Artificial intelligence for Amazonian health: validation is the price of entry
Abstract
At the AI as Catalyst meeting in Manaus (July 28–29, 2026; https://ia-como-catalisadora-manaus.netlify.app), participants ran a simple exercise. Working in teams, they wrote about ten clinical cases and put each one to widely used general-purpose chatbots twice: first in standard medical terminology, then in the terms and expressions a patient from the Amazon might use, with the same clinical facts, but carried by local language. The register of the answers changed with the framing, and so did their technical depth; more consequentially, the models inferred conditions from how the patient spoke rather than from what the case described.
// Source
Authors: Matheus Serrao, Michella Lasmar, Luis Nakayama, Gustavo Monnerat, Leo Celi
Institutions: Universidade de São Paulo, Harvard University, Regional Health, Massachusetts Institute of Technology, Universidade Federal de São Paulo, Universidade do Estado do Amazonas, Brazilian Society of Computational and Applied Mathematics