PhD Defence for Grace Turtle: Queering AI x Designing for co-predictive relations

​Abstract

​Predictive AI systems increasingly shape social and technological life by abstracting from situated knowledge and experience, encoding human and nonhuman entities into fixed categories, and constraining indeterminate and queer senses of futurity. This dissertation develops a more-than-human design practice of queering AI, reconfiguring predictive systems toward co-predictive, relational ways of knowing and worlding. Drawing on autotheory as a queer methodology grounded in lived experience, the research advances three experimental engagements—Mutant in the Mirror, Undoing Gracia, and Sounding Territories—that recode predictive systems from within. Mutant in the Mirror uses a style-based Generative Adversarial Network (StyleGAN) to generate self-portraits that explore AI as a nonbinary entity, resisting fixed recognition systems. Undoing Gracia inhabits algorithmic borderlands through a multi-agent simulation grounded in autobiographical data, tracing how plural subjectivities emerge through interaction with a digital twin. Sounding Territories foregrounds embodied, more-than-human relations through a sound-generative model that embraces dis/identificatory codings and resists algorithmic capture. Working from a situated queer mestiza ethics, these experiments traverse AI development from data capture through post-training interaction by composing datasets, configuring models and simulations, and enacting co-performances.

​Synthesizing what these experiments make perceptible in practice, the dissertation distills three queer practice (re)orientations—trans/mutations (toward indeterminacy), algorithmic borderlands (toward thresholds), and dis/identificatory codings (toward illegibility)—and translates them into tactics for design research with predictive systems. The research yields three contributions: 1) introduces co-predictive relations as a theoretical framework countering logics of separability in predictive systems; (2) advances autotheory as an epistemically generative queer design method; and (3) offers redirective pathways for designers and scholars to cultivate desirable un/predictability and queer futurities. By positioning queering as a mode of intervention within AI research, the dissertation contributes to Human Computer Interaction, Society and Technology Studies, design, and futures studies. Through this, the thesis shifts queer discourse from identity-based legibility toward a politics of possibility, concerned not only with who is rendered recognizable by AI, but with how worlds are made and transformed through the design of co-predictive relations.

To everyone who accompanied me through gestures of care, long conversations, and the occasional good dance: gracias. Futurity, I've learned, is cultivated patiently, collectively, relationally. My hope is that this work offers a few signals for the border subjects, threshold peoples, and edge organisations among us, for designing and relating a little differently in a world entangled with AI.