Generative AI, Skills, and Older Workers

• In Event: Labor Impacts of AI

Abstract

The emerging literature on the effects of generative AI on labor supply often evaluates the labor force in its entirety, overlooking its potentially distinct effects on older workers. There is small but growing evidence that generative AI has begun to widen income inequality among workers, attaching a wage premium to AI-complemented and newly created jobs (Autor, 2026).In this paper, I examine the impact of generative AI on older workers’ labor market outcomes (i.e., whether they differ substantially from those of younger workers), how occupation-specific tasks and primary skills condition these effects, and how resulting outcome differentials translate into retirement decisions. Inferring from the technology-driven predictors of labor market outcomes from the pre-AI era, I conjecture that variation in labor market outcomes is driven primarily by differences in AI adoption and usage, as well as by the degree of AI–task complementarity within occupations. I further conjecture that negative labor market outcomes associated with slower AI adoption, lower AI usage, and weak AI–task complementarity are exacerbated by perceived skill obsolescence (limited retraining prospects), AI anxiety (psychological factors), and poorer health (health constraints), all of which are more prevalent among older workers.The data are drawn from the Survey of Income and Program Participation (SIPP) 2014–2024 and merged with O*NET occupation identifiers for corresponding years to construct measures of occupation-specific tasks and average skill profiles. An exogenous shock—the launch of GPT in November 2022, widely recognized as a key inflection point in generative AI adoption—is leveraged to identify causal effects. Using a sample of working-age individuals (ages 30–80) who are currently employed, I first implement a difference-in-differences (DiD) design to compare outcomes for younger and older workers around the AI shock, controlling for time-invariant individual heterogeneity as well as occupation- and year-specific macroeconomic conditions. Preliminary results indicate that the AI shock generates worse labor market outcomes for older workers, both in hourly wages and annual earnings, relative to younger workers. Next, using a restricted sample of older workers (ages 50–80), I examine how variation in occupational task-specific AI complementarity conditions these effects, employing a triple-differences (DDD) framework. Preliminary findings suggest that, among older workers, non-routine physical adaptability and non-routine interpersonal adaptability—i.e., dimensions of occupational tasks that enhance AI complementarity—provide protective effects against the AI shock, mitigating declines in earned income.Further analyses will examine how these first-order labor market effects influence voluntary and involuntary retirement timing among older workers. This study provides empirical evidence that generative AI may reduce the viability of extended working lives, particularly for individuals whose skills and occupational tasks are more easily substituted than complemented by AI. The findings are expected to shed light on the underexplored impact of AI on retirement behavior and decision-making.

Keywords:  Generative AI; older workers; labor market outcomes; skills; occupational tasks; retirement; aging workforce

Author

• Zeewan Lee, National University of Singapore – Lee Kuan Yew School of Public Policy
  Presenting Author