Researchers model an ‘AI layoff trap’
A team at UPenn and Boston University published a paper called 'The AI Layoff Trap' arguing mass automation can create a prisoner’s dilemma where firms cut jobs to stay competitive but collectively destroy demand. The paper frames mass automation as a systemic risk with limited easy policy fixes and suggests ideas like an automation tax as part of the debate. (x.com)
A new paper argues that widespread artificial intelligence layoffs can become a trap: each firm saves money by automating, while all firms together shrink the customer base they need to sell to. (arxiv.org) The paper, “The AI Layoff Trap,” was posted to arXiv on March 21, 2026, by Brett Hemenway Falk of the University of Pennsylvania and Gerry Tsoukalas of Boston University. Its abstract says firms can displace workers “well beyond what is collectively optimal” even when managers understand the risk. (arxiv.org) (upenn.edu) (bu.edu) The basic mechanism is simple: workers are also consumers, so wages do double duty as labor costs for one company and demand for every other company’s products. The authors model that feedback as a “demand externality,” meaning each firm ignores part of the economy-wide damage when it cuts labor. (arxiv.org) Their framework uses a prisoner’s-dilemma logic: if one firm automates tasks and rivals do not, the automating firm gains an edge; if all firms automate aggressively, total output may rise while purchasing power falls. In the paper’s setup, competition pushes companies into an automation race they would not choose if they could coordinate on the broader outcome. (arxiv.org) The authors say the result gets stronger when markets are more competitive and when artificial intelligence gets “better,” because both conditions make it harder for any one firm to hold back. They also write that lower wages and new firm entry do not solve the problem inside their model. (arxiv.org) (ideas.repec.org) That puts the paper in the middle of a live policy fight over whether artificial intelligence mainly boosts productivity or also creates a macroeconomic demand problem. The authors explicitly test several popular responses and say capital income taxes, worker equity participation, universal basic income, upskilling, and Coasian bargaining do not eliminate the distortion in their model. (arxiv.org) Their headline policy result is narrower and sharper: they write that “only a Pigouvian automation tax can,” referring to a tax meant to price the spillover cost firms impose on the rest of the economy. The paper does not present that as an easy political fix; it presents it as the only remedy that closes the gap in the model. (arxiv.org) The paper is theoretical, not a measurement of current layoffs, and that matters for how to read it. It does not claim that every artificial-intelligence deployment destroys jobs or that the economy cannot create new work; it claims that under specific competitive conditions, individually rational automation can produce a worse collective outcome. (arxiv.org) That distinction is likely to shape the next round of argument over artificial intelligence and employment: not whether firms have an incentive to automate, but whether the economy has a way to absorb the losses before demand gives way. (arxiv.org)
