Human Capital in AI - An Analysis of Earnings and Skills Among Data Science Professionals

As artificial intelligence (AI) reshapes economies and labour markets, the development of human capital for AI-relevant occupations has become a central policy and research priority. This thesis investigates how earnings, gender disparities, and skill formation interact in the AI workforce, focusing on data science professionals. It argues that outcomes cannot be explained by education and experience alone, but by the joint effects of AI-specific capabilities, the pathways through which they are developed, and the frictions that shape their translation into pay.

Three empirical studies are presented. The first extends the Mincer earnings model by incorporating coding depth and machine-learning expertise. Results from OLS and quantile regression show that these domain-specific skills yield higher returns than formal schooling, especially in the upper half of the earnings distribution. The second applies Oaxaca-Blinder and Machado-Mata decompositions to gender pay gaps. Findings reveal that while technical endowments explain part of the disparity, a persistent unexplained gap remains, particularly at the top, consistent with assignment and bargaining frictions. The third evaluates skill-acquisition pathways using causal inference methods. Evidence indicates that blended pathways (university plus non-traditional learning such as MOOCs and project platforms) are most strongly associated with high-earning outcomes, while non-traditional routes alone do not consistently outperform university education.

The thesis makes three major contributions: (1) theoretical, by proposing a Capabilities-Pathways-Frictions framework, which integrates human capital, technological change and institutional perspectives; (2) methodological, by combining decomposition, distributional, and causal techniques; and (3) empirical, by offering novel evidence on AI-specific skills, gender inequality, and learning pathways.

By linking skill formation, gender inequality, and institutional frictions into a unified framework, the thesis demonstrates how human capital theory must be reconfigured for the AI era, and sets out an agenda for future research using firm-level, longitudinal, and experimental data to further test and refine this account.