From Paper to Card: Transforming Design Implications with Generative AI

Title of the Paper

From Paper to Card: Transforming Design Implications with Generative AI

Paper Information

• Domain: Design cards; HCI design practices and generative AI
• Keywords: Design cards, generative AI, large language models, text-to-image models, HCI design tools

Research Background and Problem

• Identified Problem or Challenge: Design insights in the HCI field are often presented in the form of academic papers, which designers find difficult to understand or apply. Designers frequently perceive academic papers as overly specialized, hard to read, and lacking practical applicability.
• Importance of the Problem: Efficient dissemination of design insights can bridge the gap between research and practice, enabling designers to better leverage academic research for creative design, thereby enhancing the practical significance and impact of research.
• Research Motivation and Related Work:
  • Design cards are an effective format for transforming research insights into design tools, but they typically require significant time and resources to create manually.
  • Generative AI (e.g., LLMs and text-to-image models) may offer a scalable solution for disseminating design insights.

Solution

• Proposed Method or Solution: The study designed and implemented an end-to-end system that leverages large language models (LLMs) and text-to-image generation models to automatically generate design cards, including components such as titles, descriptions, images, paper abstracts, and design evidence.
• Innovations:
  • This is the first application of generative AI in the field of automated design card generation, enabling designers to understand design insights without delving into complex academic papers.
  • The system is directly tailored to the practical needs of designers, considering card structure, content readability, and visual appeal.
• Implementation Steps and Key Technologies:
  • Preliminary User Research: Conducted interviews to understand designers' preferences for design card content and format.
  • Integration of Generative AI:
    • Large language models (e.g., GPT-3) were used to generate titles, descriptions, and paper abstracts.
    • Text-to-image models (e.g., DALL-E 2) were used to generate images relevant to the design insight themes.
  • Validation and Optimization:
    • Cards were optimized through visual hierarchy adjustments (e.g., color matching and double-sided design).
    • The system achieved a fully automated design card generation process, from parsing research paper content in PDF format to producing the final cards.

Research Outcomes

• Specific Achievements:
  • Developed an effective system capable of automatically generating design cards that meet designers' needs.
  • Experiments demonstrated that design cards make design insights more inspiring and generative while maintaining academic rigor (e.g., validity, originality).
  • The authors concluded that the system is highly effective in disseminating design insights, supporting improved communication of their research findings.
• Comparative Advantages Over Existing Solutions:
  • Compared to traditional academic papers, design cards significantly enhance the readability and appeal of design insights, reducing the cognitive load on designers.
  • The use of generative AI technology provides an efficient and scalable solution, addressing the high time costs associated with manually creating design cards.
• Experimental or Evaluation Results:
  • When evaluating the design insights of the cards, designers generally found them to be more inspiring (significant difference, p < 0.001) and somewhat more generative (p < 0.1).
  • The textual content of the cards generated by the system was highly accurate, and the visual elements helped designers quickly understand the context.
• Limitations and Future Directions:
  • There are still issues with visual representations being inconsistent with or ambiguous in relation to the research intent, requiring further optimization of the image generation and selection mechanisms.
  • The specific impact of the generated cards on actual design processes (e.g., influence on design outputs) has yet to be evaluated.
  • In the long term, addressing potential biases in generative AI training data (e.g., gender and racial biases) and studying the distribution and cognitive effects of the cards on designers are necessary.

Conclusion

This paper proposes an effective method for automatically generating design cards using generative AI, significantly improving the efficiency and effectiveness of disseminating academic research insights. Future work could evaluate the system's performance in more practical design scenarios and expand its functionality to support diverse design needs and user groups.