Bridging learner and supervisor perspectives of entrustment in health professions education with artificial intelligence

Abstract

This thesis builds upon the concept of entrustment in clinical training by considering both trainee and supervisor contributions to the trusting relationship at its core. When trust flourishes in this relationship, trainees perceive improved learning and patient care outcomes. Bidirectional feedback dialogs built around the negotiation of entrustment-supervision levels appear to be a key ingredient to support trainee-supervisor relationships and foster effective entrustment. These insights are facilitated by traditional qualitative and quantitative methods, supplemented by newly developed artificial intelligence (AI) tools such as natural language processing (NLP). In connecting these findings to broader trends in AI automation—including recent momentum towards “precision education”—this thesis ultimately expands the scope of entrustment beyond human trainees to AI tools and agents, setting the stage for a deeper exploration of how such technologies can be integrated into clinical practice and education while preserving the human-centered trust at its core. Trainees perceive their learning and patient care to be profoundly affected by their supervisors’ trust in them (Chapter 3). Yet, trainees are not always cognizant of their supervisors’ level of trust—instead inferring it from cues deriving from their interactions. When trainees and supervisors explicitly discuss trust, it appears that entrustment based assessment can prompt rich discussion of trainee competencies, which supervisors use to justify their assignment of entrustment levels (Chapter 4). However, trainee and supervisor viewpoints of both the entrustment level and its justification do not always align (Chapter 5), with trainees focusing on a more holistic assessment of their performance and supervisors focusing on how trainees perform specific tasks. In examining the narratives, biases related to gender and underrepresented in medicine (UIM) status can also affect the tone of narrative feedback, but importantly, do not appear to significantly influence the assignment of entrustment-supervision levels. These findings suggest potential benefits when supervisors and trainees explicitly discuss their similarities and differences while determining appropriate entrustment-supervision levels. Trainees and supervisors are largely concordant in their determinations of the appropriate level, with supervisors encouraging greater adaptability and contextual awareness than trainees may initially perceive (Chapter 6). In this way, bidirectional feedback around entrustment fosters a positive feedback loop in trainees’ trustworthiness, reinforcing the premise that effective entrustment is strengthened when decisions move beyond a supervisor’s exclusive determination toward an active, shared collaboration between trainee and supervisor. Finally, the rapid emergence of artificial intelligence (AI) tools within health professions education (HPE) creates a need for clear, ethical frameworks to guide their safe integration. Although guidelines for integrating AI and machine learning (ML) algorithms exist in other domains—including those built around trust—they have not yet been systematically applied within HPE. Addressing this gap, Chapter 7 examines how principles of trustworthiness used to evaluate human trainees can be adapted to AI tools and proposes a structured framework that extends entrustable professional activities (EPAs) to support the safe integration of AI into clinical education.

Ähnliche Arbeiten