Design Principles for Generative AI Applications

Document Title

Design Principles for Generative AI Applications

Document Information

• Subject Area: User Experience Design for Generative Artificial Intelligence
• Keywords: Generative AI, Design Principles, Human-Computer Interaction (HCI), User Experience Design, Foundation Models, Trust and Dependence, Co-Creation, Variability

Research Background and Problem

• Problems or Challenges Identified by the Authors:

  • Generative AI technology is becoming increasingly prevalent, yet existing human-computer interaction design guidelines do not provide specialized solutions tailored to the characteristics of generative AI.
  • Generative AI products not only possess the ability to produce diverse outputs but also involve challenges such as calibrating user trust, understanding complex mental models, and mitigating potential user harm.

• Why This Problem is Important:

  • Specialized design principles for generative AI can help users effectively and safely utilize generative AI tools, enhance user experience, reduce potential harm, and improve collaboration efficiency between humans and AI.

• Research Motivation and Related Work:

  • While there are existing guidelines for designing AI systems, they primarily focus on discriminative AI (e.g., classification models) rather than generative AI, whose generative characteristics introduce new design challenges.
  • A literature review reveals that over the past decades, the HCI field has accumulated rich design guidelines for conventional computing systems, web interfaces, mobile applications, and AI interaction design. However, a design framework specific to generative AI has yet to be established.

Solution

• Proposed Method:

  • The authors propose a set of six principles specifically designed for generative AI applications to guide user experience design. These principles reinterpret existing AI system issues and address the unique characteristics of generative AI.

• Innovative Aspects:

  • Provides targeted design guidance for the unique features of generative AI, such as variability in generation and co-creation capabilities with users.
  • Combines design strategies that can be implemented both through the design process and specific user experience functionalities.

• Implementation Steps:

  • The authors developed the design principles through the following iterative steps:
    1. Literature review and initial framework design.
    2. Collection of external and internal feedback for iterative revisions.
    3. Testing the relevance and comprehensiveness of the revised principles in real-world applications using a modified heuristic evaluation method.
    4. Evaluating the practicality of the principles through application in generative AI project teams.

Research Outcomes

• Specific Outcomes:

  • Six design principles were proposed: Design Responsibility, Mental Model Design, Appropriate Trust and Dependence Design, Design for Generative Variability, Co-Creation Design, and Imperfection Design.
  • Each principle is accompanied by specific strategies and examples to help designers apply them in practical design scenarios.

• Advantages Over Existing Solutions:

  • This design framework not only extends existing AI design principles but also validates their applicability and practicality for generative AI through multiple experiments and real-world applications.
  • The research thoroughly considers the unique capabilities of generative AI, such as diverse output generation, rethinking user mental models, and human-AI collaboration potential.

• Experimental or Evaluation Results:

  • Through heuristic evaluations of nine commercial generative AI applications (e.g., ChatGPT, Adobe Firefly), the six principles were found to address core user experience issues and exhibit broad applicability.
  • Successfully helped design teams identify blind spots and generate numerous actionable design suggestions in two types of real-world generative AI product designs.

• Limitations and Future Directions:

  • The study does not sufficiently address experimental generative AI applications and non-commercial innovative forms, such as narrative generation or dynamic action generation interactions.
  • The design guidelines need further expansion to cover other stages of the generative AI lifecycle, such as model selection, optimization, and deployment monitoring.
  • As generative AI technology evolves, new challenges and domain-specific needs may emerge, requiring continuous updates to the principles to adapt to new forms and uses of generative AI applications.

Conclusion

• This study proposes a comprehensive set of design principles and strategies for generative AI, reinterpreting existing issues in AI interaction and offering innovative design solutions tailored to the unique characteristics of generative AI. The iterative development process of the principles combines theoretical frameworks and practical validation, providing designers with a practical toolkit to meet diverse user goals (optimization and exploration). These principles establish a solid foundation for the future design and application of generative AI technologies.