New WHO discussion paper sets out opportunities and risks of AI in evidence-informed health policy

The World Health Organization has published Artificial intelligence and evidence-informed policy - emerging challenges and opportunities, a discussion paper examining how AI is reshaping the way health policy is made and what is needed to ensure those changes strengthen, rather than weaken, the evidence base on which decisions rest.

"The policy conversation on AI has focused on clinical care. This paper redirects attention to where the evidence base is actually being shaped: how problems are defined, how options are designed, how impact is assessed. Member States need a common framework for governing AI across that entire cycle. This paper provides a starting point." said Dr Alain Labrique, Director of Data, Digital Health, Analytics and AI at WHO.

The new paper, developed jointly by the Department of Data, Digital Health, Analytics and AI and the Department of Science for Health, sets out to fill the gap. It is intended to be applicable to a diverse audience, including policy-makers, regulators, health managers and AI developers.

Mapping AI's role across the policy cycle

The paper organizes its analysis around the stages of the policy cycle - understanding the problem, designing solutions, and achieving impact through implementation, monitoring and adjustment - identifying at each stage the distinct opportunities, risks, governance considerations and practical responses that AI brings, alongside cross-cutting dynamics that apply across all stages.

AI brings significant analytical capabilities to the policy process. The same capabilities, however, introduce risks that vary by phase: data bias can skew problem definition; over-optimization of measurable objectives can narrow solution design; digital divides and cybersecurity vulnerabilities can undermine implementation; and subtle biases in monitoring tools can gradually shift policies away from their original goals. A recurring cross-cutting concern is epistemic injustice: the tendency of AI systems to privilege quantifiable, data-rich evidence while marginalizing lived experience, local expertise, Indigenous knowledge and community-based insight.

Aligning evidence and AI governance

The paper takes a deliberately practical approach: identifying where existing evidence-informed policy-making (EIP) tools and AI governance frameworks already converge: on transparency, participatory engagement, rights protection and risk-based oversight. It draws on WHO's own AI ethics guidance, the GRADE Evidence-to-Decision framework, FAIR data principles, and the OECD AI Principles, among others, to help policy-makers adapt existing instruments for the purpose.

"AI is entering health policy work faster than most institutions have built the capacity to govern it," said Sameer Pujari, who is leading AI in the Department of Data, Digital Health, Analytics and AI. "This paper offers Member States practical guidance for that gap: it maps AI's place across the policy cycle and draws on governance frameworks many countries are already using, so the work of adaptation builds on what is in place rather than starting from scratch."

From principle to practice

The paper translates these governance principles into operational guidance. Before AI tools are deployed, it recommends algorithmic impact assessments and technology readiness reviews. Once in use, it calls for living evidence workflows that pair automated retrieval with human verification, human-in-the-loop decision gateways, and multidisciplinary oversight panels combining domain, methods and ethics expertise.

A unifying principle runs through this guidance: AI should augment, not automate. Humans remain responsible for framing the questions, judging the quality of evidence, interpreting results in context, and weighing ethical considerations.

"Evidence-informed policy-making has always depended on judgement, context and a plurality of voices," said Dr Tanja Kuchenmüller, Unit Head, Research and Ethics Ecosystem Strengthening in the Department of Science for Health. "AI can extend our reach into larger datasets, living evidence syntheses, and faster scenario modelling, but it should strengthen human deliberation, not replace it. This paper aims to help policy-makers harness that potential while preserving the transparency, inclusiveness and trust that underpin evidence-informed policy-making."

The discussion paper is intended as a foundation for continued dialogue with Member States, researchers, developers and civil society as AI capabilities and the contexts in which they are applied continue to evolve.