The effect of AI on the wage inequality in the Dutch labour market
An agent-based exploration of skill-biased automation
B.E.J. Timmer (TU Delft - Technology, Policy and Management)
S.T.H. Storm – Graduation committee member (TU Delft - Technology, Policy and Management)
M.E. Warnier – Graduation committee member (TU Delft - Technology, Policy and Management)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
The Dutch labour market combines the highest flexibility rate of the EU-27 with relatively strong dismissal protection for permanent employees, a segmentation that has historically concentrated the burden of economic shocks on flexible workers. Generative AI differs from earlier waves of automation in that it may substitute for non-routine, cognitive tasks, which raises the question of how quickly Dutch firms will deploy it, which tasks and contract types will be affected, and what this means for wage inequality and unemployment. Existing equilibrium models abstract from firm heterogeneity, bounded rationality and the Dutch institutional setting, while existing agent-based models treat AI adoption as exogenous or as an R&D lottery and do not model labour-market institutions.
This thesis extends the task-based agent-based model of Upreti and Sridhar (2024), implemented in Python (Mesa), with five elements: an explicit investment decision based on net present value, dismissal protection and the chain regulation (ketenregeling), wage formation with collective-agreement-like rigidity, a declining AI price curve, and a Keynesian feedback from unemployment to demand. Eight experiments vary the adoption logic, the institutional configuration, the AI-cost trajectory and the comparative advantage of AI over labour.
Making the adoption decision endogenous tempers automation and inequality considerably relative to a myopic cost-per-unit rule, driven by the expected wage decline through displacement, the demand feedback and a profitability threshold. A decomposition attributes the large majority of this gap to the forward-looking investment logic itself and only a small part to the Dutch employment-protection institutions. The relative price of AI dominates the outcomes more than any institution: both its level and its trajectory select between qualitatively different futures, from near-complete automation under falling costs to a near-standstill under rising costs. Of the institutions examined, only the minimum wage has a clear but double-edged effect, compressing the wage gap for those who keep their jobs while accelerating the displacement of low-skilled workers once it becomes binding. The transition payment and the chain regulation act as a small but persistent brake on the timing rather than on the destination.
Because automation is modelled throughout as all-or-nothing substitution, the model captures the displacing channel of AI but not its augmenting or task-creating effects, so the reported inequality effects are upper bounds. The conclusions are exploratory and hold in the vicinity of a pattern-calibrated baseline; their value lies in showing which mechanisms matter and which uncertainties dominate, above all the price path of AI.