0 cumulative citations
View corpus contextAutomation magnifies technology shocks to cut hours and wages across skill levels, with routine-task exposure lowering pay even for largely non-routine jobs; because AI disproportionately substitutes high-paid cognitive work, subsidizing AI may—counterintuitively—reduce income inequality.
Technologies d'automatisation et marché du travail : analyses théoriques et empiriques
openalex
other
medium evidence
7/10 relevance
Summary only
summary available; pdf_status=paywall
Source
PDF
Across theory and empirical work, the thesis finds that automation (including AI) amplifies technology shocks to reduce hours and depress wages across the skill distribution—routine-task exposure lowers pay even for largely non-routine workers—and optimal policy analysis suggests AI subsidies could reduce inequality by replacing high-paid cognitive labor.
Citation observations
Cumulative provider counts captured on specific dates; providers are never combined.
This thesis explores the implications of technologies capable of replacing human labor in various job tasks, such as artificial intelligence (AI), on employment and wages. Across four chapters, I address this question from both macroeconomic and microeconomic perspectives, combining theoretical and empirical analyses based on different data sources and econometric methods. The results can be summarized as follows. First, automation technologies amplify the impact of technology shocks on hours worked. Second, they reduce the wages of high-skilled occupations, as well as those of low-skilled ones, through ripple effects. Third, the automation of routine tasks puts downward pressure on workers’ wages, even when routine tasks represent only a minor share of their job, thus revealing a partial but significant exposure of non-routine workers to automation. Finally, I provide a normative analysis of AI taxation. Since AI differs from other automation technologies by its ability to perform complex cognitive tasks performed by high-skilled workers, the results suggest that subsidizing AI would be optimal, as it replaces the highest-paid workers and thereby contributes to reducing income inequality.
Summary
Main Finding
Automation technologies that can replace human tasks — and AI in particular — materially reshape labor markets: they amplify the effect of technology shocks on aggregate hours, depress wages across the skill distribution (directly and through ripple effects), expose even nominally non-routine workers to downward pressure when routine tasks are automated, and (because AI substitutes for high-paid cognitive work) create a welfare case for subsidizing AI rather than taxing it.
Key Points
- Automation magnifies the impact of technology shocks on hours worked: when technologies can substitute for labor tasks, productivity shocks propagate more strongly into labor supply and hours.
- Wage declines occur across skill levels:
- High-skilled occupations experience direct wage losses when automation/AI can perform complex cognitive tasks they do.
- Low-skilled occupations also suffer indirectly through ripple effects across the labor market.
- Routine-task automation reduces wages even for workers with only a small routine-task share in their job — non-routine workers are partially exposed, producing broader downward pressure than task-shares alone would suggest.
- Normative policy conclusion for AI differs from classic automation:
- Because AI substitutes for high-paid cognitive labor, its distributional effect can reduce top incomes.
- Under the model and welfare analysis in the thesis, subsidizing AI is optimal (rather than taxing it), since it can reduce income inequality by replacing higher-paid work.
Data & Methods
- Multi-chapter approach combining macro and micro perspectives.
- Uses both theoretical modeling and empirical analysis:
- Theoretical models to characterize how task substitutability changes propagation of technology shocks and to evaluate welfare/policy trade-offs.
- Empirical work exploiting cross-occupation and cross-region variation in task exposure to automation and AI.
- Employs diverse data sources and econometric tools (occupation/task-level measures, wage and hours data, and identification strategies to separate technology shocks from other shocks). Methods include structural and reduced-form estimation to uncover direct substitution effects and indirect ripple effects.
- Counterfactual and normative policy analysis to compare taxation versus subsidy of AI, taking distributional outcomes into account.
Implications for AI Economics
- Rethinks automation policy: AI is not just another capital-augmenting technology — by substituting high-skilled cognitive tasks it has qualitatively different distributional consequences, which can justify policies (e.g., subsidies) that would be counterintuitive under standard automation framings.
- Labor-market models need to account for partial exposure: even workers in predominantly non-routine jobs face measurable downside risk from routine-task automation, so task-based exposure metrics should be used more broadly in impact assessments.
- Short- and medium-run adjustment: amplified hours responses and ripple wage effects imply potentially large transitional costs for workers and regions; policies to ease reallocation (training, mobility, social insurance) remain important.
- Redistribution and fiscal design: the optimal fiscal response to AI depends on its distributional impact; if AI reduces top incomes substantially, progressive redistribution may be less costly, and subsidies to AI adoption can be welfare-improving when evaluated against inequality and aggregate-output effects.
- Future research directions: refining measurement of AI-specific task substitutability, dynamic general-equilibrium effects of large-scale AI adoption, and empirical testing of policy prescriptions in real-world adoption episodes.
Assessment
Paper Typeother
Evidence Strengthmedium — The thesis blends theoretical models with multiple empirical analyses across macro and micro data, which strengthens plausibility; however, the summary lacks detail on key identification strategies, robustness checks, and sources of exogenous variation, so causal claims cannot be judged as strongly established from the description alone.
Methods Rigormedium — Use of both macro and micro approaches and task-level exposure measures suggests methodological breadth and sophistication, but the absence of specifics about econometric techniques, identification assumptions, sample construction, and sensitivity analyses prevents a rating of high rigor.
SampleMultiple datasets spanning macro time-series (hours worked, aggregate technology shocks) and micro-level worker/occupation data (wages, hours, task/occupation routineness and automation exposure), plus constructed task-level measures of automation/AI exposure; exact countries, time periods, and data sources are not specified in the summary.
Themeslabor_markets productivity inequality governance adoption
IdentificationCombines structural/theoretical modeling with empirical econometric analyses that exploit variation in task-level automation exposure and time-series technology shocks; the summary does not specify the precise empirical identification (e.g., instruments, natural experiments, diff-in-diff designs) used to make causal claims.
GeneralizabilityUnclear geographic/sampling scope (results may be country- or period-specific)., Task/occupation automation exposure measures may be noisy or rapidly outdated as AI capabilities evolve., Theoretical model assumptions (preferences, production structure, substitution elasticities) may limit applicability to different labor market institutions., Normative tax conclusions depend on model parameterization and distributional assumptions that may not hold universally., Empirical results may not generalize to future AI technologies with qualitatively different capabilities.
Claims (4)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| Automation technologies amplify the impact of technology shocks on hours worked. Employment | positive | hours worked |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Automation reduces the wages of high-skilled occupations, and it also reduces the wages of low-skilled occupations through ripple effects. Wages | negative | wages |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Automation of routine tasks puts downward pressure on workers’ wages even when routine tasks represent only a minor share of their job, indicating a partial but significant exposure of non-routine workers to automation. Wages | negative | wages |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Because AI can perform complex cognitive tasks done by high-skilled workers, subsidizing AI would be optimal: AI replaces the highest-paid workers and thereby contributes to reducing income inequality. Inequality | positive | income inequality (distributional welfare) |
Reading fidelity
high
Study strength
speculative
|
not reported
|