AI and Non-Western Art Worlds: Reimagining Critical AI Futures through Artistic Inquiry and Situated Dialogue

Research Background and Issues

• What problems or challenges did the authors identify?
  Current generative AI technologies face multiple issues in their application to non-Western cultural expressions, including the dominance of Western aesthetics and values, which may lead to cultural erasure and marginalization. Furthermore, the evaluation of generative AI often overlooks its localization and social dynamics in non-Western societies. Existing research typically views artists as the primary stakeholders, neglecting the fact that art is embedded within broader social and cultural contexts.

• Why is this issue important?
  The potential impact of AI on the fields of art and cultural expression is critical, as it may reshape understandings of cultural identity, history, and political expression. In non-Western contexts, this impact is particularly urgent as it involves the strong influence of structural power dynamics on technology use and cultural representation, potentially raising global concerns about cultural equity.

• Research Motivation and Related Work
  Drawing on sociologist Howard Becker's "art worlds" theory, the authors aim to study how generative AI is adopted, adapted, and reimagined by artists and critics in non-Western cultural contexts from a more socialized and localized perspective. This research also extends existing literature on the impact of AI on creative practices, particularly in non-Western environments.

Solutions

• What methods or solutions did the authors propose?
  The authors conducted an 8-week experiment and dialogue project, involving artists, art historians, and curators in the practice of generative AI. Artists explored the cultural adaptation and creative "hacking" of generative AI tools to produce culturally embedded artworks. This process was supplemented by evaluations and guidance from local art communities on technology and art.

• What are the innovative aspects of this solution?
  The study highlights the following innovations:

  1. Localized adaptation and transformation: Artists are not merely users of technology but also shapers of its possibilities.
  2. Socialized evaluation methods: Expanding the scope of existing AI evaluations through "art world" dialogues that include social and cultural contexts.
  3. Future design perspectives: Using artistic inquiry and collaboration to imagine and design alternative technological pathways for generative AI systems, moving beyond traditional audit-based evaluation frameworks.

• What are the implementation steps and key technologies used?

  1. Experimentation and tool usage: Artists used generative AI tools (including MidJourney, DALL-E, Stable Diffusion, etc.) to create culturally embedded media artifacts, experimenting with fine-tuning specific datasets and prompt engineering.
  2. Workshops and dialogues: Workshops were conducted with local art critics to co-develop artworks and evaluate the generated results, ensuring the alignment of technology applications with cultural contexts.
  3. Data analysis: Qualitative thematic analysis methods were applied to code and summarize artists' journals, outputs, and reflections from dialogues.

Research Outcomes

• What specific outcomes were achieved?

  • Participants successfully developed creative strategies for using generative AI tools, including linguistic testing for cultural contexts, "hacking" techniques, and dataset fine-tuning.
  • The study revealed how generative AI can be reimagined in non-Western contexts as a critical tool, a medium for political expression, and a means of reparative representation.
  • Proposed visions for future AI design, including decolonized archival data, customizable and controllable tools, transparency and user education, and structural reforms for power-sharing.

• What advantages does it have compared to existing solutions?

  • Compared to existing evaluation methods, this solution improves the assessment of generative AI applications in non-Western cultural contexts through localized and socialized perspectives.
  • It proposed a more interactive, multi-stakeholder participation model rather than focusing solely on individual artists.
  • It introduced a vision that integrates cultural and social dynamics, opening new pathways for generative AI design.

• What were the experimental or evaluation results?

  • During the creation process, participants identified limitations in existing technologies, such as the inability to handle multilingual content (e.g., the failure to generate "Woman, Life, Freedom" in Persian), stereotypical outputs, and a lack of local cultural embedding.
  • Artists improved generated results through hacking strategies, such as fine-tuning models or combining multimodal inputs, partially overcoming issues of stereotyping.
  • Dialogue outcomes revealed current technological limitations while providing critical insights for future development.

• Limitations and Future Directions

  • Limitations:
    • Current generative AI tools require significant manual intervention to meet the needs of local cultural and political expression.
    • The centralized control of data by tech companies may hinder communities from effectively influencing model design.
  • Future Directions:
    • Develop more open and user-friendly model architectures and interfaces to support localization.
    • Design decentralized data-sharing and governance models to promote community power-sharing.
    • Combine structural critiques of generative AI with technical design to fundamentally address issues of representation and equity.