Seeing comes before words. The child looks and recognizes before it can speak.

John Berger starts Ways of Seeing (1972) with this still highly relevant quote, establishing that our ability to see precedes our use of language as a system to understand our relationship with our surroundings. While this can be an accurate evaluation of the human relationship with the traditional image, technological advancement and digital media has now led to the development of AI technologies which are highly capable of reproducing realistic imagery and art through a simple sentence or prompt. It seems in the last year, humans have developed a way of seeing through text, destabilising Berger’s initial theory, as well as no longer holding the monopoly on understanding and interpreting what we see. DALL-E 2 is an AI system which relies on reading public data, having scanned millions of pairs of images and text captions found on the Internet. The system then ranks and decides which images are most appropriate for the initial text prompt the user has entered, creating an AI generated image reproduced from millions of visual and text references. The most interesting part of this calculated system is that through its use of the most popular combinations of source images and text captions found on the internet, the imagery it produces from certain prompt words start to reveal how society uses and set trends within digital media: one of the biggest revelations being the censorship and commodification of the female body. Like Instagram and TikTok, DALL-E 2 has strict guardrails in place to stop the production of sexually explicit images, and as always, these guardrails target the female body more than anything else. Due to the influx of female imagery being unfairly associated to explicit and banned content amongst the internet, DALL-E 2 takes this into consideration when scanning the internet for source images for a certain prompt: naturally it will dismiss banned or censored content, and therefore automatically reduce the chances of generating female imagery even if the prompt is ungendered, creating what is known as an algorithmic bias.

So, the question remains, can we wage a war with a prompt-to-image AI system in reproducing the female nude using prompts which are not necessarily filtered out from the AI’s language system, bypassing the AI’s linguistic and visual system to give back women the visual and physical authority of their bodies which they have been craving for centuries?

I explore this subversion through the experimentation of different initial text prompts and the creation of a series of explicit images of the female body, which would typically be censored on social media platforms, using the DALL-E 2 AI system. While demystifying what explicit content means to AI technology, this process is also highly representative of society’s collective understanding of what explicit content is within the digital world: a study of socio-politics surrounding gender stereotypes. There is also a running commentary on the ownership of the explicit image and an effort in overthrowing the violent power imbalance which has been imposed upon the image of the female body for centuries. There is an interesting contradiction between the empowering act of placing the control of explicit content creation in my hands as a woman (which is already subverting the power dynamic of the nude image between the powerless female body and violent male authoritative figure), but also still allowing the AI algorithm to reproduce an image reflecting on stereotypical, and misogynistic principles repeated within our digital society. This conflict is left to the viewer’s interpretation.

The installation itself invites the audience to interact with an abandoned personal laptop, indulging themselves in the viewing of these obsessively stacked explicit AI images within the foreign computer desktop. The interactive audience is forced to play into the male gaze, becoming an active participant in the problematic viewing and distribution of these nude images, putting them into the same voyeuristic and perversive position as the pornographic archive’s mystery owner: a QR code is also present beside the laptop for those perversive enough to save these images to their own archives.