Abstract AI is rapidly transforming the way humans interact with technology. As we navigate this new landscape, understanding the cognitive processes that guide how users perceive, conceptualize, and rely on AI systems has become an increasingly important field of human–AI interaction research. Specifically, mental models, the internal representations that individuals construct to explain and predict the behavior of complex systems, including their expectations of the system’s capabilities and limitations, are crucial in determining the extent to which users rely on AI systems appropriately—neither excessively nor insufficiently. This research field is still relatively new, and a comprehensive understanding of its theoretical foundations and methodological approaches is still emerging. As a consequence, the potential lack of conceptual and methodological alignment limits comparability and generalizability across studies. To address this gap, this paper presents a scoping review of empirical research on users’ mental models in human–AI interaction, examining the field’s scope, theoretical foundations, and methodologies. The findings suggest a growing diversity in scope and approach, highlighting the field’s interdisciplinary nature. Notable insights include the diversity in theoretical underpinnings, including diverse dimensions and uses of the mental model concept, and the range of methodological approaches to investigate our understanding of AI and its complex effects on user interaction. By comparing and contrasting the various theoretical and methodological approaches, we propose directions for future research, including deeper qualitative and longitudinal investigations of mental models, consideration of contextual factors and application purposes to align with user goals and needs, and efforts to consolidate or expand conceptual frameworks.
Bodamer et al. (Sat,) studied this question.