Session Design Submission Review

No

Yes

Health professions educators are seeking safe, scalable ways for trainees to practice difficult conversations, such as disclosing medical errors, while reducing the costs and scheduling demands of using standardized patients with faculty‑led debriefs. This interactive, 90‑minute workshop demonstrates a simple, no‑coding approach to building and running AI‑based role‑plays in ChatGPT, then guides participants to create a take‑home version for their own programs.

We begin with a brief overview of the training gap and how AI simulation can complement, rather than replace, human simulation. In this session, a general surgeon with no computer‑science background will demonstrate how the simulator was built and showcase the automated debrief it generates, followed by a live demo with a facilitator and a volunteer participant. A short, step‑by‑step walkthrough will then prepare teams for group work.

This is a hands-on session designed for 4–8 tables, each with approximately 6–10 participants.Teams will use a concise prompt guide (~1 page) and a full instruction template (≤8,000 characters) that includes privacy, equity, and scope guardrails. Working collaboratively, participants will:
(1) Draft a de‑identified 150–200‑word surgical error case.
(2) Create a title and ≤300‑character description (refined with ChatGPT).
(3) Insert the case into the instruction template with guardrails (refined with ChatGPT).
(4) Add conversation starters (e.g., introduction, function/role, acknowledge, plan next steps).
(5) Optionally upload brief documents to inform the simulator (if your plan supports file upload; otherwise paste short excerpts).
(6) Run 5–7‑minute test interactions, record observations, make improvements, and re‑test the adjustments.

If timing allows, one or two tables will role‑play their scenarios with volunteers from another group. The session concludes with a brief debriefing, offering practical prompting tips for eliciting constructive feedback from the simulator and outlining next steps for local implementation. Participants will leave with a working prompt/simulation for error‑disclosure role‑play, the concise prompt guide and full instruction template; exemplar conversation starters, and a short debrief checklist.

Target audience: health professions educators (with emphasis on surgery), faculty involved in simulation and assessment, and trainees.

Preparation & setup: bring a de‑identified case (or use our provided examples as a starter). In the session, at least one internet-connected laptop per team is recommended (helpful but not mandatory). A free ChatGPT account is sufficient for the prompt‑based build; teams with paid ChatGPT (e.g., Plus, Pro, Team/Edu/Enterprise) may optionally save their prompt as a Custom GPT after the session. Free users can use but cannot create new Custom GPTs.

This session supports ASE’s priorities by fostering active learning, practical skill‑building, and a scalable, reproducible approach to expanding access to difficult conversation practice across diverse institutional settings.

90-minute workshop

No

No

To learn a simple, no-coding, reproducible approach to building and running AI-based role-plays in ChatGPT (Custom GPT).

To expand complemental access to difficult conversation practice with Generative AI across diverse institutional settings.

To create a take-home AI simulation (Custom GPT) for learners' own programs in training medical error disclosure.