Episode 7 Transcript Shirin: Hello Nutrition Conversation listeners! Artificial intelligence (AI) is a rapidly evolving field that offers unparalleled opportunities for progress and applications in many areas of health, including nutrition. In this episode, I'm pleased to chat with Benoit Lamarche, full professor at the School of Nutrition and holder of the Chair of Nutrition at Université Laval. He is also Scientific Director of the FRQS-funded Centre de recherche sur la nutrition, la santé et la société, also known as NUTRISS. He has published over 400 peer-reviewed articles on physiological, clinical, epidemiological and public health issues related to diet and health. He has contributed to the training of more than 65 master's, doctoral and post-doctoral students. Benoit has a reputation for being an excellent mentor to many trainees, and I have had the privilege of working closely with him on several opportunities and activities related to trainees at CNS. I can't overlook the fact that Benoît is a Canadian speed skater and competed in both the 1984 and 1988 Winter Olympics. With that, I welcome you, Benoit, to episode 7 of Conversations Nutrition. Benoît: Thank you for the kind invitation. It's my pleasure. Yes, I must clarify that I'm not an expert in artificial intelligence. However, as you mentioned, I oversee the nutrition center, which has researchers from diverse backgrounds, including science, engineering, and AI specialists. Through collaboration with these researchers, I've been exposed to the potential of using AI in nutrition. This exposure sparked my interest in the field. I've always been passionate about new measurement and analysis methods. The confluence of these factors led me to embrace the use of AI tools to address challenges in nutrition. Shirin: That's fantastic. Now that we have some context, let's begin with some general questions that might interest those of us who aren't directly involved in this field. Could you provide a brief explanation of what the term "artificial intelligence" actually means? Benoît: Certainly. Artificial intelligence, in essence, encompasses methods and approaches that seek to mimic human thinking, reasoning, and intelligence in order to make decisions and analyze phenomena. It involves the development of computer programs and algorithms that attempt to emulate human cognition in various contexts. This can range from medical applications to driving vehicles. AI is a vast and rapidly evolving domain. Shirin: Now that we have a foundational definition, could you give us an overview of the main methods employed within the field of artificial intelligence? Benoît: While I'm not an AI expert, I've gained insights from collaborating with experts over the years. AI encompasses a wide range of methods and approaches. These are often categorized into three primary groups. The first is machine learning, which involves algorithms and programs that facilitate predictions. For example, they can analyze medical images to predict whether a person is healthy or ill based on associated data. A more advanced version of this is deep learning, which employs multiple layers of algorithms that learn from each other, enabling more complex predictions. Deep learning involves neural networks, which resemble the brain's interconnected neurons and can predict diverse phenomena. Natural language processing is another category, as evidenced by AI-driven chatbots that understand and respond to human speech. All these methods have applications in various domains, including driving cars and even sending rockets to the moon. While technology fields are highly advanced, nutrition is still catching up. Shirin: Excellent. Thank you for shedding light on that. If we consider our own field, does nutrition research stand to benefit from AI? Benoît: Absolutely. Nutrition research can certainly gain a lot from AI, even though its application is not yet optimized. AI is a relatively new field with expanding access to expertise and methods. In nutrition, there are many areas where AI can be of great help. For instance, we deal with massive datasets in fields like omics, genomics, metabolomics, proteomics, and epidemiology. These datasets are rich with numbers and information, which AI tools can effectively process and analyze. Additionally, there's the challenge of assessing dietary habits, an area where traditional tools like 24-hour recalls and frequency questionnaires have limitations. AI could potentially assess food intake by analyzing photos taken with smartphones. Although it's challenging, some applications can already estimate calorie content. While there's progress to be made, AI holds promise in nutrition research. Shirin: Certainly. So, you've started to touch on AI's potential applications in our field. Could you elaborate on how you and your team have utilized AI in your research? Benoît: We're currently working on several projects. For example, I mentioned the assessment of dietary habits through smartphone apps. People can photograph their meals, and the app predicts nutritional aspects like fiber content or glucose levels. There's room for improvement in this area. For instance, for individuals with diabetes, we might predict glucose levels based on the meal's contents and then estimate their glycemic response—a sort of personalized prediction of their insulin needs. However, there's a challenge in distinguishing between foods in a photo—identifying steak versus salmon, for instance. It's not easy, but we're researching solutions. Another project involves analyzing social media content. Using AI, we study Twitter and Instagram messages related to nutrition and can geolocate them, revealing regional discussions on specific dietary topics. This can inform intervention strategies. The possibilities are vast, from personalized nutrition to public health initiatives. Shirin: Those are fascinating examples. Now, if someone uses an app to photograph their food and assess its nutritional content, could you discuss how this approach compares to traditional nutritional assessment methods? Benoît: There haven't been many formal comparisons between these approaches yet. The core challenge is discerning the real data. While a food photo seems like the real data, it doesn't indicate whether the person ate everything or whether others shared the meal. We can't fully know the true data, making comparisons complex. However, there are some studies. One example involves an app called Kinoa, which compared its assessments to traditional 24-hour recalls or questionnaires. While it's not definitive, the results showed some