There's a new paper making the rounds called "The AI Layoff Trap", by Brett Hemenway Falk and Gerry Tsoukalas, and it arrives with a headline that's catnip for anyone looking to wave off basic income. Among the policies that can't fix the problem of AI-driven layoffs, they list universal basic income. Right there in the abstract, sitting next to capital income taxes and upskilling and worker equity. Only one tool makes the cut: a Pigouvian tax on automation.
I want to take this paper seriously, because the core idea is genuinely good. And then I want to show you exactly where it goes wrong about UBI, because it's the kind of error that's easy to miss. It's simply a flawed assumption.
The Good Part
Here's the automation trap the authors describe, and it's a real one.
When a company replaces workers with AI, it pockets the full savings. But those laid-off workers were also customers. Their vanished paychecks mean vanished spending, and that lost spending doesn't only hit the company that did the firing. It hits everyone selling anything. The catch is that each firm eats only a small slice of the demand it destroys. The rest lands on its competitors.
So every firm faces the same math: automate, capture all the savings, and shove most of the pain onto rivals. Everybody reasons this way at once. Everybody automates past the point that's good even for the owners. This is the classic tragedy of the commons problem. Everybody eventually ends up worse off. And it gets worse the more competitive the market is, and worse the better the AI gets. "Better" AI doesn't save you. It just speeds up the race toward the cliff.
I think it’s useful to explain all of that. It puts some math under something a lot of us have been saying in plainer words for years: an economy that automates away its own customers is sawing off the branch it's sitting on. Credit where it's due.
The Flaw That’s Baked In
Now watch what happens to UBI.
In their model, a basic income is a single thing: a constant. They call it A, for "autonomous demand." Everyone gets the same check whether they're working or laid off, so it adds a flat amount of spending and then drops out of the equation that decides whether a firm automates. Their conclusion follows immediately. A flat number that doesn't change the marginal cost of firing one more person can't change the decision to fire one more person.
But look at what just happened. They defined UBI as a thing that can't affect behavior, and then discovered that it doesn't affect behavior. The conclusion was embedded inside the assumption. To their credit, the authors say as much in a quiet caveat that the result "should not be read as a verdict on all UBI designs." That caveat deserved to be the headline. It wasn't.
Their own model hands me my rebuttal. Buried in the math is a dial they call η — the share of a displaced worker's lost income that comes back through "reemployment, transfers, or other sources." Turn that dial up and the trap shrinks. Turn it up far enough and the whole thing flips into under-automation. The authors let wage insurance and severance touch that dial. They just don't let UBI touch it.
Why not? Because they modeled UBI as a check that changes nothing about what a laid-off person does next. But that's not what a basic income does. A floor under everyone changes who can afford to retrain, who can afford to hold out for a good job instead of grabbing the first bad one, who can afford to start new businesses, who has the power to say no to low wages and insist on better pay for jobs that can’t be automated yet. Every one of those is η. UBI doesn't just raise A. It raises η. And the paper's own framework says raising η is exactly what closes the trap.
What the Model Can’t See
There's a second thing the model can't see, and it's the one I care about most.
In their world, a human-made thing and an AI-made thing are the same thing. So income only ever shows up as how much people spend. Never what they choose to buy. Money is money.
That's not how consumers work. Give someone a stable floor and they don't just spend — they spend differently. They pay for the live band instead of just the Spotify subscription. They pay extra for the handmade thing, the local thing, the human experience. Human labor becomes a premium — the new "Made in the USA,” but instead “Made by Humans”. That's a market signal for human labor even when it costs more, and spending power is what creates it. The model can't represent any of it, because its one lonely demand term can't bend in that direction. A real basic income doesn't just adjust the volume knob. It changes the song.
And we're not guessing about whether basic income ripples through the wider economy. Alaska has paid every resident a universal dividend since 1982, and when economists Damon Jones and Ioana Marinescu studied it, they found it had no effect on overall employment — and actually nudged part-time work up, consistent with the cash stimulating the local economy. A general-equilibrium effect. Which is precisely the dynamic feedback the model assumes out of existence the moment it freezes UBI into a constant.
