IdeationWeb: Tracking the Evolution of Design Ideas in Human-AI Co-Creation

Research Background and Issues

• What problems or challenges did the authors identify?

  • Current large language models (LLMs) face the following issues in conceptual design interaction:
    1. Multi-round creative exploration and selection are difficult to carry out effectively, leading to random outputs and unclear iterations.
    2. Existing systems lack support for broad exploration during the early stages of ideation, which can result in design fixation and early rigid thinking.
    3. The vast amount of generated content and complex prompt operations cause cognitive overload.
  • Collaboration between designers and AI often suffers from limitations in control and interpretability, affecting the understanding and refinement of generated ideas.

• Why is this problem important?

  • Conceptual design is a critical phase of creative activities, and high-quality conceptual design directly supports innovation and effective outcomes.
  • Human-computer collaboration has significant potential in the design domain, but its complexity necessitates new interaction models to improve user experience and enhance creativity.

• Research Motivation and Related Work

  • The authors integrate existing research on design representation, analogical reasoning, and information visualization to propose a more systematic tool for creative generation and evolution, focusing on structured ideation representation, analogy mechanisms, and visualization support.
  • Many cutting-edge studies focus on optimizing LLMs or prompt design but overlook the development of specific tools aimed at supporting early-stage nonlinear and broad exploration.

Solution

• What methods or solutions did the authors propose?

  • A human-computer collaboration framework called IdeationWeb was proposed, featuring three core design objectives:
    1. Define a consistent hierarchical structure to maintain conceptual coherence in generated ideas.
    2. Implement efficient reasoning mechanisms to support rapid exploration and iteration.
    3. Use interactive visualization to map the design space, enabling parallel divergence and reflection.

• What are the innovative aspects of this solution?

  • Structured Ideation Representation: Utilizing the Function-Behavior-Structure (FBS) model to explicitly divide generated ideas, enhancing comparison and analysis efficiency.
  • Analogical Reasoning Mechanism: Adjusting analogy distances to guide AI in generating diverse and progressive ideas.
  • Interactive Visualization Design: Employing a mind-map-style interface to intuitively present ideation pathways, showcasing the overall scope and development trajectory of ideas.
  • Convenient Interaction Methods: Allowing quick collaborative ideation and systematic exploration through simple operations (e.g., clicking, dragging).

• What are the implementation steps and key technologies used?

  1. Ideation Representation: Generating structured ideas based on the FBS model, including four components: "Object," "Function," "Behavior," and "Structure."
  2. Analogical Reasoning: Developing a semantic feature-based analogy reasoning algorithm to generate ideas with varying degrees of innovation by controlling similarity distances.
  3. System Interface:
     • Visualizing nodes and relationships of each ideation pathway.
     • Providing dual-view modes (global and focused) to support designers in tracking thoughts and systematically reflecting.
  4. The system is powered by GPT-4 and collaborative reasoning technologies for idea generation, combined with Chain-of-Thought reasoning and Few-shot Learning to optimize the controllability of the language model.

Research Outcomes

• What specific outcomes were achieved?

  • System Evaluation: The IdeationWeb system was tested with 40 designers, demonstrating significant advantages in supporting human-computer collaboration and improving idea quality:
    1. The average number of generated ideas increased to 16 (a notable improvement compared to ChatGPT), with broad exploration reaching up to 28 ideation nodes.
    2. The system enabled deep and broad exploration across multiple parallel directions, avoiding early thematic convergence.
  • Improved Idea Quality:
    • Compared to ChatGPT, ideas generated using IdeationWeb scored higher in novelty, value, relevance, and surprise.
    • The study observed a higher proportion of systematic and logical ideas, leading to more mature design outcomes.

• What advantages does it have over existing solutions?

  • It significantly outperforms traditional text-based interaction modes (e.g., ChatGPT) in supporting flexible exploration, generating diverse ideas, and tracking ideation development pathways.
  • The visualized mind-map interface reduces cognitive load, enabling designers to think globally and reflect on ideation evolution.

• What were the experimental or evaluation results?

  • Experiments showed that users rated IdeationWeb significantly higher than ChatGPT in terms of interaction experience (transparency, collaboration, controllability), creative impact, and trust in AI.
  • Expert evaluations revealed that final solutions scored significantly higher in novelty, completeness, and quality, with completeness being particularly beneficial for novice designers.

• Limitations and Future Directions

  • Limitations:
    1. The authenticity of analogy capabilities requires further validation, as current analogy reasoning relies heavily on text-based language models, which need improved interpretability.
    2. The current FBS structure is primarily tailored for product design and does not fully adapt to other design domains.
    3. The tool currently supports single-user operations, with no research on multi-user collaboration scenarios.
  • Future Directions:
    1. Develop a more reliable and interpretable analogy reasoning framework and validate its cross-domain applicability.
    2. Expand to other design domains (e.g., art, architecture) by providing more customized design structures and supporting multimodal content creation.
    3. Explore the tool's potential in team design scenarios, including AI's role in facilitating team collaboration and communication.

Overall, IdeationWeb opens new avenues for the application of LLMs in human-computer collaborative design through structured output, analogical reasoning, and interactive visualization. It addresses current tool limitations and offers significant potential for delivering higher-quality design support.