Ethics and governance of artificial intelligence for health

What this guideline says

GMJ summary

This WHO guidance document, developed by a 20-member expert group over two years, establishes the first global consensus principles for the ethical use of artificial intelligence in health. It identifies six core ethical principles: protect autonomy, promote human well-being and safety, ensure transparency and explainability, foster responsibility and accountability, ensure inclusiveness and equity, and promote AI that is responsive and sustainable.

The guidance addresses key ethical challenges including data protection and privacy, algorithmic bias and discrimination, the digital divide, accountability for AI-driven decisions, appropriate use cases, and the impact on the health workforce. It provides frameworks for governance of data, regulation, public and private sector oversight, and liability regimes. The document emphasizes that humans must remain in control of health-care systems and medical decisions, that AI should not encode or exacerbate existing biases, and that appropriate regulatory oversight and transparency are essential.

Intended for ministries of health, regulatory agencies, health-care providers, technology developers, and patients, the guidance calls for collective action to ensure AI technologies improve health outcomes equitably while safeguarding human rights, autonomy, and dignity. It stresses that ethical principles and human rights obligations must be prioritized at every stage of AI design, development, deployment, and evaluation.

Key points

For clinicians — doses and thresholds as stated in the guideline

1. Six core ethical principles must guide AI for health: protect human autonomy, promote well-being and safety, ensure transparency and explainability, foster accountability, ensure inclusiveness and equity, and promote responsive and sustainable AI
2. Humans must remain in full control of health-care systems and medical decisions; AI should augment, not replace, clinical judgement and the provider-patient relationship
3. AI technologies must satisfy regulatory requirements for safety, accuracy, and efficacy before deployment, with continuous monitoring and quality improvement measures
4. Valid informed consent is required for use of AI in experimentation or manipulation; essential services must not be denied if consent is withheld
5. AI systems must be transparent with published information on assumptions, limitations, operating protocols, data properties, and development methods; regular auditing and evaluation by race, ethnicity, gender, age, and other characteristics is mandatory
6. Training data must be accurate, complete, and diverse to avoid sampling bias; oversampling of underrepresented groups may be necessary to achieve equal quality of results across populations
7. AI technologies must not encode or exacerbate biases based on race, ethnicity, gender, age, income, ability, or other protected characteristics; disproportionate effects on specific groups must be monitored with mechanisms for redress
8. Data protection laws must safeguard individual privacy and confidentiality, establish patient control over data, and prevent repurposing for uses beyond original consent; health data collected by governments or companies must meet the same protection standards as clinical data

From the guideline

Verbatim

“In the context of health care, this means that humans should remain in full control of health-care systems and medical decisions.”

Editor’s perspective

Why this matters for Georgia and the region

Applicability

Georgia and the South Caucasus

Questions this guideline answers

Quick answers

Related on GMJ

Interconnected resources

GMJ Guidelines Library records are original summaries written by the GMJ editorial team from the issuer’s published document and reviewed by a physician. They do not replace the full guideline. The document is the copyright of its issuer and is linked, not hosted. Spotted an error or a newer edition? Tell the Guidelines Desk.