Oksana Dorofeeva
From things to relations: co-construction of machine learning work with labels, ethics, and biomedical data
In recent years, technologies known as machine learning and artificial intelligence (AI) have attracted a great deal of attention, ranging from excitement to growing societal concerns, as well as emerging governance. This dissertation explores what it means to work with machine learning in this societal context based on interviews with researchers, data scientists, and engineers in Denmark, the Netherlands, and the United Kingdom. It focuses on three aspects of machine learning practitioners’ meanings and experiences: how they view ‘AI’ and ‘machine learning’ labels as representations of their work, how they understand the notion of ethics and its place in machine learning, and the frictions they experience when working with biomedical data. In doing so, the study argues that none of these phenomena, from categories (machine learning, AI, ethics) to material phenomena (biomedical data), are concrete, independent ‘things’, but rather acquire their meanings in relations that together constitute machine learning work. The dissertation elucidates how machine learning work is entangled in bundles of relations, from local interactions to being embedded in broader cultural and institutional contexts, and argues that considering these relations is important for understanding the sociotechnical systems machine learning is involved in, as well as for future dialogue and promotion of responsible research and technology development.