Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models

Summary

WHO published guidance on large multi-modal models used in health: systems that take one or more kinds of input (text, images, or video) and generate varied outputs. The document is written for governments, technology companies, and health-care providers. It sits on earlier WHO ethics guidance for AI in health (2021) and offers a long set of recommendations rather than a product specification. WHO is cited here as a source and as an example of health-specific public guidance. That citation does not mean WHO endorses, partners with, or is affiliated with AI Concerns.

The principles it rests on

In our words.

1. Health as the frame: the question is care, not a model card.
2. LMMs as a distinct class: generative systems that take mixed inputs need their own guidance.
3. Several audiences: governments, companies, and providers are each addressed.
4. Ethics and governance together: the title holds both, not one without the other.

Scope

Large multi-modal models developed, provided, or deployed for health. Recommendations are directed at governments, technology companies, and health-care providers.

Why it is exemplary

It is written from a health mandate for a named class of system. It addresses more than one kind of reader. It is public guidance, not a vendor white paper.

Concerns in our register it speaks to

• Diagnostic errorsD13
• Mental-health chatbot safetyD13
• Pediatric AI safetyD13
• Clinical-note summarization errorsD13
• AI documentation "ambient scribes"D13

Frameworks it relates to

• WHO ethics and governance of AI for health (2021), as named by the authors

How a reader might use it

• A school

  Use it to show that a general-purpose model still needs a sector-specific duty of care.

• A health system

  Use it as a starting map for how a health system might read LMMs before buying or building.

• A public body

  Use it as a model of public guidance aimed at government, industry, and providers in one document.

Source

WHO publication page

World Health Organization, Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models, 18 January 2024.