Displacement
The labor-side counterpart to capital-labor-divergence: the share of human work that AI substitutes for rather than augments, and where the cost of that substitution actually lands in the labor market. Displacement is distinct from augmentation along a single methodological axis — augmentation raises productivity of an existing worker, displacement reduces the marginal demand for that worker — but in practice the two run together, and the observed-exposure-measure is constructed precisely to weight them differently (full weight for automated use, half weight for augmentative use). The point of separating the concepts is to refuse the bull-frame elision under which “AI just helps people work better” describes everything happening at once.
Theoretical vs observed: the measurement debate
The pre-2026 displacement literature (acemoglu-simple-macroeconomics, Eloundou et al., Felten–Raj–Seamans) projected displacement risk from theoretical task exposure — what an LLM could do, scored from O*NET task descriptions. Acemoglu predicted incidence would fall on low-education clerical women. massenkoff-mccrory-labor-market-impacts-2026 introduces the observed-exposure-measure which weights theoretical capability by realized Claude work-context usage, and finds the actual exposed cohort is more educated, more white/Asian, older, and 47% higher-paid than the population (exposed-worker-demographics). The gap between the two measures is itself the story: capability has run ~3× ahead of deployment in the most-exposed category (theoretical-vs-observed-capability-gap), and which projection eventually plays out depends on whether the diffusion lag closes evenly or stays structurally uneven. This is a live disagreement, not a closed question — the wiki’s posture is to track both predictions in parallel.
Why unemployment dashboards under-detect it
The headline empirical finding from the same 2026 source is that conventional unemployment statistics do not detect AI displacement at the current scale. The unemployment-did-exposed-workers DiD is flat post-ChatGPT (+0.002, SE 0.002, insignificant at the ~1pp differential detectability threshold). The only signal that survives is ai-young-worker-hiring-slowdown: a ~14% drop in 22–25-year-old job-start rates in exposed occupations, with senior incumbents unaffected. Brynjolfsson et al.’s ADP-payroll companion finding (6–16% employment drop in the same cohort) corroborates it. Two consequences fold back into the rest of the wiki: (1) displacement currently lands as suppressed entry-level hiring inside knowledge work, not as broad-based layoffs — the incidence is intra-class (incumbent vs entrant), not inter-class; (2) the surplus is captured by capital owners of L4–L6 application-layer firms (application-layer) plus the firms saving on junior salaries, with labor seeing neither side of the wedge. The task-based-framework macro envelope still bounds the magnitude (<0.71% 10-year US TFP), but the distributional story sits inside that envelope and is sharp regardless.
Related
- capital-labor-divergence — the macro frame this concept operationalises on the labor side.
- observed-exposure-measure — the usage-weighted measure that reframes the incidence question.
- theoretical-vs-observed-capability-gap — the ~3× gap that makes the projection contested.
- exposed-worker-demographics — who the realized-exposure cohort actually is.
- ai-young-worker-hiring-slowdown — the only detectable signal; entry-level channel.
- unemployment-did-exposed-workers — the null finding that makes the displacement invisible.
- acemoglu-simple-macroeconomics — the prior theoretical projection partly contradicted by observed data.
- massenkoff-mccrory-labor-market-impacts-2026 — Anthropic 2026 source.
- eloundou-beta-exposure — theoretical scoring used as input to observed measures.
- task-based-framework — macro envelope inside which the displacement story sits.
- fusion-skills — the augmentation pole this concept is defined against.
- autonomy-slider — the design surface that makes the displacement/augmentation distinction visible.
- diffusion-vs-innovation — diffusion is what scales displacement as deployment catches capability.
- application-layer — where the L4–L6 surplus-capturing products live.
- bear-case-synthesis — the broader frame this concept updates.