From scientific events to shared knowledge: An AI-based methodology with clinical supervision
Alberto Molteni, Manuela Girina, Priscilla Cannone, Miriana Alberga, Veronica Bandera, Luciana Pisano, Maria Elena Vismara, Manuela Cuoghi, Grazia Maria Attinà, Luigi Ricciardelli, Andrea Caramma, Marco Catarci, Vincenzo Bottino, Nicola CillaraBackground:
Scientific events generate substantial educational content that remains largely confined to attendees, while formal congress proceedings have progressively disappeared. Modern artificial intelligence (AI) tools offer the technical possibility of converting recorded events into structured written knowledge, but their reliability depends on expert clinical supervision. We describe a clinician-led AI workflow designed to bridge this gap.
Methods:
We applied a four-stage tool-agnostic workflow to the diverticulitis module of the ACOI Bergamo 2026 Congress: (1) integral audio recording and slide collection; (2) parallel transcription with two independent AI systems and reconciliation; (3) AI-assisted structuring into a narrative article with human verification against the original audio; (4) AI-assisted reference retrieval and source verification. A clinician supervised every stage.
Results:
The workflow produced a structured clinical chapter in approximately three working days. The principal AI failure mode was the generation of speaker conclusions not actually stated in the recorded session—a phenomenon we provisionally refer to as “hallucinated synthesis”—corrected through audio relistening. The principal efficiency gains were AI-assisted reference retrieval and voice-based interaction with the model.
Conclusions:
The workflow appears reproducible by individual clinicians without dedicated technical staff. The critical component is the clinician supervisor, whose role shifts from primary writer to verifier and editor. By lowering the cost of converting scientific events into written outputs, this method may expand access to content currently confined to attendees.