Recalibrating AI in clinical decision-making: a process view of human–AI engagement in healthcare

Abstract

This study examines how clinicians continuously recalibrate when, how, and to what extent artificial intelligence (AI) shapes clinical decision-making, moving beyond static adoption and resistance models that treat engagement as a stable outcome. Based on 29 interviews across diverse clinical roles, our analysis indicates that clinicians do not hold stable positions of adoption or resistance. Instead, they enact five recurring decision orientations in practice: withdrawn, concealed, selective, cautious, and routine, through which they adjust the extent, visibility, and authority of AI in clinical judgement. These orientations reflect distinct decision logics grounded in situational assessments of risk, responsibility, and clinical context, and clinicians may shift between them as conditions change. By theorising AI engagement as an ongoing process of decision calibration rather than a one-time acceptance decision, this study offers a process-oriented explanation of human–AI engagement and outlines implications for the design, implementation, and governance of clinically accountable AI.