Exploring How Users Attribute Responsibilities Across Different Stakeholders in Human-AI Interaction

Abstract

With the increasing integration of artificial intelligence (AI) in various systems and applications, understanding how individuals perceive and assign responsibility in both successful and failed AI interactions is crucial. This study examines the attribution of responsibility among relevant stakeholders - companies, developers, and AI - drawing on the attribution theory. Through an online survey (n = 1,173), we investigated user perceptions of normal and abnormal recommendations on YouTube Kids and their attributions across these stakeholders. Our findings reveal significant differences in perceived ethical responsibility among these stakeholders, with AI consistently bearing higher accountability in both scenarios. This underscores the presence of a complex attribution mechanism in human-AI interactions, calling for a refinement of existing attribution theories to better capture the nuanced dynamics of ethical responsibility in this context.