Lunchtime Talk: Unsupervised Pleasures: Conscientious Datasets for Queer Futures

What does generative AI say about people like me? Can it speak so that I recognize myself? „Unsupervised Pleasures“ tries to understand the texts that shape machine learning (ML), by creating exploratory tools and accessible resources to analyze their large-scale datasets and models. It investigates why generative AI systems like GPT-4o have such limited information about anyone who falls outside their norms, and how to realign their values toward digital justice. “Unsupervised Pleasures” creates software interfaces for interrogating existing training datasets, and it suggests methods to develop alternative datasets using intersectional approaches like community input and contributor consent. It uses critical code studies to examine existing datasets and models; and draws on theories, ethics, and tactics from Black feminists, queer and trans activists, anticolonial and anti-ableist scholars to analyse and reimagine ML tools to better support the communities they serve.