Faster AI adoption can make technological transitions worse: compressing displacement into a short window overwhelms retraining pipelines, raises permanent exits and depresses the labor share, and social welfare is maximized at an intermediate adoption speed—implying a role for policy to slow adoption or expand retraining capacity.

Main Finding

Faster AI adoption—holding the eventual level of automation fixed—worsens transition outcomes because it compresses the displacement window, overwhelms finite retraining capacity, and raises permanent labor-force exit through worker discouragement. As a result, adoption speed alone determines transition welfare: social welfare is strictly concave in adoption speed and is maximized at an interior rate below the unregulated (market) speed. Stronger retraining institutions raise the socially optimal adoption speed.

Key Points

• Model setting: displaced routine workers enter a retraining pipeline with finite throughput; adoption speed is an exogenous parameter that controls the rate at which routine tasks are automated.
• Congestion mechanism: faster adoption increases the flow of displaced workers into the retraining queue over a shorter time, creating congestion that reduces successful re‐entry into employment and raises permanent exits from the labor force via discouragement.
• Transition vs. long run: two economies with identical long-run automation shares can have very different transition outcomes purely because of differences in adoption speed.
• Labor market dynamics:
  • Labor force participation falls more sharply and persistently under faster adoption because of larger discouraged-worker stocks.
  • The aggregate labor share is compressed throughout the transition window under fast adoption.
  • Non-routine employment and wages show a crossing pattern: initially higher under fast adoption (due to faster reallocation of tasks/wages for survivors) but later lower; faster adoption can both raise long-run wages for surviving non-routine incumbents and permanently reduce overall participation.
• Welfare and externalities:
  • Firms setting adoption speed do not internalize the congestion externality on retraining, the irreversibility of permanent exits, or the wage effects on non-routine incumbents; hence the private (market) adoption speed is higher than the social optimum.
  • Social welfare as a function of adoption speed is strictly concave, with an interior maximum below the market rate.
• Policy complementarity: the socially optimal adoption speed increases when retraining capacity (or other institutional supports) is stronger—i.e., adoption speed and retraining capacity are complements in welfare optimization.

Data & Methods

• Methodology: a dynamic theoretical model of technological transition and labor reallocation. Core ingredients:
  • Routine workers displaced by automation enter a retraining pipeline with limited capacity.
  • A parameter governs the speed of AI/task adoption (rate of displacement); the long-run automation level is held fixed across counterfactuals.
  • Worker discouragement captures permanent labor-force exit when retraining waits are long or prospects dim.
  • Welfare analysis compares decentralized (market) adoption choices to a planner who internalizes externalities.
• Analysis approach: analytical characterization of dynamic equilibria and comparative statics with respect to adoption speed and retraining capacity; supplemented by numerical simulations to illustrate transition paths and quantify welfare differences (model-calibration details, if any, were not provided here).
• No new microdata were required to develop the mechanism; the results are driven by model structure (finite retraining throughput + irreversibility of exit + adoption timing).

Implications for AI Economics

• Modeling: macro and labor models of automation must treat adoption speed as a key policy-relevant margin, not just long-run automation shares. Short-run congestion and hysteresis matter.
• Policy design:
  • Slow, managed adoption (or throttling mechanisms) can improve social welfare by reducing retraining congestion and discouragement.
  • Investments that expand retraining throughput or reduce the cost/duration of re‑skilling (training capacity, accelerated programs, on‑the‑job retraining) increase the socially optimal adoption speed and mitigate transition harms.
  • Policy tools could include time-phased tax incentives/subsidies for automation, temporary hiring or retraining subsidies, public provision/expansion of retraining capacity, or regulation that internalizes the externality (e.g., automation taxes earmarked for retraining).
  • Protection should consider both displaced workers (reduce permanent exit) and non-routine incumbents (wage effects), since non-routine wage depression is a welfare channel often overlooked.
• Empirical predictions and testing:
  • Faster AI adoption episodes (e.g., rapid deployment of automation in a sector) should be associated with sharper declines in participation, larger discouraged-worker stocks, and more persistent reductions in the labor share than slower adoptions with the same ultimate automation level.
  • Wages and employment in non-routine occupations may show a crossing pattern across adoption-speed regimes: short-term gains followed by longer-term underperformance under faster adoption.
  • Testing requires data on adoption timing/intensity, retraining program capacity and throughput, labor force participation dynamics (CPS, LFS), discouraged-worker measures, sectoral labor shares, and wage dynamics.
• Policy implication for institutions: strengthening retraining institutions is doubly beneficial—it directly mitigates transition harms and allows society to tolerate (and benefit from) faster adoption without generating large welfare losses from congestion and discouragement.