What is Worth Seeing?

Evaluations and Negotiation over Classification Rules in an AI-Based Medical Imaging Device

Abstract

We examine the implementation of an AI-based imaging system in a clinical microbiology laboratory, focusing on the negotiation between AI developers and laboratory members over classification rules to be embedded into the system prior to its deployment in a clinical setting. The negotiation was triggered by a discrepancy identified during a performance evaluation, when the AI detected microcolonies that laboratory technicians had overlooked. While STS research has explored how medical imaging technologies, once in use, shape healthcare professionals’ diagnostic knowledge, we adopt a technology-in-the-making perspective to analyse how the negotiation process shapes microbiologists’ visual practices even before the system becomes part of routine laboratory work. Drawing on ethnographic fieldwork in the supplier company’s R&D department, we analyse how different evaluations of the visibility status of microcolonies emerge through product customisation and developer–client interactions, and how contrasting viewpoints of what is worth seeing are negotiated into agreement. We argue negotiation over technological capabilities shapes visual routines by unfolding as a process of co‑producing a socio-natural reality. As a contribution to ongoing STS debates, we introduce the term ‘image reversibility’ to capture a mode of problematising conventions underlying AI system development. We also discuss the implications of mobilising the notion of ‘artisanal intelligence’ to encapsulate an implementation style that favours adaptability by embedding specific laboratory practices within the system.