BioSpark: Beyond Analogical Inspiration to LLM-augmented Transfer

Research Background and Problem

• Problems and Challenges:

  • Analogical inspiration is a critical method in design innovation, enabling individuals to gain insights from other domains to solve problems. However, identifying meaningful analogical inspiration, mapping it to the target domain, and creatively adapting it is a complex cognitive task. This process requires moving beyond surface similarity to delve into structural similarity.
  • Many existing tools only support the early stages of inspiration discovery, lacking support for subsequent complex processes such as mapping, transformation, and trade-off adjustments.

• Significance:

  • Analogical innovation has driven numerous breakthroughs in science, technology, and design. Addressing this problem will reduce creative bottlenecks and facilitate the development of new products and designs.

• Research Motivation and Related Work:

  • Previous studies or systems often rely on manually designed datasets or a limited range of predefined inspirations, restricting diversity and flexibility.
  • Some researchers have attempted to expand analogical design through structure mapping theory, natural language processing, and data-driven approaches. However, these efforts primarily focus on the discovery phase of inspiration while neglecting transformation and practicality.
  • This study aims to develop a system leveraging the capabilities of large language models (LLMs) to assist users throughout the entire process of analogical inspiration—from discovery to transformation and adaptation.

Solution

• Approach and Core Contributions:

  • A system named BioSpark is proposed for analogical innovation. This system integrates the powerful capabilities of LLMs to support users in deepening the identification, transformation, and trade-off analysis of analogical inspiration.
  • Core innovations include:
    1. Utilizing the "Tree of Life" to construct diverse strategies for generating biological inspirations.
    2. Providing "Spark Cards" to help users understand the mapping between analogical mechanisms and their problems.
    3. Offering trade-off visualization tools to support in-depth evaluation of the advantages and disadvantages of inspiration mechanisms.
    4. Developing a Q&A module that allows users to explore details through contextualized free-form questioning.

• Key Technologies and Implementation Steps:

  1. Biological Inspiration Generation:
     • Inspiration expansion is achieved using the AskNature website and the biological evolutionary classification tree (Tree of Life).
     • Strategies are developed for both "breadth" (expansion to sibling nodes) and "depth" (expansion to lower hierarchical levels).
     • LLMs (e.g., GPT-4) are employed to generate new inspirations while actively avoiding redundancy.
  2. Inspiration Presentation and Interaction:
     • A Pinterest-like interface is designed to enhance user familiarity.
     • Inspirations are grouped and displayed with rich visuals and text, with GPT-4 extracting the "active ingredients" of each inspiration.
  3. Analogical Transformation Support (Sparks):
     • When users save an inspiration of interest, the system immediately generates two distinct "Spark" suggestions, demonstrating how the inspiration can be adapted to the target design problem.
     • Semantic diversity optimization of previously generated results is used to avoid content repetition.
  4. Trade-offs and Q&A:
     • Automatically generates tables of advantages and disadvantages for inspiration mechanisms in specific problems.
     • Dynamically answers user questions and provides in-depth context through natural language interaction.

Research Outcomes

• Specific Results:

  1. Using BioSpark, participants nearly doubled the number of design ideas generated, significantly improving the creativity quality of designs.
     • Compared to the baseline, users' ideas showed significant improvement in novelty (+3.2 points) and practical value (+3.2 points).
  2. The system helped users expand into more diverse design spaces.
     • The average variety of biological inspiration sources (species) covered by users increased nearly twofold during BioSpark usage.
  3. Users were more focused during tasks.
     • BioSpark's stream interface and inspiration cards effectively reduced users' cognitive load, guiding them to deeply consider the applicability of inspirations.

• Advantages Over Existing Solutions:

  • Compared to the baseline condition of AskNature + ChatGPT, BioSpark not only performed equally well in inspiration discovery but also significantly reduced users' cognitive burden in analogical transformation.
  • The integrated interface and pre-contextualized AI support greatly enhanced task completion efficiency and the quality of design outcomes.

• Experiments and Evaluation Results:

  • User studies revealed that the "Spark" feature in BioSpark was the most frequently used and was rated as the most helpful part of the system.
  • Among the ideas generated by users, more examples successfully transformed analogical inspiration into practical design concepts (e.g., a bicycle rack inspired by the jumping mechanism of frogs).

• Limitations and Future Directions:

  • BioSpark does not yet fully cover subsequent stages of the design process (e.g., prototyping and evaluation).
  • The system's personalization and controllability need further optimization. For instance, users expressed a desire to customize the diversity level of inspiration generation or limit the scope of generation.
  • The current study was conducted in a controlled experimental setting; future work should embed the system into real-world design processes for validation.
  • Enhancing support for technical details in inspiration cards and developing tools for qualitative comparisons from multiple perspectives remain areas for improvement.

Conclusion: BioSpark demonstrates significant potential in supporting the discovery, transformation, and refinement of analogical inspiration in design. Its innovative integration of analogical strategies and AI-assisted methods points to a new direction for creative design tools.