ABSTRACT
The generation of an AI image is not a simple transactional process in which text input by humans results in the production of visual output through neutral processes performed by computers using neutral, value-free data. Generative artificial intelligence (GenAI) images emerge from social and material entanglements of data, latent space, diffusion, model weights, prompting, system interfaces, user actions, and affordances for the political and affective possibilities these images carry as they circulate culturally. GenAI images can also emerge through the action of a range of sociomaterial forces where alterations to model performance, efficiency, and safety can reconfigure the aesthetic and reflexive dimensions of computationally generated images as they perform representational and political ‘work’ in their contexts of reception. This chapter examines how pataphysical interventions in GenAI image systems can shape or unsettle dominant imaginaries of technical capability and human interaction with image-making technology. Just as practices such as speculative design have been mobilised in envisioning futures and alternatives in other areas of media technology, GenAI interfaces, model architectures, settings, and defaults offer similar envisioning possibilities. Pataphysical projects that experiment with prompting, dataset curation, or model tuning can be read as examples of critical image-making practice offering ways to revel and reconfigure the infrastructures that sustain narrow or overstated technology industry imaginations of what GenAI image systems do and how they do it. The design of GenAI image systems is not just instrumental but open to competing visions of human-machine creativity and control in the practice of image production.
