AI Exposure and the Future of Work: Tasks, Skill Demand, and Demographics

• In Event: AI, Labor Markets, and Economic Impacts

Abstract

As generative AI and large language models reshape how work is performed across the U.S. economy, a pressing question is whether the resulting disruptions will fall evenly across the workforce or deepen existing inequalities along lines of education, occupation, and demographic characteristics. This paper develops a task-based measure of occupational exposure to machine-executed cognitive work and applies it to investigate how AI-driven labor-market shifts may differentially affect workers by sex, age, education, and race/ethnicity. 

In prior work, we constructed a composite exposure index by scoring over 2,000 O*NET Detailed Work Activities for susceptibility to both LLM-enabled tools and mature digital automation, aggregating to occupations using incumbent-reported importance weights. Applying this index to 2019–24 Occupational Employment and Wage Statistics data and 2024–34 BLS Employment Projections yields two central findings. First, exposure is concentrated in knowledge-intensive domains: legal, business and financial, and computer and mathematical occupations rank highest, while physically intensive and site-specific work remains systematically low-exposure. Second, projected employment growth is positive across the full exposure distribution but strongly conditioned by education: within the highest-exposure quartile, occupations typically requiring a bachelor's degree or higher are projected to grow by 7.2 percent, whereas those requiring a high school diploma or less are projected to decline by 3.6 percent. Observed employment change from 2019 to 2024 corroborates this trend, with bachelor's-level occupations in the highest-exposure tier growing 17.1 percent and high school–level occupations declining 8.8 percent. 

These findings raise important questions about which workers stand to benefit from AI-driven change and which may face displacement. We extend the analysis by linking exposure scores to Current Population Survey microdata to examine how workers are distributed across the exposure spectrum by sex, age, education, and race/ethnicity. The paper concludes with policy recommendations for workforce development programs that respond to employer skill needs, with attention to the uneven demographic incidence of AI-driven occupational change. 

Authors: Erik Vasilauskas, Michael Horrigan, Alfonso Flores-Lagunes, The W.E. Upjohn Institute for Employment Research

Author

• Michael Horrigan, W. E. Upjohn Institute for Employment Research
  Presenting Author