Agency Configurations in Generative AI Ideation: How Textual and Visual Idea Concretizations Shape Idea Creativity and Ideator Effort
实验发现，AI生成的文本具体化比视觉具体化提升创意18%，但需多付出30%努力；对不成熟想法，视觉具体化能提升创意而不增加努力。
Ideators increasingly turn to generative artificial intelligence (GenAI) to improve the creativity of their ideas and reduce the cognitive effort required to refine them. This collaboration is based on fine-grained configurations of human-AI agency that allow for partial automation and augmentation of the creative-ideation process. In this study, we examine how representational differences in AI-generated inputs into human ideation in the form of textual and visual concretizations influence the creativity of the jointly produced ideas and the effort ideators must expend. These AI-generated concretizations transform initial raw ideas into more mature representations that ideators can inspect, interpret, and evaluate. In an online experiment, we found that AI-generated textual concretizations improved idea creativity by 18% relative to AI-generated visual concretizations but required 30% greater effort from ideators. This effect was most pronounced for more mature ideas (i.e., specific and actionable). For very immature ideas, visual concretizations enhanced idea creativity without increased effort. We explain these differences through different configurations of human-AI agency in ideation. Textual concretizations correspond to augmentation: GenAI produces coherent textual concretizations that ideators must interpret and complete with their imagination. The material agency of GenAI is matched by human agency, increasing idea creativity but requiring greater effort. As ideas mature, richer textual concretizations provide greater substance for creative elaboration, boosting both idea creativity and ideator effort. In contrast, the use of visual concretizations follows an automation logic. GenAI exerts material agency by autonomously specifying all required details for a visual concretization. While this can boost creativity for very immature ideas by offering stimulating details for creative exploration, these details become constraining as ideas mature. This constrains human agency and ideators’ abilities to integrate novel elements into ideation. As its main contribution, our paper shows that representational differences in AI-generated concretizations shape idea creativity and ideator effort by producing distinct configurations of human-AI agency.