Towards a Diffractive Analysis of Prompt-Based Generative AI

Title of the Paper

Towards A Diffractive Analysis of Prompt-Based Generative AI

Paper Information

• Subject Area: Human-Computer Interaction (HCI), Impact of Generative AI on Creative Production
• Keywords: Generative AI, Creative AI, Diffusion Models, Diffractive Analysis, Creativity Support Tools, Prompt Interfaces, Art and Technology, Human-AI Collaboration, Identity and Authorship, Creative Production Efficiency

Research Background and Problem Statement

• Problems and Challenges: The rapid development of generative AI, particularly systems that generate images based on text prompts, such as Stable Diffusion and MidJourney, has sparked ethical debates (e.g., copyright and plagiarism), cultural impacts, and economic issues. A key concern is how these tools are transforming the creative processes of artists and potentially threatening the uniqueness of human creativity.

  The commercial dominance in AI development often overstates technological capabilities while neglecting the profound impacts on diverse user groups. This creates a critical space for academic research, particularly in analyzing how these tools influence cultural production.

• Significance: The integration of art and technology represents a significant cultural disruption brought about by generative AI. In the broader context of AI's widespread adoption, understanding the interaction between human creativity and machine generation is crucial for shaping the next generation of generative AI.

• Research Motivation and Related Work: This study aims to move beyond a purely technical optimization perspective, focusing on how current technologies influence artistic practices and exploring the cultural and social implications of generative AI. Unlike previous research that primarily examines digital artists, this paper seeks to understand how artists working with physical materials perceive and utilize generative AI tools.

Proposed Solution

• Proposed Solution/Research Methodology: The authors adopt a critical theoretical approach—Diffractive Methodology (DM)—to explore the impact of generative AI on creative practices in a more sensitive and open-ended manner. The study involves seven artists who use a fine-tuned Stable Diffusion generative model, customized based on their own works.

• Innovations:

  • Proposes a novel qualitative analysis method for HCI (i.e., diffractive methodology), emphasizing differences and contextual relationships in the data rather than singular outcomes or statistical generalizations.
  • Examines the tension between generative AI and traditional material-based creative practices, considering the artists' agency and adaptation to technology in their creative processes.

• Implementation Steps and Key Techniques:

  1. Model Customization: Fine-tuning the Stable Diffusion model using each participant's artistic work data, facilitated by the open-source EveryDream tool.
  2. User Testing: Artists use the model over a two-week period, documenting natural observations and experiences without restrictions on usage frequency or strategies.
  3. Semi-Structured Interviews: In-depth interviews are conducted with the artists after the two-week experiment, employing qualitative coding analysis and diffractive methodology to uncover divergences, diversity, and contextual connections.

Research Findings

• Specific Findings:

  • Artists primarily use generative AI tools in two ways: ① Ideation, and ② Production. Each mode requires different levels of human-AI agency.
  • Specific design considerations for these two interaction modes are proposed (e.g., interface optimization, model specialization, domain customization).
  • Offers a neutral analysis of current ethical controversies surrounding generative AI (e.g., plagiarism and data acquisition methods).

• Advantages Compared to Existing Literature:

  • This study places greater emphasis on the deep and contextualized interactions between artists and models rather than focusing solely on technical performance.
  • Highlights how the uniqueness and limitations of models can inspire creativity rather than striving for "perfect replication."

• Experimental and Evaluation Results:

  • Artists exhibited strong user preferences for the model, but their dominant emotional response was ambivalence. Generative AI was perceived as both a "threat" and a "support."
  • The content generated by the model was generally not considered standalone artwork but rather a tool integrated into the artists' creative processes. This underscores the importance of co-creative AI.

• Limitations and Future Directions:

  • Limitations: Small sample size, with participants limited to specific professional fields; no evaluation of the model's impact on general users or art audiences.
  • Future Directions:
    1. Expand to other generative AI models (e.g., MidJourney) to compare the creative support features of different tools.
    2. Investigate the reception of generative AI-created works from the perspective of art audiences.
    3. Explore further integration between AI and traditional creative practices, such as "Can AI fully replace humans?" and the associated cultural and social consequences.

Conclusion

This study employs diffractive analysis to deeply explore the potential interactions and impacts of generative AI on artistic practices. It provides concrete references for the design of generative AI interfaces and technologies while offering new perspectives on the ethical issues surrounding this technology. Generative AI is not merely a tool but an extension of human creative capacity, prompting complex discussions about agency, creativity, and cultural production.