argue this could eliminate discounts and leave consumers worse off. Walk us through your basic argument.

BA: The starting point is why you want to offer different prices. In many markets, you don’t see this. Most firms have what economists call market power, which just means that if the firm lowered its price a little bit, someone out there would be willing to buy who didn’t buy at the current price. But you don’t do that because if you drop the price, you have to give up money from the people who were already buying it at the higher price.

There is a fundamental tension: businesses want to sell more but don’t want to lower prices for everyone. So, what do you do? You offer coupons. A coupon creates two different prices. The way it prevents everyone from taking the lower price is that some people just won’t fuss around with cutting coupons. This shows up in airline tickets, different sizes of coffees, etc. You create different products that allow people to sort themselves into different bins.

Personalized pricing gets around the wasteful activity of coupons. It says, “I am going to target a coupon to you.” You don’t have to cut it out. Because I am targeting it to you, I don’t have to offer it to everyone else. Therefore, I can gain the additional sale, and you as the consumer are better off because you bought something you wouldn’t have otherwise. A pure ban is not justified.

ES: Price discrimination is not a new idea. Walk us through the various flavors of it.

BA: Economists sometimes talk about first, second, and third-degree price discrimination. I would call it group-based discounts, like student or senior discounts. It is not grounded in their individual data per se; it is offered to all seniors. That comes with a cost because you need to verify those things.

There is a self-selection system, sometimes called second-degree price discrimination. That is my airline ticket example. Everyone could buy first class or the basic economy price, but people with company accounts or more money will select into the higher-priced bins. Then there is individualized pricing, which is the newer version where I have data particularly on you, and I send you a coupon. In practice, the difference between you and you as a senior is not particularly clear.

ES: One of the issues here is “surveillance pricing.” People seem very worried about firms collecting data on them. People say they don’t like when advertisers collect data, but it turns out they really like seeing ads for things they like. There is a sense of fairness that if rules are transparent, I understand why a senior gets a discount and I don’t. But when it happens in the background, people think it is unfair.

BA: I want to push back on two things. First, let’s acknowledge that if we can’t do this sort of price discrimination, the alternative is not that we all get the lowest price. No one believes that if we banned coupons, everyone would get the coupon price. There is some trade-off there. When Consumer Reports published a study on Instacart, they claimed AB testing on prices was costing consumers a certain amount. Their implicit counterfactual was that everyone would get the lowest price seen by anyone. That is not serious.

Second, just calling this “surveillance pricing” to scare people is not helpful. We need to think about the core economic trade-offs. Those trade-offs are there in the coupon case too, but that doesn’t come with a scary term. Just because the FTC used the word “surveillance pricing,” let’s not fall into that.

ES: Prices are always set based on data, even if it is just a guy at a counter noticing they aren’t busy and deciding to drop the price. When you optimize prices, the firm does better, they might hire more people, or invest in innovation. You can’t assume you are better off if the price is the lowest possible and the item sells out so some people can’t get it.

BA: It is just a different way of competing for the good. At the end of the day, you have a scarce resource, like World Cup tickets. If the price were zero, the number of people who want to buy would be much greater than the tickets available. You have to have some way of rationing that. One way is to set prices. Most firms let the price do a lot of the work.

Taylor Swift does not choose to take all the possible profit on ticket sales. She sets the price such that people have to log on at a particular time to get them. You need some way to allocate these goods. One of the beautiful things about prices is that if I am willing to pay more, that gets transferred to the seller. If I am waiting in line, that time is just wasted. The seller doesn’t get any benefit from me waiting in line; that is a deadweight loss. You want to harness prices when you can so you don’t generate all this waste. The waste is the coupon-cutting or the waiting in line. In the sake of some fairness, we might be willing to deal with some of that waste, but we should recognize the trade-offs.

ES: The digital economy personalizes everything—recommendations, search results, podcasts. Firms have been more cautious about personalizing the transaction price itself. Why has that remained the sensitive frontier of personalization?

BA: It seems to be consumer-driven. Consumers have a different reaction to price differences. When Wendy’s was thinking about implementing dynamic pricing—which is not the same as personalized pricing, but rather higher prices during busy hours—there was a big outcry. People worry about uncertainty. It could be that now I worry I could get a better price elsewhere and I search around. Amazon has an incentive not to make me search around. I think it is really coming from the consumer side; people have a sense of fairness that changes over time, and sellers respond to what consumers are going to get upset about.

ES: Brian, this was great. How can people engage with you online?

BA: They can yell at me on BlueSky, or follow me on Twitter at BrianCAlbrecht. Economic Forces is the newsletter on Substack. Those are the main things.

ES: I will link to both of those articles and your account in the show notes. Brian, thank you so much for joining me today and sharing your wisdom.

BA: Thanks so much.