To critically examine the role of AI in historical representation and resist anti-LGBTQIA+ biases and erasures, we leverage un/making and propose a tactic we name material reconfigurations. We share an autoethnographic account of un/making and materially reconfiguring AI-generated images of queer histories: the lead author's memories of queer places and events. Through hand annotating, scratching, burying, submerging, and walking with physical images, they un/make and reconfigure, highlighting embodied aspects of archival records unparsable by generative AI. We propose that un/making and materially reconfiguring synthetic archival images can resist generative AI's increasingly hegemonic role in misrepresenting historical data and erasing queer identities. We contribute reflections on un/making and material reconfigurations as tangible tactics for queering AI, attuning to queer temporalities to unsettle AI-generated histories, using embodied, autoethnographic practices as critical strategies, and working through tensions of use and refusal in Queer AI research.
ACM CHI Conference on Human Factors in Computing Systems