LMMs in Education, Research & Health System Risks

Medical and nursing education

LMMs can generate tailored learning texts, simulate patient conversations (including rare conditions or disabilities), and provide chain-of-thought explanations. Yet incorrect or fabricated content can undermine education quality, and students may defer judgement to the model (automation bias). WHO also notes a new digital literacy burden — professionals must learn to use AI-supported tools while maintaining core clinical skills.

Scientific research and drug development

LMMs can draft manuscripts, summarise literature, analyse datasets, proofread grants, and support de novo drug design. Major publishers and the World Association of Medical Editors restrict authorship to humans because AI cannot assume accountability. Other risks include:

• Hallucinated references — citations to papers that do not exist
• High-income country bias in training data and outputs
• Undermining peer review if opaque AI-generated reviews proliferate
• Paywalled access exacerbating knowledge divides among researchers

Systemic health system risks

WHO groups broader risks that affect entire health systems:

Risk	What it means
Technological solutionism	Overestimating LMM benefits while underinvesting in proven non-AI interventions
Accessibility & affordability	Subscription fees, English-centric models, and digital divides limiting equitable access
System-wide bias	Larger models may encode biases that propagate across institutions
Workforce impacts	Retraining needs, annotation labour concerns, and potential de-skilling
Dependence	Health systems reliant on models not maintained for local contexts
Cybersecurity	Prompt injection, data breaches, and attacks on AI-dependent infrastructure

For Hong Kong healthcare professionals

When using LMMs for CME, teaching, or research writing, disclose AI use per journal and institutional policy. Prioritise human accountability and verify every citation. In hospital governance discussions, ask whether LMM investments divert resources from evidence-based programmes that already improve population health in Hong Kong.

Source: WHO — Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models (2024)