CoExploreDS: Framing and Advancing Collaborative Design Space Exploration Between Human and AI

Research Background and Problem

• Problem and Challenges: In the early stages of product design, effective generation of high-quality design ideas is critical for Design Space Exploration (DSE). However, due to the complexity and information-intensive nature of design problems, designers must possess extensive knowledge and balance various solutions under multiple constraints. Although existing Large Language Models (LLMs) have improved in knowledge and reasoning capabilities, they lack process structure and systematic support in collaborative design, resulting in solutions that often lack depth or relevance.
• Significance: Design space exploration is a core step in product design, directly influencing development costs and the innovation of the final product. The co-evolution of problems and solutions during the design process, along with the application of reasoning methods, plays a decisive role in achieving optimal design outcomes.
• Research Motivation: While LLMs have been applied to assist in design and knowledge generation tasks, their potential remains underutilized in complex design reasoning and global exploration. This study aims to establish a systematic framework by introducing design reasoning methods and a problem-solution co-evolution model to improve human-AI collaborative design space exploration.

Solution

• Proposed Method: The authors developed a system called CoExploreDS, which formalizes design problems and solutions as nodes and employs four reasoning methods (deductive, inductive, abductive, and analogical reasoning) to provide dynamic suggestions for the design space. The system's core features include a design task view, a main canvas (node organization area), a design space visualization map, and a "Quick Assist" panel.
• Innovations:
  • Combining the problem-solution co-evolution model with multiple reasoning methods to provide theoretical support for human-AI collaborative design space exploration.
  • Enhancing the system's adaptability to the current design process through dynamic node suggestion generation.
  • Offering real-time visualization of the design space to help designers identify underexplored areas and improve their thought processes.
• Implementation Steps:
  1. Node Formalization: Map design problems and solutions into a node structure on the main canvas.
  2. Dynamic Suggestion Generation: Generate real-time problem or solution suggestions for design tasks based on current content and reasoning methods.
  3. Visualization Map Construction: Dynamically construct a hierarchical design space map to help designers understand the logical relationships between problems and solutions.
  4. User Interaction Design: Provide both "active generation" and "passive generation" suggestion modes, enhancing AI interpretability through transparent logic presentation.

Research Outcomes

• Results and Experimental Validation:
  • Through user studies, the authors validated the effectiveness of the CoExploreDS system in enhancing design quality and fostering creativity. Compared to baseline systems, CoExploreDS significantly improved the novelty and practicality of design outcomes.
  • Expert evaluations showed that designs generated using the CoExploreDS system scored significantly higher in novelty (𝑁) and practicality (𝑈) (𝑁: 5.00 vs. 4.08; 𝑈: 3.28 vs. 2.49).
  • The task load (NASA-TLX) scores for CoExploreDS were significantly lower than those of the baseline system, indicating reduced cognitive burden for users in collaborative design.
• Advantages:
  • Systematic Exploration: CoExploreDS enhances the systematic nature of design exploration through frequent interactions between problems and solutions.
  • Diverse Reasoning: Encourages designers to incorporate multiple reasoning methods in each iteration, avoiding path dependency in design thinking.
  • Boosting Confidence and Balanced Dependence: By providing transparent logic and real-time feedback, the system moderately influences designers' confidence and reliance on AI, fostering a healthy human-AI collaboration relationship.
• Limitations and Future Directions:
  • Limitations: The current system primarily focuses on text-based design expressions, which limits its ability to capture and simulate designers' tacit knowledge and intuition. Additionally, the system's semantic depth is constrained, lacking full integration of multimodal data such as visual and semantic inputs.
  • Future Directions:
    1. Introduce multimodal input and output capabilities to enhance support for non-verbal dimensions of the design process, such as visual sketches or product models.
    2. Explore the integration of knowledge graphs and case libraries to enrich semantic information and deepen design space representation.
    3. Improve system automation and evaluation capabilities, such as incorporating LLMs for automatic scoring and iterative optimization of generated nodes.

Through this study, CoExploreDS provides a novel perspective, enabling designers to leverage AI collaboration in a structured and systematic manner for complex design tasks, significantly enhancing the potential for human-AI co-creative innovation.