AI4GH paper urges LMIC AI governance

A paper published on May 21 in npj Digital Medicine argues that artificial intelligence work in global health across low- and middle-income countries is being held back by fragmentation rather than a lack of technical promise. The authors said projects, training, governance and implementation efforts often develop in parallel, leaving countries to navigate adoption without enough shared infrastructure or coordination. They proposed a community-of-practice model built around local leadership, shared governance and cross-country exchange. ### What are the authors saying is broken? The paper by Mylena Maria Guedes de Almeida, Mercedes Rumi and Trudie Lang says AI for global health in LMICs has expanded into a “complex and fragmented” landscape of funders, researchers, implementers and policymakers. According to the article abstract, that fragmentation makes it harder to translate AI work into durable health-system use. (nature.com) IDRC, one of the institutions behind the broader AI4GH initiative referenced in the paper, says demand for health-focused AI solutions in LMICs is not matched by enough locally generated evidence, skilled workers, credible data supplies and computing infrastructure. It also says the risks of poor governance can fall hardest on women and other equity-seeking groups. (nature.com) ### What is the “community of practice” they want instead? The Nature article says the Artificial Intelligence for Global Health Community of Practice, or AI4GH CoP, was established in 2023 to connect researchers, implementers and policymakers across the Global South. The paper presents that network as a practical way to coordinate learning, governance and implementation rather than treating AI deployment as a series of isolated pilots. (idrc-crdi.ca) The AI for Global Health Research hub describes the community as a forum for collaboration and information exchange among participants from Asia, the Middle East and North Africa, Sub-Saharan Africa, Latin America and the Caribbean. It says the goal is to support AI tools that improve health and health care, reduce disparities and strengthen health systems, with work focused especially on sexual, reproductive and maternal health and epidemic or pandemic preparedness and response. (nature.com) ### Why does the paper put so much weight on governance? The paper’s argument is that responsible AI in health is not only about whether a model performs well in a benchmark or pilot. It places equal emphasis on the organizational work around implementation — who sets priorities, who shares knowledge, how decisions are made and whether local institutions can sustain the system after launch. That is an inference from the paper’s focus on communities of practice, shared governance and LMIC-led implementation. (ai-globalhealthresearch.tghn.org) The WHO-led Global Initiative on AI for Health uses similar language in describing the field’s needs. WHO, ITU and WIPO say the initiative aims to build governance structures, policies, technical guidance and standards to support evidence-based adoption of AI for health at country level. ### What does “LMIC-led” mean in practice here? IDRC says its seven-year, CAD22.3 million AI4GH initiative funds researchers in the Global South and is designed to build regional and global networks across researchers, policymakers, practitioners, private-sector actors and civil society. (nature.com) The emphasis is on contextualized, responsible AI rather than importing one model of deployment across countries with different health systems and constraints. (who.int) The AI4GH knowledge hub also describes a steering committee, thematic working groups and regional participation structures, indicating that the model is meant to distribute decision-making rather than centralize it in a single donor or institution. ### Where does this leave the broader AI-for-health debate? The paper arrives as global agencies are building more formal AI governance structures. (idrc-crdi.ca) WHO says its Global Initiative on AI for Health was launched in July 2023 with ITU and WIPO to create long-term institutional support for standards, guidance and country-level implementation. For readers following the field, the paper is less a call for a new model than a push to organize the existing one around local leadership, shared process and sustained collaboration. (ai-globalhealthresearch.tghn.org) The article, “The AI4GH community of practice: strengthening LMIC-Led artificial intelligence for global health,” was published by npj Digital Medicine on May 21, 2026. (nature.com) (who.int)