What is it about?
The surveys and institutions that tell us what is really happening to people’s health are losing money and respect, just as AI tools churn out confident answers with no source. We argue that sorting real evidence from plausible text must become a funded, defended job for public health bodies.
Featured Image
Photo by Nathan Cima on Unsplash
Why is it important?
Two things happened at once in 2025. The United States stopped paying for the Demographic and Health Surveys, which for forty years measured child deaths, nutrition and vaccination in more than ninety countries, at a cost of less than 0.1% of its aid budget. In the same year, AI chatbots arrived inside the UN, WHO and national health agencies. Ask one of these tools for a child malnutrition figure in any district and it will give you one instantly, whether or not anyone ever measured it. So the machinery that produced checked facts is being switched off just as machinery that produces convincing guesses is being switched on, often by the very organisations whose job is to tell the two apart. At the same time, the scientists and statisticians who do the checking are being cut and publicly attacked. Decisions about vaccines, famine relief and health budgets will be made on whatever numbers are available. Whether those numbers are real matters now, not in some future AI scenario.
Perspectives
The piece reframes what looks like a technology question as a question about power. Whoever controls which information counts as evidence shapes what governments act on, what never reaches the agenda, and what the public comes to accept as common sense. Tobacco and ultraprocessed food companies used that power for decades; AI now makes it cheap, fast and global. Seeing it this way helps readers understand why defunding a survey programme or sacking a chief statistician is not an administrative detail but a shift in who gets to define reality. It also points to a practical fix. Public bodies should label every figure they publish as measured, modelled or machine-generated, invest in collecting real data rather than relying on estimates, fund the people who verify evidence, and speak up when those people are attacked. Institutions that do this will still be trusted in ten years. Those that do not will have handed the question of what is true to whoever owns the model.
Kent Buse
Monash University
Read the Original
This page is a summary of: Who sifts? Knowledge as a public good in the age of AI, PLOS Global Public Health, October 2026, PLOS,
DOI: 10.1371/journal.pgph.0007381.
You can read the full text:
Contributors
The following have contributed to this page
