BioSpark: Beyond Analogical Inspiration to LLM-augmented Transfer

TL;DR

BioSpark presents an LLM-enhanced creativity partner for analogical design, integrating tree-of-life-based diversification, a Pinterest-like interface, Sparks for transfer, trade-offs, and a contextual Q&A flow to support end-to-end analogical ideation. Through formative studies and a within-subject user study, BioSpark demonstrates increased creative output, higher-quality ideas, and greater diversity of biological inspirations compared with a baseline relying on AskNature and ChatGPT. The work highlights how embedding AI in familiar interaction paradigms can augment designer workflows while foregrounding user agency, managing cognitive load, and addressing fixation through progressive elaboration and contextual grounding. Limitations include concerns about fixation, controllability, and generalizability, with future work focusing on personalization, deeper integration with later design stages, and broader real-world deployment.

Abstract

We present BioSpark, a system for analogical innovation designed to act as a creativity partner in reducing the cognitive effort in finding, mapping, and creatively adapting diverse inspirations. While prior approaches have focused on initial stages of finding inspirations, BioSpark uses LLMs embedded in a familiar, visual, Pinterest-like interface to go beyond inspiration to supporting users in identifying the key solution mechanisms, transferring them to the problem domain, considering tradeoffs, and elaborating on details and characteristics. To accomplish this BioSpark introduces several novel contributions, including a tree-of-life enabled approach for generating relevant and diverse inspirations, as well as AI-powered cards including 'Sparks' for analogical transfer; 'Trade-offs' for considering pros and cons; and 'Q&A' for deeper elaboration. We evaluated BioSpark through workshops with professional designers and a controlled user study, finding that using BioSpark led to a greater number of generated ideas; those ideas being rated higher in creative quality; and more diversity in terms of biological inspirations used than a control condition. Our results suggest new avenues for creativity support tools embedding AI in familiar interaction paradigms for designer workflows.

Figures (19)

• Figure 1: The modal view of a clicked mechanism cluster shows additional mechanism and active ingredient details (Ⓐ). The same action buttons featured on the main page of the interface (Ⓑ) are shown, as well as the 'See more details on Perplexity.ai' for finding additional details and related scientific researech (Ⓒ), and a carousel displaying other species that belong to the cluster which can be viewed by clicking on any of the images (Ⓓ).
• Figure 2: Goal-driven Mechanism Inspiration Generation Pipeline. The pipeline begins from ① a query problem, such as a function 'Manage Impact' from the BioMimicry Institute's Taxonomy which covers a broad range of problems. The problem is used to ② search AskNature.org, which organizes species or innovations according to the functions, to provide seed data. ③ This data is structured into the (problem, mechanism, species) schema and the species names are used to prompt an LLM to construct the tree-of-life hierarchy. ④ The hierarchy is traversed to identify expansion sites at the frontier, here instantiated as sparse branches on the tree with high diversification opportunities, determined in the breadth- or depth-focused manner. The sites along with the problem context and existing mechanisms in the dataset are provided to contextualize the generation and to jointly enforce diversification and relevance filtering. The generation continues iteratively until a stopping condition is met. While the pipeline uses specific data sources in this example, the approach is source-agnostic and adaptable to other contexts (see text).
• Figure 3: Regression results show that the estimated overall quality for the baseline condition is 4.48 and 4.48 + 2.04 = 6.52 for the BioSpark condition. The effect of condition was significant, $t = 16.22$, $p < .0001$, suggesting a substantial difference between the baseline and the BioSpark condition.
• Figure 4: The number of participants' ideas during the experiment.
• Figure 5: (Left) The bar graph shows that the average number of deep engagement was significantly higher in BioSpark than in Baseline.; (Right) The bar graph shows that there were equally many shallow engagement types in both conditions.