Artificial Intelligence, Skills, and Labor Mobility: Understanding the Transformation of Work

The first three chapters of the PhD dissertation analyze the labor-market consequences of AI in the United States and Europe. Chapter four investigates the predictors of cross-border employment and its labor-market impact on returnees.
Chapter one examines the effects of AI on the U.S. labor market from 2015 to 2022, distinguishing between automation AI—technologies that substitute for labor—and augmentation AI—technologies that enhance workers' output. I construct novel longitudinal measures of occupational exposure to each type of AI by mapping developer activity on Stack Overflow to occupational descriptions. I also introduce a new measure of emerging work based on successive updates of job titles in O*NET. Using an instrumental-variables strategy based on lagged computer science research intensity, I find that augmentation AI stimulates the creation of new work. Automation AI increases employment but depresses wages. These effects vary across skill groups: automation AI harms low-skilled workers, while augmentation AI raises wages and generates new work primarily for high-skilled occupations.
In chapter two, we use 75 million online job postings from four European countries over the 2018–2023 period to investigate how skill and occupational demand have shifted in response to AI. Our analysis leverages data science methods to extract and classify skills from job postings across multiple languages. Instrumenting AI exposure with lagged research intensity in computer science, we document significant growth in demand for AI, Data, and Prediction skills within exposed occupations, alongside a decline in Social skills. Guided by our theoretical framework, the results are consistent with the interpretation that AI, Data, and Prediction are AI-exposed skills, while Judgment, Decision-Making, and Leadership are complementary, and Social skills are substituted. We further document an increased co-occurrence of AI-exposed and complementary skill bundles. Our findings indicate that exposed occupations expand.
Chapter three examines how AI exposure influences the selection of Bachelor's programs in the United States from 2010 to 2022. We distinguish between AI used for task automation and AI that complements human work and analyze three margins: aggregate graduation patterns, student demand for education, and college program supply. Using an instrumental-variable strategy based on lagged computer science research intensity, we find that augmentation AI increases graduations in exposed fields, attracts more and higher-ability students, and stimulates program openings. Automation AI has no significant effect on aggregate graduations but is associated with the likelihood of not pursuing postgraduate studies and field-switching after graduation, raises program closures, and reduces new openings. These results indicate that AI-driven labor-market signals propagate upstream to educational decisions, with automation and augmentation generating distinct responses on both the demand and supply sides of higher education.
In chapter four, using linked Belgian administrative registers that identify cross-border spells in Luxembourg, we provide individual-level evidence on the determinants of entry and exit and the first causal evidence on post-return outcomes, including returnees' interaction with the residence-country welfare system. Random-forest models reveal sharply nonlinear transition patterns in commuting time, prior employment instability, earnings, and household cross-border exposure. Returnees face a short-run employment penalty that fades with cross-border tenure and time since return. They are also more likely to receive unemployment benefits than comparable stayers, with higher daily benefit levels among recipients. We find no evidence of a sizeable effect on wages following return.