Abstract: Group creativity is widely viewed as an emergent product of individual creativity and the interaction processes among group members. Throughout ongoing interactions, individuals continuously build on one another’s prior ideas, enabling the creative process to emerge dynamically at the group level. Within this process, perspective-taking has been recognized as a core cognitive mechanism. Perspective-taking refers to understanding problems, ideas, and situations from others’ viewpoints. With the rapid development of generative artificial intelligence (AI), AI is increasingly participating in creative tasks as a collaborative partner rather than merely as a tool. However, whether human-AI collaboration differs from human-human collaboration in terms of creative performance and perspective-taking behavior remains unclear. In addition, the actor-partner interdependence model (APIM) posits that an individual’s creative performance can be influenced by both their own perspective-taking behavior (actor effects) and their partner’s perspective-taking behaviors (partner effects). Whether these actor and partner effects differ between human-human and human-AI collaboration has not yet been systematically investigated.
This study employed a between-subjects experimental design with a sample of 169 university students, who were randomly assigned to either the human-human collaboration or human-AI collaboration groups. Participants completed two creative tasks, the Alternative Uses Task (AUT) and the Product Improvement Task (PIT), During the tasks, they interacted with their partners (human or AI) via a custom-built online chat platform and took turns to generate creative ideas within a fixed time limit. Creative performance was evaluated in terms of novelty, usefulness, fluency, and flexibility. Perspective-taking behavior was quantified based on idea-category overlap between partners. Immediate perspective-taking was coded when participant’s idea fell into the same category as the partner’s immediately preceding idea, whereas delayed perspective-taking was coded when it matched a category introduced by the partner earlier in the interaction.
The results showed that: (1) In both AUT and PIT tasks, human-AI collaboration showed significantly higher idea novelty and more frequent perspective-taking than human-human collaboration. (2) The relationship between perspective-taking and creative performance varied across the two collaboration types, tasks, and creativity indicators. In human-human collaboration group, both the actor and partner effects of perspective-taking on idea fluency were significantly positive across both tasks. The partner effect of perspective taking on idea flexibility was significantly positive only in the PIT. In human-AI collaboration group, AI’s immediate perspective-taking exhibited significant positive actor and partner effects on idea fluency in the AUT task, whereas humans’ immediate perspective-taking exhibited significant negative actor and partner effects on idea flexibility. In the PIT task, humans’ delayed perspective-taking exhibited significant positive actor and partner effects on idea fluency and flexibility, whereas AI’s delayed perspective-taking exhibited significant negative actor and partner effects on idea novelty.
These findings suggest that the effects of perspective-taking on creative performance differ between human-human and human-AI collaboration. Specifically, its effects vary across creativity tasks, creativity indicators, and collaborator roles. Our findings extend theoretical accounts of collaborative creativity and provide new insights into the creative dynamics of human-AI collaboration.

Key words: human-AI collaboration, group creativity, perspective-taking