Forcing an AI to reconstruct the exact steps of a previous invention makes it significantly more likely to come up with a brand-new, original idea.
Conventional wisdom suggests that creativity requires freedom and open-ended prompts. This study shows that strict structural constraints actually lead to more novel and diverse research outputs from LLMs. By training models to follow a specific ideation path, they develop a better understanding of how to bridge the gap between existing knowledge and new discoveries. The results prove that targeted reconstruction is more effective than open-ended brainstorming for scientific innovation. This means that the best way to use AI for R&D is to give it a rigid framework rather than a blank canvas.
Structure Liberates: How Constrained Sensemaking Produces More Novel Research Output
arXiv · 2605.00557
Scientific discovery is an extended process of ideation--surveying prior work, forming hypotheses, and refining reasoning--yet existing approaches treat this phase as a brief preamble despite its central role in research. We introduce SCISENSE, a sensemaking-grounded framework that operationalizes ideation as a structured sequence of eight cognitive stages (Pirolli \& Card, 2005). We construct SCISENSE-Traj, a 100K-scale dataset of citation-conditioned research trajectories in two modes: Target,