When AI Faces Clinical Uncertainty

What a periodontal diagnosis study teaches us about LLMs in dentistry

Introduction

Imagine asking two experienced clinicians to assess a patient with incomplete records. No radiographs yet. Some probing depths are missing. The medical history is vague. One clinician might lean conservative. Another might assume higher risk and escalate the diagnosis.

This gray zone is not a corner case in dentistry. It is the daily reality of clinical practice.

Now replace the clinicians with large language models. What happens when AI is asked to reason under uncertainty? And more importantly, does it behave differently depending on who the patient is?

A recent paper published in Frontiers in Digital Health explores exactly this question in the context of periodontal diagnosis. The results are subtle, thoughtful, and highly relevant for anyone building or evaluating AI systems in dentistry.

Screenshot of the research paper by Teerachate Nantakeeratipat. Image adapted by the author.

What is the paper about?

The study titled Large language model bias auditing for periodontal diagnosis using an ambiguity probe methodology investigates how modern large language models behave when diagnosing periodontal disease under ambiguous conditions.