Seminar, Hosted by McGill University Department of Epidemiology, Biostatistics and Occupational Health
From the LHIVING Lab to the World: Building and Evaluating Locally Grounded AI for HIV Care and Prevention
Bertrand Lebouché and Yuanchao Ma
Department of Family Medicine, McGill University; CVIS-MUHC; CORE, RI-MUHC; LHIVING Lab; McGill University
Abstract
The talk traces a path from a living lab to international evaluation of health AI. The LHIVING Lab brings together people living with HIV, clinicians, researchers and engineers to co-develop and test digital health tools in real-world care. The speakers describe the master protocol and multidisciplinary co-design behind MARVIN, an AI chatbot for HIV self-management and prevention, and MARVIN-CONNECT, a CIHR-funded study adapting it to PrEP guidelines and care pathways in Canada, Germany and Ghana. The guideline-grounded chatbot gave locally appropriate advice in 80-89% of cases, whereas ChatGPT (GPT-5.6) did so in 49% of cases in Canada and in none in Ghana despite answers that sounded locally tailored. The talk explains why general-purpose AI can give locally inaccurate advice, how a shared architecture with clinician-curated knowledge bases can be adapted to other areas of medicine and care settings, and how locally grounded health AI can be evaluated and governed so that it is equitable and accountable.