Reconfiguring through Ruptures: Material Reconfigurations and Un/Making as Tangible Tactics for Queering AI-Generated Histories

Paper Title

Reconfiguring through Ruptures: Material Reconfigurations and Un/Making as Tangible Tactics for Queering AI-Generated Histories

Publication Info

• Topic area: Queering AI and critical reflection on generative AI's role in historical representation.
• Keywords: Queer AI, generative AI, un/making, material reconfigurations, queer temporalities, archival erasure, bias in AI, autoethnography, arts-based inquiry, queer HCI.

Background and Problem

• Problem / challenge: Generative AI reproduces biases and erasures, especially against LGBTQIA+ identities, in historical representations, leading to misrepresentation and epistemic harm.
• Significance: Addressing these biases is critical to preserving and accurately representing marginalized identities and histories, particularly in the face of increasing algorithmic mediation of memory and archives.
• Motivation and related work: Prior work in Queer HCI and critical AI has explored queering technologies, un/making practices, and artistic critiques. However, gaps remain in developing tangible, embodied strategies to resist the normalizing forces of generative AI and its erasure of queer identities.

Solution

• Proposed approach: The authors propose "material reconfigurations" as a tactic for queering AI, combining un/making practices with embodied, autoethnographic methods to critique and resist biases in AI-generated historical representations.
• Novelty:
  1. Introduction of material reconfigurations as a tactic to critique generative AI's erasures and biases.
  2. Exploration of queer temporalities to destabilize AI-generated histories.
  3. Use of autoethnographic, arts-based inquiry to reflect on embodied unease and critique generative AI.
  4. Examination of tensions between use and refusal of generative AI in queer archival contexts.
• Procedure and key techniques:
  1. Generate AI-based images of personal queer historical memories using ChatGPT-4o, ChatLGBT, and GayGPT with DALL-E 3.
  2. Print and annotate the images by hand to critique their biases and erasures.
  3. Perform material reconfigurations, such as submerging, scratching, and burying the images, to highlight ruptures and resist AI's normalizing tendencies.
  4. Film the process to create video vignettes, combining voiceovers of AI-generated dialogues with embodied, performative actions.

Results

• Concrete findings:
  • AI-generated images exhibited biases, such as Whitewashing, cisnormativity, and erasure of queer diversity.
  • Material reconfigurations (e.g., submerging, scratching, burying) highlighted ruptures in AI-generated representations and reasserted embodied queer perspectives.
  • The project was exhibited in a public gallery, reaching over 285 viewers and sparking dialogues on generative AI's biases.
• Advantage over baselines:
  • Unlike prior methods, material reconfigurations provide a tangible, embodied critique of generative AI, emphasizing the physical and temporal dimensions of resistance.
  • The approach bridges personal, autoethnographic experiences with collective engagements through public exhibitions.
• Experiments / evaluation:
  • Three vignettes were created based on personal queer memories, each involving distinct material reconfigurations (e.g., submerging in water, scratching with tools, burying at a site).
  • The process was documented through video and exhibited alongside a zine containing reflections.
• Limitations and future work:
  • The approach is exploratory and not yet a fully developed method.
  • Future work will explore collective engagements with material reconfigurations and further examine tensions between use and refusal of generative AI.

Summary

This paper introduces "material reconfigurations" as a tangible tactic for queering AI, addressing biases and erasures in AI-generated historical representations. Through autoethnographic and arts-based inquiry, the lead author generated synthetic images of queer memories, annotated them by hand, and performed material reconfigurations (e.g., submerging, scratching, burying) to critique generative AI's normalizing tendencies. These practices highlighted ruptures in AI-generated histories, reasserted embodied queer perspectives, and explored alternative temporalities. The work was exhibited publicly, fostering dialogues on generative AI's ethical issues. This research contributes to Queer HCI and critical AI scholarship, inviting further development of tangible strategies for queering AI.