The Harm Isn’t the Automation
Here's my deeper problem with the paper, and it's the one that matters most.
The authors call this a negative externality and reach for the textbook fix: a Pigouvian tax, the kind you'd slap on pollution — tax the harmful thing. But automation isn't pollution. Automating human labor is usually a good thing. It's how we get more for less. It's how a crew of twenty becomes a crew of five — and, in a sane system, how all of them get some of their lives back. Depending on the work, automation runs from neutral to genuinely wonderful. Almost none of it is harmful in itself.
So why does the model see harm everywhere a job disappears? Because of one assumption it quietly shares with our entire economy: that a person's income is coupled to their job. Cut someone loose from their paycheck and yes, their spending collapses, and yes, that ripples out onto everyone. But that chain reaction isn't a property of the robot. It's a property of the rule — our rule, the one we wrote ourselves — that says you only get to eat if you have a job, and you only get money if someone pays you to work.
That's the externality. Not the automation. The income-from-labor link. We built a system where one person's lost wage is the next person's lost customer, and then we act surprised when the system starts eating itself.
Automation can be a problem. I won't pretend otherwise. But it doesn't have to be. It only is so long as things stay exactly as they are — where people need jobs to live, and jobs need consumers who can spend. Break that link, even partway, and the trap loosens. Give people a floor that doesn't vanish the moment their job does, and a laid-off worker is still a customer, still spending, still part of the economy. A basic income is how you keep at least some of the wage without requiring the work. That non-zero floor will then enable more choices and also more income.
So Tax It — but Not as a Sin
Notice what the authors want their tax to do: slow the robots down, pull automation back to the level that's collectively "optimal." I don't want that. I want the robots to go as fast as they safely can. What I want to change isn't the speed of automation — it's who it's for.
So tax it, sure. Just not as a penalty for the crime of automating. Tax it the way you'd tax any rich, jointly-created stream of value: to fund the dividend and spread the gains. And here's the contest the paper sets up that was never a contest — UBI versus the tax. UBI has always needed funding, and the funding is the tax. One taxes and one distributes. They aren't opponents. They're two sides of the same coin. The paper even admits it, calling UBI "a complement to the automation tax, not a substitute" — a strange thing to bury after stamping "neither can universal basic income" across the abstract.
And I'd go further than one tax, because no single tax should have to carry an entire economy. Tax land value — the value the community creates, not the building on top of it, what Milton Friedman called the "least bad tax". Put a tiny tax on every financial transaction. And for the AI age specifically, I'd add a stock-dilution tax: require public companies to issue something like 1.5% of new shares each year into a national wealth fund whose dividends flow to all of us. If capital is going to eat labor, then let us own the capital, together. Fund the dividend out of the very automation that threatens the paychecks, and watch the incentives flip — because the more we automate, the more everyone gains.
What the Paper Actually Proved
Stated honestly, the finding is narrow and fine: in a stripped-down model where AI eats consumption and UBI is defined as a number that can't change anyone's behavior, UBI doesn't change anyone's behavior. Coherent, and almost entirely beside the point.
Because the real question was never whether flat cash, all by itself, changes a firm's automation math. The real question is what kind of world we build on the far side of all this automation. The trap the authors describe is real. The way out isn't to pick the tax over the dividend. It's to tax automation and hand the proceeds back to the people as a dividend that rises as the robots do.
UBI isn't the policy that fails to stop the trap. UBI is how we make sure that when the machines do the work, the people still get to live, and thrive.
And if you want to do more than nod along, this is exactly the moment I built the AI Pledge for Humanity for. It's a public stand for the simplest version of the idea — that the gains from AI belong to all of us, as a rising basic income floor framed as an AI dividend — and a way to hold the AI leaders who keep saying UBI is inevitable to actually helping make it real. Sign it. Share it. The people building the machines keep telling us this is coming. Let's make sure they help build the floor that catches us and uplifts us when it does.
Remember: jobs are for machines; life is for people.
