ABSTRACT VIEW
TEACHING THE UNTEACHABLE: PURPOSEFUL AI HALLUCINATION AS A PEDAGOGICAL FRAMEWORK FOR 21ST CENTURY SKILL DEVELOPMENT
K. Yap1, K. Fo2, J. Wong3
1 La Trobe University, Singapore General Hospital, Duke-NUS Medical School, SingHealth Academy (SINGAPORE)
2 Singapore General Hospital, National Cancer Centre Singapore (SINGAPORE)
3 Singapore General Hospital, National Cancer Centre Singapore, SingHealth Duke-NUS (SINGAPORE)
Introduction:
Healthcare professional (HCP) education faces a challenge in teaching 21st century skills (e.g. empathy, narrative understanding) to students who may never have experienced chronic illnesses. Younger Generation Z and Alpha learners struggle to "put themselves in patients' shoes", yet understanding patients’ lived experiences often matters more for person-centered care than medical knowledge alone. Traditional approaches to teaching empathic responding and narrative competence (NC) lack depth and personalized feedback. In recent years, Generative AI (GAI) has encountered resistance in education due to hallucination concerns, particularly in healthcare fields where accuracy and evidence-based information is crucial. This study reconceptualizes GAI hallucination from a technical “flaw” into a pedagogical advantage through OncoNarratus, a serious game designed to develop 21st century skills in healthcare education.

Methodology:
OncoNarratus is an in-house Gamebot developed using Claude 3.5. It combines the fields of 'Arts' (humanities/social sciences), 'Science' (medical/pharmacy knowledge), and 'Tech' (digital health/edtech) to cultivate skills in health communication, empathic responding and NC, through a novel “Cognitive Authenticity” framework that combines fantasy with reality. We also propose a Purposeful Hallucination Pedagogy (PHP) that employs controlled hallucinations, effective prompt engineering, narrative immersivity, and formative feedback to enhance student learning through GAI games. In OncoNarratus, learners play tailored scenarios of cancer supportive care through personalized avatars, with real-time feedback on their NC. An alpha test with HCPs included pre- and post-surveys assessing changes in empathy and NC, as well as user experience and technology acceptance.

Results:
Among the 17 HCP participants, half (58.8%) had no formal empathic responding training, and 47.1% rated themselves as neutral or not very empathetic. Post-gameplay assessments (n=8) revealed improvements in emotional involvement (median rating scores: 4.81→4.95 on a 7-point scale) and perspective-taking (4.67→5.05). Participants valued the gamebot's meaningful feedback (4.0/5.0) and engagement (4.0/5.0). Technology Acceptance Model showed that participants had strong perceived ease of use (6.0/7.0) and behavioral intention (5.0/7.0), despite 76.5% having limited digital health technology familiarity. Their intention to recommend OncoNarratus to other educators/learners was also high (6.0/7.0), suggesting the possibility to scale across learner groups.

Conclusion:
OncoNarratus demonstrates how controlled AI hallucination can become a powerful educational tool for developing 21st century skills like NC, health communication and empathic responding – all of which are important in person-centered healthcare. While initial testing focused on practitioners, this approach shows promise for the younger digital-native learners. We propose our novel pedagogical framework to educators who want to engage younger learners in immersive, engaging ways that traditional approaches struggle to deliver.

Keywords: Cognitive Authenticity, Empathy Training, 21st Century Skills, Generative AI, Purposeful Hallucination Pedagogy, Narrative Competence, OncoNarratus.

Event: EDULEARN25
Track: Innovative Educational Technologies
Session: Generative AI in Education
Session type: VIRTUAL