The Substitution Effect of Artificial Intelligence on Middle-Skill Jobs: A Core Source of Labor Market Polarization

Abstract

The rapid integration of artificial intelligence into workplace settings has intensified concerns about its asymmetric impact on occupational structures. This study investigates how AI adoption systematically displaces workers in middle-skill occupations—those requiring moderate levels of education and routine cognitive or manual tasks—while simultaneously reinforcing demand for both high-skill abstract roles and low-skill interpersonal or physical roles. Drawing on nationally representative labor force data spanning a decade of accelerating AI deployment, we identify a robust substitution effect concentrated precisely within the occupational core traditionally associated with clerical, administrative, production, and technical support functions. The displacement is not random but follows a distinct task-based logic: occupations dominated by codifiable, sequential, and predictable activities exhibit the strongest negative association with AI intensity measures. This pattern contributes directly to the widening hollowing-out of the labor market middle, independent of broader macroeconomic shifts or educational expansion trends. Our findings clarify the mechanistic role of AI as a structural driver—not merely a correlate—of labor market polarization.