UC Berkeley

The introduction of AI into working processes has resulted in workers increasingly being subject to AI-related harms. By analyzing incidents of worker-related AI harms between 2008 and 2023 in the AI Incident Database, we find that harms get addressed under considerably restricted scenarios. Results from a Qualitative Comparative Analysis (QCA) show that workers with more power resources, either in the form of expertise or labor market power, have a greater likelihood of seeing harms fixed, all else equal. By contrast, workers lacking expertise or labor market power, have lower success rates and must resort to legal or regulatory mechanisms to get fixes through. These findings suggest that the workplace is another arena in which AI has the potential to reproduce existing inequalities among workers and that stronger legal frameworks and regulations can empower more vulnerable worker populations.