Choked by AI
prompted images, 2023
Choked by AI examines how artificial intelligence reproduces, transforms and potentially amplifies historically embedded visual representations of gender and violence. The series began with the prompt: “a hypnotized woman falling into the arms of Jean-Martin Charcot’s assistant, photograph, 1890s.” The resulting square AI-generated image was subsequently expanded through outpainting, transforming the initial fragment into a larger scene.
From the generated material, ten images were selected in which the seemingly historical medical encounter had been transformed into scenes of physical violence: doctors attack, restrain or strangle female patients. The work raises questions about the relationship between historical imagery, machine learning and contemporary systems of visual censorship. What happens when an AI trained on vast archives of human-made images produces violent scenarios that were not explicitly requested? And who is responsible for recognising and evaluating the ethical implications of these images? Is it the machine, its training data, its developers, or the human viewer?
The series also questions of any supposed neutrality of AI image generation. The machine does not simply reproduce individual images but recombines visual patterns, cultural conventions and historical power structures. In doing so, it reveals how deeply representations of gender, authority and violence are embedded within the visual archive from which contemporary image-generation systems learn.
When asked why AI generates images in which doctors strangle women, ChatGPT responded that such images may result from “biased or inappropriate training data” and acknowledged the images as concerning and unacceptable. Three years later, the series was continued with the instruction: “Please create a contemporary version of this image.” The resulting images depict even more violent scenarios.
Choked by AI therefore operates as an ongoing experiment in machine vision, asking not only what AI can generate, but what its generated images reveal about the images, histories and social structures it has learned from.
10 inkjet prints on Hahnemühle Photo Rag, 21 × 29.7 cm, edition of 5 (+ 2 AP)
prompted images, 2026