Chapter 10 - Human-machine interaction: AI-assisted medicine, instead of AI-driven medicine

Abstract

A growing number of Artificial Intelligence (AI) systems are being deployed in clinical settings to improve patient care. AI systems use models learned from large amounts of data to offer insights to clinicians to consider in the process of making clinical care decisions. While the development of AI systems for medicine is advancing rapidly, their integration into clinical workflows is limited, with few examples of widespread and effective use. Unless clinicians perceive AI systems as useful, easy to use, and trustworthy, they are unlikely to adopt them. This chapter examines ways to support the adoption and effective use of AI systems in clinical practice from a human-machine interaction perspective. Explainable AI techniques can help overcome algorithm aversion, develop trust, and deliver timely justifications when clinician expectations are violated by the AI system's output. The success of AI-assisted medicine hinges on our ability to address the sociotechnical challenge of designing AI systems that are embedded in current clinical workflows instead of disrupting them. To address this challenge, we explore the potential of framing the AI system as a member of the clinical team to develop a useful heuristic to guide the design and integration of AI in medicine.