Nutrition in the age of AI

AI enables precise digital diet assessment

Off-the-shelf multimodal AI now supports real-time, multilingual dietary tracking with image, barcode, and text input. This markedly improves data quality and reduces reliance on human annotation in digital nutrition cohorts.

Glucose prediction improves with richer diet data

Machine-learning models predict postprandial glucose responses more accurately when detailed diet, glucose, and demographic data are combined. In this cohort, microbiome data are not required to achieve high predictive performance.

Diet regularity matters for microbiome diversity

Daily Healthy Eating Index regularity shows a stronger association with gut microbiome alpha diversity than the overall amount of healthy nutrition. Temporal patterns of intake therefore represent an important nutritional dimension.

The continuing education session “Nutrition in the Age of AI,” delivered by Prof. Dr. Marcel Salathé and organized by Klinik Barmelweid, presents how artificial intelligence currently expands nutritional research and practice through improved dietary data capture and analysis. Prof. Dr. Salathé explains that digital cohorts enable remote collection of longitudinal data on food intake, continuous glucose monitoring, physical activity, sleep, and gut microbiota, and he reports that his group achieves high participant retention in a Swiss cohort of more than 1,000 completers. He shows that machine-learning models predict postprandial glucose responses with substantially greater precision than carbohydrate content alone, and that high predictive performance remains possible even without microbiome features. A central finding from the presented work is that the regularity of healthy eating, assessed through daily Healthy Eating Index measurements, shows a strong association with gut microbiome diversity and appears more informative than the overall amount of healthy eating for this specific outcome. Prof. Dr. Salathé also describes evidence from his group indicating that food additives may adversely affect the gut microbiome, although he emphasizes that this remains an evolving field. A major focus of the lecture is the transition from earlier AI-assisted food logging with human annotation to current real-time, multilingual diet tracking using off-the-shelf vision-language models, which markedly improves scalability and usability for research cohorts. Finally, he outlines ongoing and planned studies on diet, cognitive performance, and longitudinal changes in diet, microbiome, glycemic responses, and health outcomes, while noting that evidence is still lacking on whether continuous glucose monitoring in non-diabetic individuals reduces diabetes incidence or other age-related diseases.