Abstract
How are fat and disabled bodies presented through generative AI images and what does this mean for how computer vision conditions notions of normalcy. We argue that the way GenAI models refuse to display certain images offers insight into the socio-political expectations of bodies and identities embedded in the design and dataset of the models. We highlight two modes of refusal: 1) Direct refusal refers to the explicit censorship of words or images by the models, resulting in no image being outputted, and 2) Indirect refusal is when the models prioritise or ignore subsections of the prompt to centre some types of identity and marginalise others. Both types of refusal mediate what is seen as acceptable, ultimately positioning fat and disabled bodies as needing to be assimilated into normative expectations of “the human.” To challenge this version of “the human” proliferated by GenAI images, we bridge crip, an analytic from disability studies, with fat, an analytic from fat studies, to engage with the politics of representation and ideas of a good citizen (Mollow 2015). Both “crip” and “fat” emphasise the potential of non-action. Considering that inclusion in GenAI outputs requires extractivist practices, ‘CripFat’ suggests the need to reconsider which digital systems we want to be included in and why. To do this, we take a position of stillness, refusing to be intimately exploited to be accurately represented. We aim to contribute to conversations about the desire for inclusion in oppressive image systems while recognising the material impacts of aesthetic exclusion.
Presenters
Amy GaetaResearch Associate, Leverhulme Centre for the Future of Intelligence, University of Cambridge, United Kingdom Aisha Sobey
Research Associate, Leverhulme Centre for the Future of Intelligence, University of Cambridge, Cambridgeshire, United Kingdom
Details
Presentation Type
Paper Presentation in a Themed Session
Theme
KEYWORDS
Generative AI, Disability Studies, Fat Studies, Aesthetic Politics, AI Ethics
