This paper introduces Relational Perception as an observational method for detecting Role Drift in clinical human-AI interaction. Role Drift describes the shift in perceived role of an AI system during use — from assistant to experienced authority — without any technical change to the system. The method uses structured observation of language, behavior, and workflow patterns to identify when this shift occurs and how it influences clinical decision-making. Friction Codes provide a systematic vocabulary for categorizing observed patterns. The paper argues that this relational layer is currently underexamined in both clinical AI research and governance frameworks, including EU AI Act Article 14.
Henri Hommersom (Tue,) studied this question.