| Abstract [eng] |
This thesis examines the early labour market effects of generative artificial intelligence (GenAI) at the detailed occupational level in four Nordic countries: Denmark, Norway, Finland and Sweden. It asks whether, following the rapid diffusion of GenAI since late 2022, occupations with higher theoretical exposure to GenAI already show different employment and wage dynamics than less exposed occupations. The theoretical framework is based on the task-based approach, which understands occupations as bundles of tasks, and technology as affecting tasks within occupations rather than occupations as a whole. GenAI is treated as a potential general-purpose technology whose labour market effects may emerge with a delay because they depend on complementary innovations, organisational adaptation and the wider institutional and regulatory environment. The thesis distinguishes between automation, understood as task displacement and a potential reduction in labour demand, and augmentation, understood as task reinstatement, productivity growth and increased demand for complementary human work. Based on this framework, the thesis tests whether more theoretically GenAI-exposed occupations experience slower employment growth (H1.1), slower entry-level employment growth (H1.2), slower or faster real wage growth depending on whether automation or augmentation dominates (H2A/H2B), and structural employment reallocation towards less exposed occupations, both in overall employment (H3.1) and entry-level employment (H3.2). Empirically, the thesis applies a counterfactual Difference-in-Differences design to test whether, after 2023, employment and real wage indicators changed differently in more theoretically GenAI-exposed occupations than in less exposed occupations. The independent variable is the ILO 2025 Global Index of Occupational Exposure, while the dependent variables are employment and real wages data from national statistical institutions in Denmark, Norway, Finland and Sweden. The credibility of the results is assessed using event study specifications, a pre-trend Wald test, placebo tests and robustness checks controlling for occupation-specific linear trends and broad occupational group-year fixed effects. The results show partial evidence of an automation mechanism. The clearest evidence is found in Sweden, where more GenAI-exposed occupations experienced slower overall employment growth after 2023, supporting H1.1. In Finland, a statistically significant negative effect is observed for entry-level employment, providing partial support for H1.2, although this result should be interpreted cautiously because only one post-GenAI year is available. In Denmark and Norway, the estimated effects are weaker, statistically insignificant or methodologically less reliable. No consistent evidence is found that GenAI exposure has already affected real wage growth through either automation or augmentation, so H2A and H2B are not confirmed. Similarly, the hypotheses on structural employment reallocation between more and less exposed occupations (H3.1/H3.2) are not confirmed. Overall, the findings suggest that the early measurable labour market effects of GenAI in the Nordic countries are limited and heterogeneous. They are more visible in employment dynamics than in wages, and more visible in specific countries or entry-level worker groups than in aggregate labour market outcomes. This supports the interpretation that GenAI’s technological potential does not automatically translate into immediate labour market change. Instead, its observed effects depend on the interaction between occupational task exposure, firms’ GenAI adoption strategies, macroeconomic conditions, demographic trends and institutional labour market structures. The thesis contributes to the literature by providing comparative Nordic evidence and points to the need to monitor entry-level transitions, support skills adaptation and collect data on actual firm-level GenAI adoption. |
