AI Chatbots and Nutritional Advice for Obesity
Author Information
Author(s): Valentina Ponzo, Rosalba Rosato, Maria Carmine Scigliano, Martina Onida, Simona Cossai, Morena De Vecchi, Andreas Devecchi, Ilaria Goitre, Enrica Favaro, Fabio Dario Merlo, Domenico Sergi, Simona Bo
Primary Institution: Department of Medical Science, University of Turin
Hypothesis
Can general-purpose AI chatbots provide accurate and reliable nutritional advice for patients with obesity and comorbidities?
Conclusion
AI chatbots showed limited accuracy, reproducibility, and consistency in providing dietary advice for complex clinical scenarios.
Supporting Evidence
- ChatGPT 3.5 had the highest accuracy rate at 67.2%.
- ChatGPT 4.0 and Claude achieved the highest completeness scores at 77.8%.
- Reproducibility was highest for Chatsonic at 86.1%.
- Major inconsistencies were found in protein intake recommendations across several chatbots.
Takeaway
AI chatbots can give basic diet advice, but they often get it wrong, especially for people with health issues.
Methodology
The study evaluated 10 AI chatbots by asking them for dietary advice on two clinical cases of obesity over three consecutive days.
Potential Biases
The chatbots may provide conflicting or inaccurate dietary advice, especially for complex health conditions.
Limitations
The study had a small sample size and focused on only two clinical cases, which may limit the generalizability of the findings.
Participant Demographics
Two hypothetical cases: a 35-year-old male with obesity and a 65-year-old female with obesity and multiple comorbidities.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website