The AI Layoff Trap - Interview
1. Why do you describe automation as a “prisoner’s dilemma”?
Maybe an analogy helps here: Say you’re at a steakhouse with a big group splitting the bill evenly. On the menu, there’s a fancy steak. It’s way more expensive than anything you’d normally order by yourself. But since the cost gets split across everyone, your share is only a little. So you order it. Except everyone at the table is doing the same math and they all order the fancy steak too. In the end---and there’s an actual econ paper about this---everyone ends up with a higher bill. That’s a type of prisoner’s dilemma: individually rational, collectively bad. My coauthor Brett Falk and I argue AI layoffs can work the same way. The company that automates keeps the full cost saving. That’s the fancy steak. But the downside, workers with less income to spend, gets spread across every company in the industry. That’s the bill getting split. Each firm only feels a small slice of the damage it caused, so every firm has the same incentive to automate. And everyone ends up with weaker demand, because the customers they rely on are the workers they just fired.
2. Proponents of AI say that it will create new jobs we can’t even foresee, therefore we don’t necessarily need to worry about the job displacement it creates. Do you find this to be a compelling argument?
Historically yes, it has been mostly true and there are plenty of academic studies that show this. But for this AI wave, no one actually knows at the moment, everyone is just speculating, albeit sometimes with conviction. Maybe eventually things will stabilize like they have before. But even so, timing matters. There’s a gap between losing your job and finding the next one, if it ever comes, and in that gap, your paycheck is gone. Our paper is about what happens in that gap. “It’ll work out eventually” is cold comfort if you’re the one living through the “eventually.”
3. I had a hard time following why UBI wouldn’t necessarily work on its own. Can you explain it briefly? Is it because it wouldn’t change the incentives for businesses so we’d still end up with a bunch of people out of work?
You’ve got it. UBI is a check to households. Which is nice if you’re a household. But it doesn’t change the math a company runs when it’s deciding whether to replace the next worker with AI. They’re comparing what the worker costs to what the AI costs, and that comparison doesn’t change whether or not the government is sending checks to people. So you still get the layoffs. You still get the lost spending. And in our model, there’s even a scenario where UBI can make things worse. If firms can freely enter the market, the extra consumer spending from UBI attracts more firms in, and a more crowded market automates harder. So UBI might end up partially working against itself. That doesn’t cancel out its benefits, but it complicates the picture. Think of it this way. UBI is like putting buckets under a leaky ceiling. It catches the water, which matters, but it doesn’t fix the leak.
4. What would UBI be good for, in your view?
Cushioning the people who get hit. Which matters. But on its own, UBI is a painkiller, not a cure. Where it really shines in our framework is when you pair it with a tax on automation. The tax changes what companies do. UBI, funded by that tax, helps the workers who still get hurt along the way. They work together.
5. Do you have an idea of how an automation tax would actually be implemented? What would the IRS actually look at?
The implementation details deserve more study, but we have some preliminary ideas. At a high level, companies would pay a small fee when they replace a worker with AI, and enforcement could build on records the government already tracks. The bigger point, for us, with this paper, is flagging a gap in the current debate. Most of the conversation we hear is about what to do after workers are displaced. We’re pointing to the incentive that drives the displacement in the first place. In our model, a fee like this would directly reduce the pressure to over-fire. And if we’re right, both workers and firm owners would end up better off, because over-automation in the model is hurting everyone, including the companies doing it.
6. What do you say to those who say taxes like this would stifle innovation?
It gets the direction backwards. A tax like this doesn’t punish innovation. It just makes companies account for the damage their layoffs impose on the rest of the market. Carbon taxes didn’t stop energy innovation. They helped it. Same idea here. One more thing. In our model, this kind of tax should be self-limiting. As laid-off workers find new jobs, the right tax rate shrinks on its own. So you don’t need it forever. It phases itself out as the economy adjusts.
7. Does anything about your paper or what you’ve studied keep you up at night?
I should probably clarify that I’m generally pro-AI, I use it every day. But there are at least two things that worry me.
First, the displacement gap. I see it from two sides. My students ask whether the entry-level jobs they’re training for will even exist when they graduate. The executives I advise are being told to shrink their teams and don’t know how to reskill the people left. McKinsey is projecting that generative AI will push millions of people into career switches by 2030, but nobody can actually tell you yet what exactly they’re switching to. The gap between the speed of the displacement and the speed of the response, that’s something I think about a lot.
Second, and this might be the bigger one, the politics of all this. I was in banking before going into academia, so I’ve lived through what happens when an industry loses the public’s trust. What I’m watching right now reminds me of the run-up to 2008. Enormous amounts of capital pouring into a single technology, at the expense of a lot of other things. A lot of people getting left behind. A growing sense from the public that no one is listening. The parallel I’d draw is Occupy Wall Street. The banks didn’t see that coming. The ones that came out best were the ones who’d been thinking about it before it arrived. I don’t think the leading AI companies have fully internalized the cost of losing the public. Once the trust is gone, it’s very difficult to get back, and I feel it slipping at the moment. That’s not just a policy risk. It’s an existential risk. The companies that get ahead of this will end up in a very different position from the ones that don’t.
8. What do you want readers to take away from your paper?
That the real story isn’t AI versus workers. It’s companies competing themselves into a corner. In our model, seeing the problem coming doesn’t stop it. More competition makes it worse. Better AI makes it worse. UBI, retraining, profit-sharing, companies trying to negotiate with each other, none of those seem to touch the underlying problem. The only thing that does, in the model, is making companies pay for the damage their layoffs do to others. Until that happens, the race doesn’t end. Not because anyone’s a villain. Because everyone is rational and they are caught in a trap.
9. What is the wrong takeaway from your paper?
A few things. First, this is not an anti-AI paper. We use AI every day. AI that augments workers is fine. AI deployed thoughtfully is fine. The problem in our model isn’t the technology, it’s what happens when every firm in a market races to deploy it at the same time. That’s a coordination problem, not a technology problem.
Second, we’re not predicting collapse. The first word of our abstract is “if.” People keep dropping it. Whether the risk actually materializes depends on how fast displaced workers find new roles, and that’s something both companies and policymakers can actively influence.
Third, this isn’t about villainizing companies. In our model, they’re rational actors responding to competitive pressure, like everyone else. Our work is about helping the industry think through a dynamic that no single firm can fix alone. The companies that get this right, that deploy AI to augment rather than just replace, that invest in the transition, will end up in a much stronger position. We see those companies as partners in getting this right, not as targets.