AI-Assisted Formative Assessment in Clinical Education: From Algorithms to Agency

Artificial intelligence (AI) is rapidly reshaping clinical education by embedding assessment and feedback into everyday learning activities. Medical students can now use machine learning dashboards, generative AI, large language models, and emerging agentic systems to practice clinical reasoning, communication, and procedural skills while receiving individualized feedback within seconds. However, the availability of more data and more feedback does not necessarily produce better learning. This Viewpoint is intended for clinical educators, assessment leaders, curriculum committees, faculty developers, and institutional leaders who must decide how AI should be used in formative activities without reducing education to automated scoring. AI-assisted formative assessment is defined in this paper as the intentional use of AI tools to generate, organize, and support interpretation of performance information for learning rather than grading. Its distinctive contribution lies in the scale, adaptivity, conversational simulation, pattern detection, and possible autonomy of AI systems. However, AI outputs become formative only when learners and educators interpret them critically, judge their trustworthiness, and translate them into a small number of focused follow-on learning actions. This paper synthesizes the current evidence base while noting that much of it remains early, heterogeneous, and concentrated in short-term or single-setting studies. It examines key risks, including hallucination, automation bias, epistemic overtrust, hidden curricular effects, and broader concerns related to professional identity, power asymmetries, data privacy, and inequitable access. It also presents context-specific implementation examples for preclinical case-based learning, communication and objective structured clinical examination preparation, procedural skill laboratories, clerkship learning, and programmatic assessment portfolios, together with practical implications for faculty development, institutional governance, and phased local implementation. As a Viewpoint rather than an empirical study or systematic review, the framework and examples should be interpreted as evidence-informed design propositions that require local evaluation and validation. Overall, the value of AI-assisted formative assessment depends less on the volume of AI-generated feedback than on educational designs that preserve learner agency, professional judgment, and human accountability.