Source-linked AI summary
Beyond Bias: Participatory and Reflective Approaches to Cultural AI
Archana Prasad, Isha Singh, Tom Simmons
TL;DR
Generative AI systems often fail to capture cultural meanings and interpretive practices through computational metrics alone, especially when dominant norms flatten culturally specific representation. Beyond Bias addresses this gap through participatory workshops, reflective tools, collaborative datasets, fine-tuning, and governance practices, finding that participants could develop more situated outputs while critically examining AI’s cultural assumptions.
Problem
Generative AI can privilege dominant norms and flatten culturally specific aesthetics, symbolism, and contextual knowledge in its representations.
Method
Beyond Bias used participatory workshops, collaborative dataset creation, reflective tooling, artist-led fine-tuning, iterative co-design, and co-authored governance practices.
Results
Participants produced more situated cultural interpretations and used iterative experimentation to reflect on representation, authorship, and cultural interpretation.
Takeaways & Limitations
Cultural AI should foreground cultural integrity, interpretive agency, participation, stewardship, and governance rather than treating culture as extractable data.
Takeaways & Limitations
Collaborative fine-tuning remained constrained by commercial foundation-model ecosystems trained on opaque datasets.
Abstract
from arXiv · showhide
Generative AI systems increasingly shape cultural production, yet creative intentions, cultural meanings, and interpretive practices often can't be articulated through computational metrics alone. This paper presents Beyond Bias, a collaboration between Gooey.AI and Goethe-Institut India, as a participatory approach to cultural AI which includes collaborative dataset creation, reflective AI tooling, artist-led model fine-tuning, and co-authored governance practices. Across 9 workshops involving over 200 participants, artists and cultural practitioners engaged with AI systems through experimentation, iteration, and collaborative LoRA training. Participants used their AI-generated outputs and visualizations as reflective interfaces for exploring symbolism, memory, authorship, and cultural contexts. Comparing contemporary generative AI outputs with participant fine-tuned outputs helped participants reflect on cultural details missing in big tech AI systems. This paper contributes reflective AI tooling approaches foregrounding transparency, stewardship, and community participation; findings from participatory workshops examining how generative AI visualizations mediate cultural representation and interpretive practice; and a framework for cultural AI grounded in cultural integrity, and reflective practice.
2 Cultural and Participatory Approaches
Generative AI can reproduce dataset-embedded inequalities and dominant cultural norms, while participatory approaches place artists and cultural practitioners in roles spanning creation, interpretation, and governance.
- Generative AI systems can reproduce inequalities embedded in datasets and computational infrastructures.
- Big tech image models may privilege dominant Western norms and generate stereotyped representations of under-represented communities.
- Cultural AI raises concerns about local knowledge, aesthetics, and histories entering datasets without meaningful consent, attribution, or governance.
- Beyond Bias extends participatory AI by positioning artists and cultural practitioners as co-creators in dataset creation, fine-tuning, interpretation, and governance.
3 The Beyond Bias Initiative
Beyond Bias combined collaborative workshops, reflective tools, fine-tuning, and governance experiments to help cultural practitioners examine and reshape generative AI representations. Participants produced more situated outputs and used iterative co-design to inform tooling, datasets, and governance.
- The Beyond Bias Initiative: Over 200 participants and 24 organizations joined nine workshops involving dataset creation, prompt experimentation, LoRA training, and reflective discussion.
- The Beyond Bias Initiative: The collaboratively authored manifesto established principles including cultural integrity, transparency, environmental accountability, fair compensation, and community stewardship.
- Reflective Tool Design: Reflective tools enabled participants without extensive technical expertise to fine-tune image and video models with collaboratively curated datasets and multilingual prompting workflows.
- Reflective Tool Design: Tools displayed water and electricity costs for each run, while some workshops allowed participants to choose whether datasets could be shared publicly or used for future model training.
- Reflective Tool Design: Localized fine-tuning and participatory tooling created pathways for communities to influence broader generative ecosystems rather than remain end-users.
- Participatory Workshops and Cultural Outputs: Participants found that culturally specific prompts often produced ornate but flattened outputs missing compositional structure, symbolism, material texture, and regional variation.
- Participatory Workshops and Cultural Outputs: Participants created outputs better reflecting the textures, animal forms, and visual style of Lascaux cave paintings than baseline systems generated.
- Participatory Workshops and Cultural Outputs: Workshops became spaces for examining dataset bias and representation, including reinterpretations of Mughal miniatures depicting contemporary political contexts.
4 A Framework for Cultural AI
Beyond Bias proposes cultural AI as an interconnected framework centered on cultural integrity, participation, interpretive capacity, and stewardship and governance.
- The framework comprises four interconnected dimensions: cultural integrity, participation and agency, interpretive capacity, and stewardship and governance.
- Cultural Integrity: Cultural integrity requires preserving contextual meaning, symbolism, provenance, and historical relationships rather than reducing culture to surface-level aesthetics.
- Participation and Agency: Participation and agency position cultural practitioners as co-creators who shape AI through dataset curation, fine-tuning, prompting, and governance.
- Interpretive Capacity: Interpretive capacity treats generated outputs as prompts for reflection and reinterpretation, enabling critical examination of how prompts, datasets, and interfaces shape representation.
- Stewardship and Governance: Stewardship and governance foreground ownership, consent, environmental impact, and cultural extraction as concerns within generative AI systems.
5 Conclusion
The conclusion presents participatory cultural AI as expanding representation and creative agency while retaining tensions around ownership, labor, consent, scalability, and dependence on opaque commercial models.
- Participatory cultural AI creates possibilities for representation and creative agency but introduces tensions concerning ownership, labor, consent, and scalability.
- Collaborative fine-tuning enabled more culturally situated outputs, but participants remained constrained by commercial foundation models trained on opaque datasets.
- Beyond Bias calls for moving beyond harm mitigation toward cultural integrity, interpretive agency, and community stewardship.
- Reflective cultural AI frameworks may enable communities to become active co-creators rather than treating culture as extractable data.