The AI Layoff Trap
Read full paper here - https://arxiv.org/pdf/2603.20617
The basic situation
Imagine there are 10 pizza shops in your town. They all hire delivery drivers. Then a company invents delivery drones that are cheaper than paying drivers.
Here's the problem: those drivers are also customers. They spend their paychecks at pizza shops, movie theaters, clothing stores — everywhere. If they lose their jobs, they stop spending money.
So why don't the pizza shops just... not fire everyone?
This is the puzzle the paper is solving. Every shop knows that if everyone fires their drivers, the whole town gets poorer and pizza sales drop. They can see the cliff. So why do they drive toward it anyway?
Here's why: imagine you own one of the 10 pizza shops. You're thinking about switching to drones.
If you automate, you save $10,000 in wages
But the demand destruction — your fired drivers spending less money around town — gets spread across all 10 shops
So you only feel 1/10th of the damage you caused. The other 9/10ths hits your rivals
It's like littering. Throwing trash out your car window is convenient for you, but the ugliness is shared by everyone. Except here, your rivals also litter, and the whole town turns into a dump — including your own front yard.
Why can't they just agree to stop?
You might think the 10 shop owners could just meet up and agree: "nobody fires their drivers."
But here's the catch — even at that meeting, every single owner is secretly thinking: "If the other 9 shops keep their drivers, and I switch to drones, I save money AND everyone else absorbs most of my demand damage. I should defect."
And everyone thinks this simultaneously. So the agreement falls apart before it even starts. This is called a Prisoner's Dilemma — everyone acting in their own interest leads to everyone being worse off, including the shop owners themselves.
Who gets hurt?
Everyone, actually:
It's not like the owners get rich while workers suffer. The paper shows it's pure waste — like burning money that belonged to both sides.
What fixes it?
The paper goes through a bunch of ideas:
Universal Basic Income (give everyone cash): Helps people pay rent, but doesn't change the shop owner's math at all. They still automate at the same rate.
Tax the profits: Same problem — doesn't change the decision to fire drivers, just changes how much money you keep afterward.
Workers own shares in the company: Helps a little, but can't fully solve it because the damage spills over to other companies' customers too.
Shop owners make a deal: Doesn't work — as explained above, everyone has an incentive to cheat.
The only thing that actually works: A tax specifically on each worker you replace with a robot. This forces the shop owner to feel the full cost of what they're doing to the town's economy, not just 1/10th of it. Now their math changes, and they automate less.
The weird twist
Better AI actually makes this worse, not better. If drones get twice as fast, every shop rushes even harder to switch — hoping to grab customers from rivals. But since all shops do it at once, nobody gains any customers. They just all fired more people for no extra benefit. The race accelerates for no collective gain.
