Artificial intelligence for global health: there is a dangerous flaw in WHO's most discussed proposal. Read the full article: https://www.learning.foundation/2026/08/14/artificial-intelligence-for-global-health-a-tragedy-of-the-commons-brewing-in-genevas-basement/
The problem is arithmetic.
At 95% accuracy per step, a ten-step agentic workflow lands below 60% end to end.
Certifying the output does not fix that.
Reading the path does.
And reading the path requires exactly the grades WHO's restructuring plan cuts first: a projected 30% reduction in P1-to-P3 posts, the roles where people learn to distinguish a technically correct recommendation from one that is wrong for this district, this season, this population.
Reda Sadki, writing in direct response to Rifat Hossain's widely read "The Machine in Geneva's Basement," names what the certification argument misses: the savings from AI adoption are captured now, by the institutions that automate.
The cost arrives as a shortage of independent judgment somewhere between 2035 and 2045, long after anyone can be held responsible.
Nolan Lovett calls it the tragedy of the cognitive commons.
The Validation Tether is its sharpest edge: catching a plausible but wrong answer requires precisely the expertise that AI adoption erodes.
Sadki brings to this debate something most commentators do not have: direct operational evidence.
4,300 health workers in Nigeria produced more than 400 root-cause analyses in four weeks, at roughly 90% lower unit cost and seven times faster than conventional technical assistance, and more than half kept collaborating after the funding closed.
That is a proof of concept for the kind of institution that can reproduce its own capacity for judgment, at the district level, at scale, without Geneva's budget.
For anyone deciding where the next tranche of AI-for-health investment goes, Sadki's test is the one worth applying: not whether a system produces good outputs, but whether the people who will have to overrule it are still being made.
Learn more about The Geneva Learning Foundation's AI4Health certificate programme: https://www.learning.foundation/ai
20
