Always Already Imperfect: Exploring Infrastructures of Expertise and Algorithmic EMS Work

Artificial intelligence (AI) is gaining traction within many areas of expertise. Drawing on fieldwork within an Emergency Medical Response (EMS) unit, we explore the ways in which AI become embedded within an infrastructure of expertise to detect cardiac arrest. We analyze how emergency medical response work is practiced through three distinct yet interwoven forms of expertise: clinical, relational, and technological. Together, these three forms of expertise are embedded in a larger infrastructure that reaches beyond the EMS unit and that must be configured locally, in each individual call, according to what the situation in the call requires. We show that expertise in this setting revolves around a doubt/trust nexus where algorithmic inputs are used and works in several ways, also when it is not responding to a call, depending on how doubt/trust is configured.