In Design Thinking, the ideation phase plays a central role. Yet, after an initial wave of relatively unoriginal ideas, designers tend to fixate on their initial ideas and quickly exhaust their generative capacity. This decline, both in the number of ideas and in their diversity, refers to the fixation phenomenon, which hinders the search for a better final solution.
This thesis starts from the observation that interferences can contribute to defixation, although not all of them have the same reactivation potential. Their effect appears to depend on their nature, their delivery modality, and their degree of project specificity. The thesis mobilizes generative artificial intelligence (GenAI) as ideation infrastructure, i.e., as a way to generate project-specific interferences without replacing the human designer. Its objective is to explore, design, and compare these interferences in order to assess the extent to which they reactivate human ideation after fixation.
The doctoral program combines Design Science Research Methodology (DSRM) and Comparative Case Study (CCS).
The first article explores the generation with ChatGPT 4.0 of oriented textual suggestions as fixation interferences and evaluates them through forced ranking. The second article extends this logic to a public-library innovation problem by comparing a project-specific text interference, a project-specific image interference, and a neutral break. The third article positions GenAI as ideation infrastructure and compares, under a shared post-fixation protocol, text interference, image mosaic interference, video mosaic interference, and a one-minute neutral break.
The results show that it is possible to produce textual suggestions differentiated by their semantic orientation. They further indicate that, in the public-library case, text and image interferences show a positive directional relaunch pattern compared with the neutral break, with a stronger signal for text. Finally, in the multimodal protocol studied, text-based interference shows the strongest directional signal of reactivation and conceptual movement, followed by the image mosaic; the neutral break occupies an intermediate position, whereas the video mosaic appears least effective under this protocol. This result should be understood as a situated pattern showing that richer generative media are not automatically more effective, not as a universal hierarchy of interference modalities.
Taken together, the three articles show that not all interferences are equal. Their effectiveness depends on both their semantic construction and their delivery modality. The thesis contributes to the literature on fixation, defixation, and ideation assisted by generative artificial intelligence. It shows how to design project-specific interferences and relate them to an indicator of conceptual shift with respect to initial idea families (FixShift).
From a practical perspective, the thesis opens a path for practitioners who wish to use GenAI in a simple, low-cost, and non prescriptive way when idea production begins to stall. The results should nevertheless be interpreted with caution, as the program remains exploratory.
The thesis concludes not with a universal recipe, but with the identification of a promising avenue for research and action: using GenAI to produce project-specific interferences that support defixation and reactivate ideation without displacing the designer from their central role.
| Date | 10 Sept 2026 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Mickaël Gardoni (Supervisor) |
|---|
Fischer, F. (Author),
Gardoni (Supervisor),
10 Sept 2026Student thesis: Doctoral thesis › Doctorate in Engineering: Engineering