The Effects of Generative AI on Design Fixation and Divergent Thinking

Title of the Paper

The Effects of Generative AI on Design Fixation and Divergent Thinking

Paper Information

• Subject Area: Human-Computer Interaction (HCI), Creative Design Tools, Impact of Generative AI on Design Tasks
• Keywords: Design Fixation, Generative AI, Creativity Support Tools, Divergent Thinking, Image Generation

Research Background and Questions

• Research Questions:

  • RQ1: How does AI-generated imagery influence design fixation and divergent thinking compared to traditional heuristic tools and no heuristic support?
  • RQ2: How does the way users interact with AI image generators affect their effectiveness in ideation tasks?

• Significance and Current State:

  • Design fixation is a major barrier to creativity and exploration, where designers are constrained by existing ideas or external examples, leading to insufficient exploration of the design space.
  • Generative AI (e.g., Midjourney, Stable Diffusion) is believed to have the potential to inspire divergent thinking due to its ability to rapidly produce high-quality, original visual content.
  • Empirical research is urgently needed to validate the specific utility of generative AI in design tasks, particularly regarding whether it exacerbates design fixation.

• Research Motivation:

  • There is a lack of empirical data on how generative AI impacts design tasks, especially in terms of its potential to either induce fixation or inspire creativity during the ideation process.
  • The study aims to explore the potential effects of generative AI on the diversity and originality of design solutions while distinguishing between the supportive and limiting aspects of AI-generated content in ideation.

Methodology

• Experimental Approach:

  • Experimental Design:
    • Three conditions were tested: no heuristic support (Baseline condition), heuristic support via Google Image Search (Image Search condition), and heuristic support via an AI image generation tool (Midjourney, GenAI condition).
    • The experimental task involved participants sketching as many different chatbot avatar designs as possible during a visual ideation exercise.
  • Evaluation Metrics:
    • Design Fixation Score (DFS): The proportion of features in participants' sketches that matched those in example designs.
    • Fluency: The total number of sketches.
    • Variety: The extent to which the sketches covered the creative space.
    • Originality: The uniqueness of the sketches.
  • Quantitative: Bayesian statistical models were used to analyze the impact of each condition on design metrics.
  • Qualitative: Interview data were analyzed to understand participants' ideation processes and interactions with the AI tool.

• Innovations:

  • Introduced the new concept of "Fixation Displacement," where fixation may shift from initial design examples to AI-generated content.
  • Emphasized that the inspirational potential of AI design tools depends not only on the generated content but also on how users process this content and how the system guides users to avoid fixation.

Research Findings

• Key Findings:

  1. Design Fixation: Participants exposed to AI-generated images exhibited higher design fixation compared to other conditions. Their designs more frequently incorporated features from the initial examples, and a phenomenon of "Fixation Displacement" was observed.
  2. Fluency, Originality, and Variety:
     • The number of sketches (fluency) in the AI-supported group (GenAI) was significantly lower than in the no-support group.
     • The diversity and originality of sketches in the AI-supported group were also lower than in the no-support group, with no significant advantage over the image search condition.
  3. Dual Impact of Generative AI on Creative Tasks:
     • While high-fidelity images generated by AI could spark some inspiration, they also tended to induce direct imitation, thereby suppressing divergent thinking.

• Recommendations for Improving Heuristic Support:

  • Reduce the fidelity of AI-generated images (e.g., by blurring or providing partially completed images) to mitigate fixation risks caused by detailed and visually intense content.
  • Introduce guided prompting strategies to help users generate prompts that lead to more diverse and less constrained outputs.
  • Encourage users to abstract or reinterpret elements from generated content to develop new design pathways.

• Supporting Experimental Data:

  • 44% of AI-generated images directly reflected features from the initial design examples, and there was a moderate positive correlation (ρ = 0.56) between sketch fixation scores and these images.
  • Based on quantitative models, the AI-supported group exhibited significantly higher design fixation compared to the baseline group, with a Bayes factor of 124, indicating "very strong support" for this hypothesis.

• Limitations and Future Directions:

  • The short experimental duration (20 minutes) limited the evaluation of generative AI's impact in long-term iterative design scenarios.
  • Participants were primarily non-professional designers, so the findings may not fully generalize to experienced designers.
  • The study focused solely on visual design; future research should explore the effects of generative AI on other modalities (e.g., text, audio).

Conclusion

This study highlights the multifaceted impact of generative AI on design tasks and outlines directions for optimizing AI-supported design tools. Developing smarter creativity support tools that guide users to avoid fixation is key to unlocking the potential of AI technology in creative domains.