Paper short abstract:

This paper conceptually explores the construction of ‘data bodies’ as technopolitical enactments of constituted medical datasets, machine-learning models and infrastructures and their role in producing categories of ‘others’ on which (biassed) AI-based medical-decisions can possibly be made.

Paper long abstract:

The growing development and implementation of Artificial Intelligence (AI) technologies in “high-stake” domains such as medicine has increasingly raised concerns over the bias and social inequalities these systems risk to reproduce. To better understand and mitigate these issues, much critical work in AI ethics has primarily focused on scrutinising single components such as the medical datasets used to train and test machine-learning models in the field. However, while such elements are vital, AI-inscribed bias may emerge through a series of entangled processes that underpin medical AI development pipelines: the construction and annotation of datasets as well as model production and evaluation. It is therefore critical to further investigate these situated practices and how they are producing categories of ‘others’ on which biassed AI-based medical-decisions can possibly be made.

Following these observations, this paper takes a combined Critical AI and STS research approach to conceptually investigate the construction of ‘data bodies’ as technopolitical enactments of constituted medical datasets, machine-learning models and their underlying infrastructures for AI-based medical decision-making. Drawing on the notion of the ‘body multiple’ (Mol, 2002) data bodies are constituted and always ‘oscillate between multiplicity and singularity’ (Bucher, 2018). Exploring their construction, I argue that medical dataset construction and annotation protocols can be understood as ‘scripts of alterity’ (Pelizza & Van Rossem, 2023) that reveal various ideas and assumptions about medical conditions and patients. As such, they participate in the continuous AI-induced reconfiguration of the lines between self and various others that are potentially targeted for medical treatment.