Episode 1 — Where the clinician's role ends in dental AI
A conversation about clinical-decision-support boundaries and why categorical diagnoses do not belong in current dental AI.
Guest
Dr. Mara Lovre
Clinical Lead, Faurenis
Periodontist and clinical lead at Faurenis. Trained in Zagreb and Stockholm.
Episode 1 — Where the clinician's role ends in dental AI
42 min · with Dr. Mara Lovre
Web player coming soon. For now, listen on your favourite app, or read the full transcript below.
Highlights
- Why "diagnostic AI" is the wrong frame for current dental products.
- How grounding prevents most clinical hallucinations.
- What changed in the day after the clinic adopted ambient documentation.
- The validation queue as an operational dashboard, not a checkbox.
Transcript
HostWelcome to Faurenis Notes. Today we are talking about where the clinician's role ends in dental AI. Mara, thank you for joining.
MaraThanks for having me. It is a question I think about every day at this point.
HostLet us start with the framing. People talk about diagnostic AI in dentistry. What is wrong with that frame?
MaraDiagnostic AI implies that the AI returns the diagnosis. The clinician's role becomes label-reviewer. That is a different cognitive task from making the diagnosis, and it is the wrong gradient for a product where the clinician is in the room with the patient.
HostWhat is the right frame?
MaraClinical decision support. The AI helps the clinician think more thoroughly, more quickly, more completely. It does not return the decision. It surfaces the inputs. The clinician integrates them.
HostLet us talk about grounding. You and the engineering team push hard on grounding as a design principle. What is it doing?
MaraGrounding requires every claim in an AI-drafted note to point to a source — a transcript line, a chart entry, an image. Most clinical AI failures are not subtle reasoning errors. They are claims about things the model never saw. If a model can only say things it has a source for, that class of failure disappears.
HostDoes it disappear entirely?
MaraNo. Spurious grounding still happens — wrong source for a real claim, or the model omits a claim it should have made. But it removes about 80% of what people call "AI hallucination" in clinical contexts.
HostLast question. The validation queue. You called it an operational dashboard. Walk me through that.
MaraThe queue surfaces AI outputs the clinic should look at. If it is empty, either the model is performing well or the queue threshold is too loose. If it is overflowing, either the model is regressing or the threshold is too tight. Either way, the queue gives the admin role a number to track. That is what I mean by operational dashboard.
HostMara, thank you for the time.
MaraThanks for having me.
Hear the next episode.
Faurenis Notes ships on Spotify, Apple Podcasts, and any RSS reader.