Coordinating actions and sharing control with a disembodied artificial intelligence: Task-dependent differences from human partners.

As humans increasingly collaborate with artificial intelligence (AI) systems, understanding how people coordinate their actions with disembodied AI partners is critical for effective human-AI interaction.While corepresentation, the extent to which a partner's actions influence one's own responses has been well studied in human dyads, its role in human-AI interaction remains unclear.Evidence on vicarious agency, the sense of control over a coagent's actions, is similarly mixed for disembodied AI.In a mixed-design laboratory experiment (N = 115), we manipulated coagent entity (human vs. AI) and goal structure (cooperative vs. competitive).The main dependent variables were corepresentation, measured with the reaction time-based Joint Simon task, and vicarious agency, assessed via temporal delay judgments (intentional binding).Results showed that when the task required only coordinating responses, corepresentation was comparable for human and AI coagents.However, when the task additionally required judging action outcomes (in intentional binding blocks), differences emerged: Human coagents elicited stronger corepresentation and greater vicarious agency than AI coagents.Participants experienced the weakest sense of agency when collaborating with AI, revealing a dissociation between coordination and agency in human-AI interaction.Goal structure did not reliably modulate either measure.