Growing in algorithmic ruins. Contamination as queer-feminist method

Abstract

In data science and artificial intelligence, “data contamination” is typically treated as a technical flaw to be removed. This paper instead approaches contamination as a way to examine how data infrastructures organise and exclude difference. Drawing on feminist science studies and queer theory, it explores how data cleaning and classification embed normative assumptions about gender and sexuality. Focusing on systems, such as DeepL and the United Nations Parallel Corpus, the paper analyses mistranslations and erasures of queer language. These are not isolated errors but reveal how algorithmic systems impose fixed categories onto ambiguous meanings. Engaging with artistic practices that foreground error and glitch, the paper argues that such “contamination” exposes the limits of computational systems. These moments act as “queer ghosts,” traces that resist capture. Contamination thus becomes a queer-feminist method for engaging AI through disruption and the persistence of what does not fit.