Paper short abstract:

We present a participatory ethnography of the demonstration of an algorithm imbued with the promise of diagnosing brain tumors more accurately than human experts, challenging the existing hierarchy of expertise amongst physicians.

Paper long abstract:

This study focuses on a deep-learning algorithm to diagnose brain tumors. It was developed by an ambitious radiologist in a Dutch academic hospital, who imbued it with the promise of more accurate diagnosis, potentially establishing a new diagnostic ‘golden standard’ (Timmermans & Berg, 2003), a claim traditionally ascribed to pathologists. In collaboration with this radiologist, we conducted participatory ethnographic research for one year. One of the authors embodied the algorithm by vocalizing its predictive outcomes during neuro-oncology interdisciplinary team meetings. Still at about the same accuracy as the radiologist itself, for them the algorithm’s demonstration served as a means of showcasing its potential, and finding out how it would be received by other specialists.

We found that the algorithm challenged the existing hierarchy of expertise (Carr, 2010), stirring up reflections among specialists on their own claims to expertise, both during the meetings and in interviews. Following STS scholarship on AI and expert systems in medicine (Berg, 1997; Collins, 2021) and demonstrations of automata (e.g, Jones-Imhotep, 2020) we explain how both the discursive and material practices around the diagnosis of tumors were reconsidered under the promise of the ‘new AI expert’. Our embodiment of the algorithm allowed us to study this not just from the perspective of an observant of the interactions, but also as being intrinsically part of this changing expert dynamic.