The Impact of AI Personas on Human Creativity and Idea Elaboration

Supervisors: Simon Ruffieux, Denis Lalanne

Contact person: Simon Ruffieux

Student: Francesco Lam

Project status: Finished

Year: 2025

This Master's thesis investigates how different AI persona configurations influence human creativity during ideation tasks. Building on dual-process models of creativity, the study compares interactions with a divergent AI persona, a convergent AI persona, and a dual-persona architecture that sequentially combines both approaches. A within-subjects experiment with 16 participants evaluated creative outcomes and interaction dynamics using LLM-based assessments, behavioral measures, and semantic diversity metrics. The results show that the convergent persona encouraged greater user elaboration and was associated with higher automated originality scores, while the dual-persona approach did not increase semantic diversity beyond the divergent persona alone. Moreover, the dual-persona condition resulted in longer interactions and reduced participants' perceived ownership of their ideas. Overall, the findings suggest that separating divergent and convergent AI roles can encourage more structured idea elaboration, but may also introduce interaction friction and negatively affect users' sense of creative agency.

Keywords: generative AI, human-computer interaction, user study.

Document: Not yet available