Feasibility of intent-elicitation design for LLM-based CSTs

Determine whether the design vision of treating creativity support as intent elicitation—progressively eliciting and refining users’ detailed, idiosyncratic creative intent to guide generation—can be effectively realized in practical LLM-based creativity support tools.

Background

The authors argue that LLM outputs often reflect users’ brief, underspecified intents, which can lead to homogenized results when the system must make many creative decisions on the user’s behalf. They propose reframing creativity support as intent elicitation to draw out more detailed, distinctive user intent before generation, potentially mitigating homogenization and preserving authorship.

They explicitly flag uncertainty about whether such a design vision can be implemented effectively, identifying a need to investigate and validate concrete systems that realize this approach.

References

Whether this design vision can be effectively realized remains to be seen, but we believe it to be an especially worthwhile goal for the next generation of AI-based CSTs to pursue.

— Homogenization Effects of Large Language Models on Human Creative Ideation  (2402.01536 - Anderson et al., 2024) in Section 5.5 (Implications for LLM-Based CST Design)

This introduces a fundamental challenge for adaptive media: how can creators meaningfully control outputs they will never directly observe?

— MultiVerse: A Creator-Centered Approach to Steering Context-Adaptive Lyrics  (2608.19350 - Wang et al., 19 Aug 2026) in Section Discussion, subsection “Creator challenges for authoring adaptive media”

Together, these findings suggest that supporting designers in continuously expressing and adjusting design requirements over the course of generation remains an open problem.

— GUIDE: Designer-in-the-loop Authoring of Conformant Generative User Interfaces  (2609.21285 - Lee et al., 18 Sep 2026) in Section 2.1, Related Work—UI Generation