Paper warns mass automation risk
A game‑theory paper from UPenn and Boston University models a scenario where firms automate to cut costs but collectively cause demand to collapse—a Prisoner’s Dilemma that the authors say could prompt proposals like a 'robot tax'. The thread has drawn wide attention online as a concrete framing of the macro risks from coordinated automation. (x.com/pvergadia/status/2043430961057058888)
A new economics paper argues that firms can automate themselves into weaker sales, even when every firm can see the risk in advance. (arxiv.org) The paper, “The AI Layoff Trap,” is by Brett Hemenway Falk of the University of Pennsylvania and Gerry Tsoukalas of Boston University. It is dated March 2, 2026, and was posted to arXiv on March 21, 2026. (arxiv.org) The core idea is simple: workers are also customers. If companies replace people with artificial intelligence faster than the economy creates new work, household income falls and the demand for goods and services falls with it. (arxiv.org) The authors model that problem as a Prisoner’s Dilemma, the game theory setup where each player has an incentive to make the individually rational move even when everyone ends up worse off. In their version, each firm still wants to automate because holding back alone means eating the revenue hit from rivals’ layoffs without getting the same cost savings. (arxiv.org) Their result is not that automation is always bad. It is that competition can push firms to automate “well beyond” the level that is collectively optimal, harming workers and firm owners together. (arxiv.org) The paper also tests a list of common responses and says they do not remove the trap in the model. The authors write that wage adjustments, free entry, capital income taxes, worker equity participation, universal basic income, upskilling, and private bargaining still leave the core incentive to automate in place. (arxiv.org) What does change the outcome in their model is a Pigouvian automation tax, a levy meant to make firms pay for the wider economic damage their layoffs impose on everyone else. The authors say that kind of tax targets the decision to automate, not just the fallout after jobs disappear. (arxiv.org) That argument lands in an older policy debate. A 2022 National Bureau of Economic Research working paper by Martin Beraja and Nathan Zorzi also found that firms can automate too quickly because they do not internalize the costs borne by displaced workers, and it estimated welfare gains from slowing automation in its model. (nber.org) Supporters of a robot tax have been making a related case outside academia. A 2024 Brookings commentary argued that a tax on companies deploying autonomous artificial intelligence and robotics could both fund support for displaced workers and make firms think harder about marginal automation decisions. (brookings.edu) The new paper is a theory paper, not a measurement of the whole economy, and its conclusions depend on the assumptions built into the model. Its claim is narrower and sharper: if business incentives look like this, voluntary restraint will not stop an automation race on its own. (arxiv.org)
