An article in today's ECONOMIST TODAY newsletter points to an unintended consequence of improved access to reputable health and medical information on line -- due to poor AI translation of source material from English.
The article points out that “The people with the most to gain are the least able to use these tools.”
The article:, "Top AI models underperform in languages other than English" provides a real-world example. Here it is:
https://www.economist.com/science-and-technology/2026/03/18/top-ai-model...
It refers to work by Tuka Alhanai at New York University Abu Dhabi and her collaborator Mohammad Ghassemi at Michigan State University. They have developed a way to benchmark and compare: language models (LLMs) ability to understand other languages, which was reported at
- the April 2025
- AAI Conference on Artificial Intelligence
<https://www.researchgate.net/journal/Proceedings-of-the-AAAI-Conference-...
39(27):27802-27812:
- DOI: - 10.1609/aaai.v39i27.34996 <https://doi.org/10.1609/aaai.v39i27.34996>
* Jo Anne Bennett, PhD, RN*
*Consultante / Conseillère technique *
*Recherche en politique de la santé *
*Renforcement des systèmes de santé*
*Institut National de Santé Publique*
*Direction Nationale de l'Épidémiologie et de la Lutte contre la
Maladie *
*Ministère de la santé – République de Guinée*
*+224 626 95 17 28 <+224%20626%2095%2017%2028>*
* Il n’y a qu’une seule preuve de capacité : l’action*
* (*Marie von Ebner-Eschenbach)
HIFA profile: Jo Anne Bennett is a Senior Research Scientist at the New York City Dept of Health and Mental Hygiene in the USA. Professional interests: Public health, Health system strengthening, immunizations, disaster mitigation,and response preparation, workforce development, HIV. Joannebenn@gmail.com