# Technological Unemployment **Job loss caused by technical change outpacing an economy's ability to create replacement work.** [[John Maynard Keynes]] named it in *Economic Possibilities for our Grandchildren* (1930), defining it as unemployment arising because our means of economising the use of labour outrun the pace at which we find new uses for labour. Keynes called it a **temporary phase of maladjustment** and predicted his grandchildren would work fifteen-hour weeks. He got the productivity growth roughly right and the hours badly wrong, which is the first thing worth knowing about this concept: **the optimists have generally been right about aggregate employment and wrong about distribution.** ## The long argument The dispute is older than the phrase. - **The [[Luddites]] (1811–1816)** were not technophobes. They were skilled textile workers destroying machines that let unskilled operators produce inferior goods at lower cost — a labor-market dispute conducted by sabotage, against which Britain deployed roughly 12,000 troops. The word became a slur for irrational resistance to progress, which is itself a rhetorical victory worth noticing. - **[[David Ricardo]]** added the chapter "On Machinery" to the third edition of *Principles* (1821), reversing himself in public: machinery could reduce the demand for labor and injure workers' interests, and the compensating effects were not automatic. The founder of comparative advantage conceded the point the tradition claiming him usually denies. - **[[Karl Marx]]** built it into his system as the reserve army of labor — technology as a permanent disciplinary lever on wages rather than an occasional shock. - **The postwar consensus** settled on the **lump of labor fallacy**: there is no fixed quantity of work, productivity gains raise real incomes, new demand creates new occupations, and displaced workers reallocate. Historically this has been correct at the aggregate level. Two centuries of automation have not produced a permanent rise in unemployment. The honest version of the optimistic case comes with a footnote the slogan version drops: **reallocation is real, but it is slow, geographically uneven, and generationally brutal.** American manufacturing regions demonstrate that a labor market can fully "adjust" in the aggregate while specific places and cohorts never recover. Aggregate employment statistics are a poor instrument for measuring that kind of harm. ## Is AI different? Two arguments say yes, and they are not the same argument. **Scope.** Previous waves automated specific physical or clerical tasks. General-purpose AI targets cognitive work across occupations simultaneously, which compresses the reallocation window — the new jobs may arrive, but not before the displaced cohort has aged out of retraining. **The ladder.** The more specific concern is that AI eats *entry-level* work first. Junior analysts, junior developers, paralegals, and coordinators are precisely the roles whose output is most legible to a model. If the bottom rungs disappear, the profession does not simply shrink — the pipeline that produces senior practitioners breaks. **The current evidence is early but not nothing.** Brynjolfsson, Chandar & Chen's *Canaries in the Coal Mine?* (Stanford Digital Economy Lab, November 2025) used administrative payroll microdata from the largest US payroll provider and found a **16% relative decline in employment for workers aged 22–25 in AI-exposed occupations**, with employment stable or growing for experienced workers and less-exposed roles. Adjustment came through headcount rather than wages, concentrated where AI **automates** rather than augments, and held after excluding tech firms and remote-amenable jobs. Read carefully, that is not "AI is destroying jobs." It is a **specific, age-stratified pattern** consistent with the ladder hypothesis — which is a narrower claim and a more disturbing one, because it is invisible in headline unemployment numbers. Sober caveats apply: one dataset, one payroll provider, an early window, and a period with confounding interest-rate and post-pandemic hiring effects. Treat it as the best available early signal, not a settled finding. ## Why it matters to the KB **It is the load-bearing premise of the tech right's political offer.** [[Dario Amodei]] has argued displacement may be "an intrinsic property of the technology" and proposed taxing AI firms to fund a basic income (see [[Policy on the AI Exponential]]). [[Sam Altman]] funded the largest US cash-transfer study and floats "universal basic compute." [[Elon Musk]] predicts "universal high income." [[Marc Andreessen]]'s [[Techno-Optimist Manifesto]] takes the opposite line — that displacement anxiety is itself the enemy. What unites the pro-UBI versions is the **structure of the bargain**: accept the displacement, and the winners will fund a floor. What the bargain omits is worker power, ownership, antitrust, or any constraint on deployment speed. It converts a question about who controls the technology into a question about the size of a check — and defers the check. **The Alaska parallel should be argued, not gestured at.** Proposals to fund a basic income from AI rents are structurally proposals to build an [[Alaska Permanent Fund]] for a different extractive resource. The disanalogy is the important part: Alaska's subsurface minerals were **publicly owned**, which is why the state could constitutionally claim a royalty. AI's rents accrue to private firms under no equivalent public obligation. Getting from here to a compute dividend requires a taxation fight the oil case did not — and the federal government's actual 2025–2026 experiment in state equity ownership ([[Sovereign Wealth Fund]]) has been built with no distribution rule whatsoever. **And the coalition is incoherent about it.** The same donor ecosystem financing the AI buildout funds the [[Guaranteed Income Preemption]] campaign banning municipal cash pilots. If displacement accelerates, that incoherence stops being rhetorical and becomes a funding fight — the thing to watch. **Finally, the measurement problem is political.** If the effect concentrates in entry-level hiring, it shows up as young people failing to start careers rather than as workers losing jobs. That is a form of harm the [[Unemployment Rate]] barely registers and [[Labor Force Participation Rate]] registers late — which means the political system will be slow to respond even if the effect is large. See [[Wage Growth vs Productivity]] for the previous era's version of gains failing to reach labor. ## See also [[Universal Basic Income]] · [[Alaska Permanent Fund]] · [[Sovereign Wealth Fund]] · [[Luddites]] · [[Artificial Intelligence]] · [[Policy on the AI Exponential]] · [[The Tech Right Hub]] · [[Techno-Optimism]] · [[Wage Growth vs Productivity]] · [[Economics Hub]]