The arc of uncertainty across clinicians’ interactions with AI
Large language models (LLMs) are increasingly being used in various clinical settings to support clinical reasoning and decision-making. Although they have demonstrated strong performance on medical knowledge and reasoning tasks, and early studies suggest potential to reduce medical errors, emerging evidence also highlights important safety concerns, including clinically significant management errors in outputs that may not be readily apparent to users. How these tools can be integrated into clinical practice to consistently lead to improved patient outcomes remains an open question. AI-assisted reasoning unfolds across a sequence of decisions: whether to use AI for the clinical question, how to engage with the tool, how to interpret its output, and whether to act. Each stage in this sequence introduces distinct risks, including inappropriate tool selection, incomplete or biased inputs, overconfidence in fluent responses, and failure to recognize error or appreciate nuance. These risks are further shaped by the same clinician factors that generally affect clinical reasoning such as experience, workload, and uncertainty management. Viewing the AI-clinician interaction as a process rather than a single moment highlights several key points of vulnerability but also opportunities for training. Ultimately, the safe use of AI may depend less on the quality of the model and more on the clinician’s ability to recognize when its outputs should be questioned, verified, or set aside.
Authors
- Jonathan Seth Ilgen (ORCID: https://orcid.org/0000-0003-4590-6570)
- Ryan W. Hunton (ORCID: https://orcid.org/0009-0005-1835-4821)
- Tama Thé
Institutions
- University of Kentucky (US)
Publication Details
- Journal
- Discover Artificial Intelligence
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1007/s44163-026-02851-3
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- article
- Field-Weighted Citation Impact
- 0.00