Publication

Exploration vs. fixation: Scaffolding divergent and convergent thinking for human-AI co-creation with generative models

Chao Wen, Tung Phung, Pronita Mehrotra, Sumit Gulwani, Roger Beaty, Tomohiro Nagashima, Adish Singla

HCOMP2026 · 2026

@inproceedings{wen2026exploration,
  title     = {Exploration vs. fixation: Scaffolding divergent and convergent thinking for human-AI co-creation with generative models},
  author    = {Chao Wen and Tung Phung and Pronita Mehrotra and Sumit Gulwani and Roger Beaty and Tomohiro Nagashima and Adish Singla},
  booktitle = {HCOMP2026},
  year      = {2026},
}

Abstract

Generative AI is increasingly used for open-ended creative tasks such as story writing, product design, and video creation. However, popular chatbot-based interfaces often prioritize execution, generating fully rendered outputs right away. This can lead to premature convergence and design fixation, where users are anchored to initial outputs. Recent work has proposed new interfaces to address this issue by supporting exploration, though typically constrained to be semantically close to a user’s initial task framing, potentially limiting the creativity of the outcomes. We examine an approach for human-AI co-creation and instantiate it in HAICo, a system for image co-creation. HAICo structures the creative process into two switchable modes: Divergent mode supports users in exploration by surfacing remote conceptual ideas that users might not reach on their own, while Convergent mode supports users in aligning model outputs with their refinement intent through generating intent interpretations surfaced as options. Through a within-subjects study on a poster image creation task, we demonstrate that HAICo outperforms ChatGPT across multiple dimensions of creativity and usability. Our results highlight the potential of leveraging human-AI complementarity to expand the exploration space and treating alignment as an interactive process through which user intent is interpreted and progressively specified.