validity. Each tool has its advantages and drawbacks. Do people capture all their meals in photos? Do individuals accurately report all their food in a 24-hour Shirin: Yes, great. Thank you. That's very interesting. My next question is about the transformative impact of artificial intelligence across various fields. In the context of nutrition, what are the most promising areas for its application? Benoît: We have already discussed food evaluation, and I believe that leveraging the potential of these tools, particularly in image analysis and prediction, can lead to significant advancements. Furthermore, these tools can enhance our understanding of complex food patterns. For instance, we can develop nutritional quality scores like the Healthy Eating Food Index, which outlines healthy food recommendations in Canada. While the measure seems simple, the intricate nature of food relationships presents complexity. AI tools can help us grasp this complexity. Additionally, AI may potentially enhance disease prediction associated with food consumption. For instance, traditional techniques like logistic regression can analyze associations between specific food patterns, like saturated fat consumption, and the risk of heart attacks. Nonetheless, emerging data suggests that AI, with its automatic learning models and deep learning capabilities, could potentially outperform traditional methods in predicting health-nutrition links. Moreover, AI holds promise in precision areas like precision nutrition and public health. Precision nutrition involves predicting health outcomes based on comprehensive data sets encompassing aspects like genomics, proteomics, and metabolism. We can then tailor interventions for individuals based on their unique profiles. Public health precision focuses on identifying subgroups at risk within a population, offering targeted interventions. While currently speculative, this approach could potentially lead to region-specific food guidelines, acknowledging diverse health concerns. These promising applications highlight the vast potential of AI's capabilities. Shirin: Absolutely, there are undoubtedly numerous advantages and potentials, but you also briefly touched upon limitations. Could you elaborate on the limitations of AI's application in our field? Benoît: Certainly, like any tool, AI comes with limitations. While AI experts are better equipped to address these limitations, I can offer some insights. In nutrition, a fundamental principle applies - "garbage in, garbage out." If we input low-quality data, we will obtain poor-quality results. For instance, if nutritional data derived from methods like 24-hour recalls or frequency questionnaires is not of high quality, any AI model built upon such data will yield unreliable results. AI cannot enhance data quality if the data itself is flawed. Hence, prudence is essential. This issue isn't exclusive to nutrition; it spans across all AI applications. Another challenge arises with the use of complex models like deep learning. These models function as black boxes, generating results without a clear understanding of the underlying processes. Such opacity can be problematic when attempting to interpret results. While not a strict limitation, it's crucial to exercise caution. The excitement around new methodologies like AI is valid, but we mustn't rely solely on AI to solve all our analytical challenges. The constraints are similar to other tools, and it's crucial to use AI judiciously. Shirin: Indeed, you've highlighted both the potential and the limitations. There's much work to be done in this evolving field. How do you envision the future evolution of AI's role in our domain? Benoît: The future evolution of AI's role hinges on several factors. One key aspect is fostering collaboration with AI methodology experts, who possess distinct perspectives from ours. It's not about criticism but mutual understanding. We need to bridge the gap in our languages - statistical approaches differ, and harmonizing them is the initial step. It's a learning curve and requires respectful exchange. An example comes to mind: my doctoral student Melina, I won't name her entirely, faced initial challenges while working with AI experts. We share great collaboration, but initially, there was a language barrier. Overcoming this is paramount. Additionally, training the next generation is crucial. Students graduate without exposure to AI, lacking insight into its ethics, limits, and advantages. Training them to effect positive change in this evolving field is vital. It's more than just incorporating AI experts' methodologies into our field; it's about a harmonious blend between AI and our expertise. These considerations pave the way for a promising trajectory. Shirin: Absolutely, there's a learning curve in progress. I still have many questions to ask, but let me conclude with this: If you could leave our audience with a key takeaway, what would it be? Benoît: Summarizing the insights shared over the past 25-30 minutes, AI is an immensely promising field, particularly evident in other sectors. While nutrition is still emerging in its integration of AI, we must exercise caution while employing these tools. AI doesn't address the foundational challenges in nutritional evaluation, such as accurate dietary assessment. It's imperative that we collaborate with the right experts and maintain an open-minded approach. Using AI responsibly, in harmony with existing expertise, will facilitate the transformation of our field in the best way possible. Shirin: Thank you, Benoit, for sharing this enlightening message. Before we conclude, I'd like to recommend an article to our listeners. It's written by Melina Côté, a doctoral student, and Professor Benoit Lemarche in 2021. It's titled "Artificial Intelligence in Nutrition Research: Perspectives on Current and Future Applications," and it's published in the Applied Physiology and Nutrition and Metabolism journal. While it's in English, it offers a comprehensive overview of AI methods and their application in the field of nutrition. The potential of AI is vast, and witnessing how you and your team are harnessing its capabilities to accelerate discoveries is inspiring. Thank you, Benoit, for taking the time to share your insights into this evolving field. Benoît: Thank you, Shirin. And congratulations on your successful presentation in French. You did an excellent job. Shirin: Thank you for the rewarding challenge. It was a bit challenging, but I appreciate your encouragement for this French episode. Merci, Benoit. Benoît: You can continue, you're doing great. Shirin: Merci.