Artificial Intelligence and the Labor Market∗

We utilize recent advances in natural language processing to develop novel measures of workers’ task-level exposure to artificial intelligence (AI) and machine learning technologies from 2010 to 2023, capturing variation across firms and over time. We show that tasks exposed to AI subsequently experience lower labor demand. Employing a model that distinguishes between direct and indirect productivity effects of labor-saving technologies, we identify two variables that summarize the impact of AI on within-firm labor demand: an occupations mean task exposure to AI, and the degree to which this mean exposure is concentrated in a small number of tasks. Higher mean exposure reduces labor demand, whereas more concentrated exposures plays an offsetting role as it allows workers to reallocate their effort to non-displaced tasks. Leveraging exogenous variation in AI adoption linked to firms’ pre-existing hiring practices, we find empirical support for these predictions. Overall, we observe relatively modest net employment effects due to countervailing forces: reduced demand in AI-exposed occupations is offset by productivity-driven employment increases across all occupations at AI-adopting firms.

Introduction. Recent advances in artificial intelligence have re-ignited the perennial concern that technology will automate away most tasks performed by workers, leading to large declines in labor demand, depressed wages, and diminished job opportunities for workers. In contrast to prior waves of technological change, which have largely exposed middle- and low-skilled occupations (Autor, Katz, and Kearney, 2006; Autor and Dorn, 2013; Kogan, Papanikolaou, Schmidt, and Seegmiller, 2023), AI exposure appears to be concentrated in white-collar jobs (Webb, 2020; Eloundou, Manning, Mishkin, and Rock, 2023).

Discussion / Conclusion. This paper examines the impact of artificial intelligence adoption on firm dynamics and labor demand, leveraging firm-occupation level variation in AI exposure. We document three primary empirical patterns. First, AI adoption is concentrated in larger, more productive firms, which tend to have distinct growth trajectories. Instrumental variable estimates confirm that AI adoption leads to higher firm-level sales, profits, and total factor productivity. Second, at the occupational level, AI exposure is concentrated in higher-wage positions, with employment effects that depend on the concentration of exposure across tasks. Higher average exposure reduces within-firm employment, while greater concentration in exposure mitigates these declines by reallocating labor toward complementary tasks. Third, firm-wide AI adoption generates positive employment effects, consistent with AI-driven productivity gains increasing aggregate labor demand. The results suggest that while AI substitutes for labor at the task level, its net employment effects are shaped by offsetting forces. Highly AI-exposed occupations experience declines in labor demand, yet within-occupation task reallocation and firm-wide AI-driven growth help sustain overall employment levels.