Generative AI in Creative Practice: ML-Artist Folk Theories of T2I Use, Harm, and Harm-Reduction

Title of the Paper

Generative AI in Creative Practice: ML-Artist Folk Theories of Use, Harm, and Harm Reduction

Paper Information

• Subject Area: Human-Computer Interaction (HCI), Creative Practice, and Generative AI
• Keywords: Art and Technology, Folk Theories, Generative AI, Text-to-Image Models (T2I), Creativity

Research Background and Issues

• Issues and Challenges:

  • Understanding the user experience of generative AI (particularly text-to-image models) in the creative practices of the art community.
  • Generative AI may lead to socio-technical harms, including the cooling of cultural production, homogenization of artistic styles, and economic pressures on creative workers.

• Significance of the Research:

  • Generative AI is rapidly expanding into the arts, and its potential applications and societal impacts require explicit discussion.
  • There is limited research on the harms caused by T2I models, especially in terms of preventive measures.

• Motivation and Related Work:

  • This study adopts a folk theory perspective, engaging with artists (particularly those using generative AI) to explore how the target community perceives and experiences generative AI.
  • Building on prior research into the politics and practices of the ML art community, these reflections may help strengthen responsible AI development.

Proposed Solution

• Proposed Approach:

  • The authors use a folk theory framework to analyze the use, harms, and harm reduction strategies of generative AI in artistic practice.
  • They organized three workshops to gather artists' perspectives on text-to-image models.

• Innovative Contributions:

  • Identified and refined 11 folk theories regarding T2I models, encompassing use, harm, and harm reduction.
  • Explored the relationship between socio-technical issues and potential solutions within the involved community through these theories.

• Implementation Steps and Techniques:

  • Employed qualitative research methods, using Reflexive Thematic Analysis (RTA) to synthesize key themes and folk theories.
  • Designed participatory workshops to capture diverse artist perspectives, utilizing online interactive tools (e.g., Jamboard) to encourage visual expression.

Research Findings

• Specific Findings:

  • On Use:
    1. Text-to-image models serve as artistic media with specific characteristics (e.g., inspiration from errors and failures).
    2. Text-to-image models act as routine tools for communication in prototyping and collaborative processes.
    3. T2I models expand the possibilities and forms of artistic expression.
    4. True creativity requires transcending the basic functions of the model.
  • On Harm:
    1. Engineering efforts to eliminate errors and failures may undermine the artistic attributes of T2I models.
    2. Restricting model functionality and release could harm creative practices.
    3. The harm caused by T2I model use is non-deterministic and context-dependent.
  • On Harm Reduction Strategies:
    1. Increasing transparency to help artists better utilize the models while aiding other users in understanding appropriate use cases.
    2. Expanding artists' control over model parameters to preserve the medium's artistic attributes.
    3. Distributing harm reduction responsibility across multiple stakeholders, including artists, developers, and content moderators.

• Advantages Over Existing Solutions:

  • The study introduces a cross-cultural diversity perspective, analyzing how artists from different social contexts view generative AI.
  • Enriches discussions on the use and societal impact of T2I models, offering more direct recommendations for preventive measures.

• Experimental or Evaluation Results:

  • The authors found that the artist community's attitudes toward T2I models reflect complexity, recognizing their utility while also offering profound socio-cultural insights into the models' limitations and harms.
  • These theories highlight value conflicts in current technological development, such as the emphasis on "perfecting technology" and productivity.

• Limitations and Future Directions:

  • The study focuses on artists who have used T2I models, potentially excluding critics or those who have not engaged with the technology.
  • Recruitment constraints may affect sample diversity.
  • The study does not compare perspectives across different artistic communities.
  • Future research could expand to broader communities and analyze differences and commonalities in folk theories across diverse technology users.

Conclusion and Recommendations

• Folk theories, as an analytical framework, can be used to explore community perspectives on technology use, harm, and solutions.
• This study recommends:
  1. Using folk theories to analyze different communities' attitudes toward technological issues and solutions.
  2. Describing and distinguishing community viewpoints to inform equitable policymaking.
  3. Connecting folk theories with policy design to guide responsible AI practices.