Who is Responsible When AI Fails? Mapping Causes, Entities, and Consequences of AI Privacy and Ethical Incidents

Socio-Technical, Governance, and Human-Computer Interaction

This preprint analyzes real-world incidents related to AI systems through the lens of Thomas Kuhn's paradigm cycle. While it doesn't propose a radical shift in our understanding of AI (a paradigm shift), it does highlight significant discrepancies between existing theoretical frameworks and the practical realities of AI deployment. These discrepancies indicate a model drift, where the existing models of AI governance and ethical considerations are struggling to keep pace with the rapid advancement and widespread adoption of AI technologies. The preprint also identifies areas where current models are inadequate, such as incident reporting practices and the protection of vulnerable populations, suggesting a potential transition towards a model crisis. Therefore, classifying this preprint as Model Drift with a potential shift towards Model Crisis appears most appropriate. The detailed taxonomy presented in the preprint contributes significantly to refining our understanding of AI incidents, which could be instrumental in preventing future issues and improving AI governance, potentially paving the way for a more robust and adaptable model in the future.