Beyond Accuracy: A Hands-On Lab to Evaluate GenAI Safely in Healthcare

Date: Friday, 25 September 2026 | Time: 1445 - 1615 | Venue: PSL2 [Combined], Academia (SGH Campus)

Track Type: Workshop

Speaker(s):

Overview:

GenAI is increasingly used in healthcare for clinical documentation, summarisation, and answering practice-related questions. Yet many teams evaluate GenAI outputs using simplistic criteria such as accuracy or sounds correct. In healthcare, this is insufficient. Unsafe outputs may appear factually plausible while omitting contraindications, fabricating references, disclosing sensitive information, or expressing inappropriate certainty.

This 90-minute interactive workshop is a hands-on evaluation lab that equips participants with a structured framework to assess GenAI safely. Working in small groups, participants will analyse multiple GenAI-generated responses to the same healthcare scenario. Using a guided rubric, they will identify hidden risks, score outputs across multiple dimensions, and debate deployment decisions.

Facilitators will introduce a practical evaluation framework covering clinical safety, hallucination risk, privacy and data disclosure, and robustness. Participants will then translate their findings into actionable decisions such as prompt refinement, guardrail design, human review checkpoints and monitoring strategies. While grounded in healthcare scenarios, the framework is transferable to any high-stakes domain where GenAI supports decision-making or documentation.

Learning Outcome(s):

By the end of this workshop, participants will be able to:
1. Explain why accuracy alone is insufficient for evaluating GenAI in healthcare.
2. Identify common high-risk GenAI failure modes including hallucinations, unsafe omissions, fabricated references, privacy risks and misleading certainty.
3. Apply a structured evaluation rubric to assess GenAI outputs across safety, hallucination, privacy and robustness dimensions.
4. Distinguish between outputs that are acceptable, require human review, or should be blocked from deployment.
5. Design a simple evaluation plan including test scenarios, scoring criteria and acceptance thresholds.
6. Translate evaluation findings into practical implementation decisions such as guardrails, workflow integration and monitoring.

Target Audience:

This workshop is designed for healthcare educators, clinicians, nurses, allied health professionals, informatics teams, quality and patient safety leads, and operational leaders involved in AI or GenAI adoption. No coding experience is required. Participants should have used a GenAI tool – such as ChatGPT, Pair Chat, or similar – at least once before attending.