Critical Dataset and Machine Learning Art

The paper proposes that artistic practices engaging with datasets on which machine learning (ML) systems are trained can provide caring or curative resistance to digital toxicity and furnish models for imagining more equitable digital futures. The paper provides a critique of harmful practices of data mining and data extraction in ML system development by focusing on artworks that engage pharmacologically, that is critically and restoratively, with these technologies. The three works analyzed are ImageNet Roulette (2019) by Kate Crawford and Trevor Paglen, This Person does Exist (2020) by Mathias Schäfer, and Feminist Dataset (2017-ongoing) by Caroline Sinders. These works, we ague, use defamiliarization as a critical practice when engaging with datasets and ML, providing vital counter-narratives and curative strategies, yet in some cases also deepening and exacerbating the very same technological toxicity they set out to remedy.