Towards Human-AI Interaction in Medical Emergency Call Handling

Abstract

Call-takers in emergency medical dispatch centers typically rely on decision-support systems that help to structure emergency call dialogues and propose appropriate responses. Current research investigates whether such systems should follow a hybrid intelligent approach, which requires their extension with interfaces and mechanisms to enable an interaction between call-takers and artificial intelligence (AI). Yet unclear is how these interfaces and mechanisms should be designed to foster call handling performances while making efficient use of call-taker's often strained mental capacities. This paper moves towards closing this gap by 1) deriving required artifacts for human-AI interaction and 2) proposing an iterative procedure for their design and evaluation. For 1), we apply the guidelines for human-AI interaction and conduct workshops with domain experts. For 2), we argue that performing a full evaluation of the artifacts is too extensive at earlier iterations of the design process, and therefore propose to enact use-case-driven lightweight evaluations instead.