Paper short abstract:

We explore how an algorithmic system, designed to predict critical illness, is taking shape in a negotiation process between software engineers, project managers, and health professionals. We are particularly interested in the ways professional expertise is articulated, negotiated and transformed.

Paper long abstract:

Development of machine learning based algorithms for the health care sector is currently booming, promising results in terms of diagnostic accuracy, predicting illnesses, screening and triaging patients. Yet, many AI applications in clinical settings are failing due to a neglect of clinical contexts and difficulties with interdisciplinary collaborations. Anthropologists working and conducting research in such interdisciplinary futures-focused spaces can help bridge practical and epistemological gaps and contribute with new understandings of the translational roles that involved participants can take to improve collaborations and ultimately build better technology futures.

xAI-EWS is an explainable AI model, based on machine learning and using data from electronic health records, designed to predict acute critical illness and developed as part of a larger research and innovation project at Regional Hospital Horsens in Denmark involving medical doctors, nurses, data scientists, UX designers and anthropologists, among others. In this paper we explore how the algorithmic system is taking shape in a negotiation process between software engineers, project managers, and health professionals. We are particularly interested in the ways professional expertise is articulated, negotiated and transformed in this process.

The paper builds on long term participant observation in Danish hospitals and interviews and ethnographic conversations with a number of the involved partners, both programmers of the algorithm, data scientists, UX designers and health personnel. Attention to work practices and participatory mapping of the workflows, data infrastructures and use of data and other technologies help us outline the collaboration, and potential misunderstandings and conflicts, between different groups of professionals.