Three Degrees of Representation (3°R) Auditing GenAI through Lesbian Film Data

Since I see this project, Three Degrees of Representation, as an investigation of the representation of representations, looking at fan fiction and fan art and their intersection with GenAI served as an informative starting point. There is no shortage of AI-generated lesbian fan art and fan fiction. A simple Google search in the spring of 2025 provided a plethora of existing examples and GenAI tools for users to produce more. That same Google search also revealed ongoing debates about the acceptance of AI-generated writing on fanfiction forums, about creators’ anger at the continued scraping of existing fanfiction content by AI companies (Sung), and about the proliferation of AI-generated content on fan sites and its impact on the quality of the genre (Jones).

Now that these LLM-powered tools are widely available and their products are already seeping into our avenues of culture creation and consumption, the question becomes one of discernment: How can we judge the quality, limits, and affordances of the outputs GenAI offers? As is often the case with GenAI art and writing in these early days of widely available LLMs, almost any output a non-expert user receives in response to a prompt leaves them feeling astonished and satisfied as a consumer of, e.g., poetry (Porter and Machery). However, experts who experience outputs in aggregate are more keenly aware of the limits and homogenizing properties of GenAI-made works. For example, as reported in The Verge, in late 2022, the editors of the science fiction magazine Clarksworld stopped accepting submissions because their submission inbox was flooded with unoriginal, homogeneous sci-fi stories rife with genre tropes, including more than 20 stories with the same title (Sato). No doubt, the submitters assumed these stories fulfilled at least the minimum quality requirements.

Further corroborating this divergent experience of a single work vs. the experience of GenAI-created works in aggregate is a study by Doshi and Hauser investigating the perceptions of short fiction written from AI-generated ideas. As the title of their study reveals, “Generative AI enhances individual creativity but reduces the collective diversity of novel content” (Doshi and Hauser). In other words, to understand the costs and limits of GenAI as a collaborator in storytelling and visualizing processes that redefine representation, it is vital to observe AI-generated content not only on a case-by-case basis but as a collection of outputs.

This project offers precisely that (among other things). By creating and analyzing a collection of images and text outputs related to and “inspired by” Lesbian Films, the project gives users the opportunity to grasp the tendencies of these systems’ outputs in aggregate.