Nutrition Conversations · The Canadian Nutrition Society

L'intelligence artificielle dans la recherche en nutrition? Entre promesses et réalité avec Dr Benoît Lamarche / Artificial Intelligence in Nutrition Research? Between Promises and Reality with Dr. Benoît Lamarche

·29 min·2 clips

AI could predict a diabetic's insulin needs by analyzing a photo of their meal.

This French-language episode of Nutrition Conversations features Dr. Chérine as host and Professor Benoit Lamarche, scientific director of the Nutriss research centre at Universite Laval, as guest. Lamarche entered AI through interdisciplinary exposure to AI specialists within the Nutriss centre rather than formal AI training. He defines AI broadly as computer programs that mimic human reasoning to analyze phenomena and make decisions, and divides the field into three main categories: machine learning (algorithms that classify and predict, such as identifying liver disease risk), deep learning (multiple layered neural networks that handle greater complexity), and natural language processing (as in ChatGPT and Siri). Regarding nutrition specifically, Lamarche explains the field stands to benefit significantly from AI but lags behind sectors like automotive and aerospace. Key promising applications include photo-based dietary assessment, where smartphone apps photograph a meal and an algorithm estimates caloric content and macronutrient breakdown. He describes his team's work on this challenge, noting that identifying an apple versus a banana is straightforward, but distinguishing a chicken pie from a salmon pie, or estimating portion volume, remains technically difficult. Another application involves analyzing social media content using natural language processing to geolocate discussions about junk food by province, which could inform public health policy. Precision nutrition — using omics data to identify individuals at risk of specific diseases and tailor dietary recommendations accordingly — is presented as a transformative long-term possibility, along with precision public health that could generate region-specific dietary guidelines rather than one-size-fits-all national guides like Canada's Food Guide. Lamarche's key caution is the garbage-in-garbage-out principle: if the underlying nutritional data is low quality (as it often is with 24-hour recalls and food frequency questionnaires), no AI model can compensate. He also flags the black-box problem in deep learning, where outputs are produced but the reasoning behind them is opaque, making results difficult to interpret or act on clinically. He calls for cross-disciplinary education and genuine cultural exchange between nutrition researchers and AI methodologists, describing early friction between his doctoral student Melina and AI collaborators who approached data very differently. He recommends a 2021 paper by Melina Cote and himself titled Artificial Intelligence in Nutrition Research: Perspectives on Current and Future Applications in the journal Applied Physiology, Nutrition and Metabolism.

As heard by us

A sober look at AI's real uses and hard limits in nutrition research.

The episode treats artificial intelligence as a serious research tool rather than a shortcut. It moves from plate recognition and assessment to the idea that machine learning could estimate carbohydrate load, predict glycemic response, and make sense of nutrition talk on social…

Read the full review in PlayNext →

Why you'd press play

You want the promises-and-reality view of how AI fits into nutrition research.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Nutrition Conversations