S1E20 "I Don't Know" — The Three Words AI Can't Say (But Every Clinician Should)

"I Don't Know" — The Three Words AI Can't Say (But Every Clinician Should) In this solo episode, Dr. Adama Diarra makes the case that saying "I don't know" is one of the most underused and undervalued tools in clinical medicine. Using a real patient encounter from the week of recording — a gastroenteritis case in late spring where the standard viral default turned out to be the wrong seasonal answer — he walks through what happened when he paused, admitted uncertainty out loud, did the research, and came back with a better answer. The data showed a meaningful seasonal shift toward bacterial pathogens like Campylobacter and Salmonella in warmer months, which changed how he counseled the patient and shaped their shared decision to pursue supportive care with a clear plan to escalate if needed. From that clinical story, Dr. Diarra draws out three things "I don't know" actually accomplishes: it preserves a clinician's calibration and credibility, it models intellectual honesty for trainees, and most importantly, it protects patients by creating a disclosure loop that a confident wrong answer never would. The episode then pivots to a May 2026 New England Journal of Medicine perspective piece — Can AI Say "I Don't Know"? by Sikora, Celi, and Abdulnour — which argues that AI tools, as currently deployed, lack the reflex to disclose uncertainty. They generate fluent, confident responses regardless of whether the underlying evidence is solid or absent. Dr. Diarra illustrates the danger with a methotrexate dosing example and maps it back to his own gastroenteritis case: an AI would have given the annual default answer with full confidence, missing the seasonal nuance entirely. He closes with a practical framework: become AI-fluent, not AI-dependent. Use multiple sources the way clinicians always have — PubMed, Cochrane, colleagues, subspecialists. Don't outsource critical thinking. And until the tools learn to say "I don't know," that responsibility stays with the clinician.