Do Conversational Interfaces Limit Creativity? Exploring Visual Graph Systems for Creative Writing

Abstract: We present a graphical, node-based system through which users can visually chain generative AI models for creative tasks. Research in the area of chaining LLMs has found that while chaining provides transparency, controllability and guardrails to approach certain tasks, chaining with pre-defined LLM steps prevents free exploration. Using cognitive processes from creativity research as a basis, we create a system that addresses the inherent constraints of chat-based AI interactions. Specifically, our system aims to overcome the limiting linear structure that inhibits creative exploration and ideation. Further, our node-based approach enables the creation of reusable, shareable templates that can address different creative tasks. In a small-scale user study, we find that our graph-based system supports ideation and allows some users to better visualise and think through their writing process when compared to a similar conversational interface. We further discuss the weaknesses and limitations of our system, noting the benefits to creativity that user interfaces with higher complexity can provide for users who can effectively use them.