Epistemic Status: The Challenge of Integrating Machine Learning Technologies into Professional Knowledge Generation Processes

Abstract

This paper examines the integration of machine learning (ML) technologies into the work of a profession. We depart from the dominant “tool view” in which experts interact dyadically with ML technologies to accomplish a task, instead adopting a “systems view” which reveals the range of technologies, instruments, materials, and adjacent experts that inform a profession’s diagnosis process. Drawing on an ethnographic study of a gastroenterology department that welcomed ML technologies, we observe how these technologies were introduced into a pre-existing epistemic system that produces significant information about a case. We analyze how gastroenterologists (GEs) integrated the outputs of ML technologies into their diagnoses and we identify conditions that challenged this integration. Using a pragmatist semiotic approach, we find that GEs integrated the information produced by ML technologies into their diagnosis process when they deemed it to have epistemic status, that is, when it is deemed novel or useful relative to other information or to the case. More frequently, GEs did not attribute epistemic status to ML technologies because the information produced was excessive to the case, redundant with other information, or incomplete. ML technologies were also excluded from the diagnosis process when they were temporally misaligned with the pace of work set by the GE. This study contributes to knowledge of how ML technologies are changing professional work by showing how their integration is contingent on pre-existing work systems. We problematize the epistemic status of ML technology outputs relative to other information produced in the epistemic system and the case. Further, we show how the application of substantive expertise depends on the generation of relational expertise within an epistemic system which may prevent or slow ML technologies from threatening professional expertise.