Causal AI, online surveys, and synthetic data: can we ensure reproducibility in nutrition research?

Public health and nutrition fields are witnessing a notable rise in AI-powered tools for knowledge synthesis, data gathering, and generation, such as online surveys and synthetic datasets. These alternatives to traditional self-reported responses and electronic health records aim to address cost, scalability, timeliness, and privacy challenges. Online surveys promise the timely collection of opinions, beliefs, attitudes, and practices. Synthetic datasets attempt to replicate real-world data's characteristics, structure, and statistical properties without directly reproducing individual records. In public health and nutrition research, these datasets could support studying trends and risk factors while safeguarding privacy. For example, with synthetic datasets we can simulate dietary patterns, preferences, and nutritional needs across diverse populations, training AI models for personalized nutrition or meal recommendation systems. Synthetic lipidomics and metabolomics datasets can help to evaluate computational methods for understanding the effects of ultra-processed foods on health. As the next logical step, we could envision creative ways to synthesize complex information with Causal AI, a branch of AI focused on modeling, understanding, and inferring causal relationships from data. This approach has the potential to enhance evidence-based public health interventions. However, with these advancements we also should raise critical questions: Are online responses generalizable? Are synthetic data trustworthy? Are we overly reliant on these emerging tools? In this talk I will discuss these concerns, emphasizing the importance of reproducibility, replicability, and critical evaluation in using AI-powered tools. I will explore the benefits of these approaches, highlight common pitfalls in their design, implementation, and interpretation, and offer solutions for establishing robust reproducibility metrics to advance research and policy in the digital era.