Machine agency attribution in human–AI interaction: developing and validating an analytical framework

Abstract As artificial intelligence (AI) increasingly reshapes human communication, the attribution of agency to machines has become a central scholarly concern. Yet there is little consensus on how users psychologically construe machine agency in human–AI interaction. To address this gap, we propose and validate an analytical framework that organizes prior conceptualizations into three phases: (a) perceptual phase (perception of agentic behavior), (b) inferential phase (inference of agentic minds), and (c) evaluative phase (judgment of agentic influence). Two survey studies develop measures for these phases, and an experiment tests their relationships. The results support a layered process: Perceived machine independence and goal-orientation (perceptual-phase variables) are directly associated with influential capacity judgment, and mental-state inference further strengthens this association. The framework clarifies how divergent conceptualizations of machine agency may reflect different phases of a broader attribution process and helps locate related claims at more precise levels of analysis.