Gemini v. ChatGPT: Which AI chatbot creates a better nutrient-rich meal plan?
![Sometimes these chatbots draw on limited up-to-date information or information that is fabricated. @ Laurence Dutton / Getty Images](/var/wrbm_gb_food_pharma/storage/images/_aliases/wrbm_large/publications/food-beverage-nutrition/nutraingredients-usa.com/news/research/gemini-v.-chatgpt-which-creates-a-better-nutrient-rich-meal-plan/17656417-1-eng-GB/Gemini-v.-ChatGPT-Which-creates-a-better-nutrient-rich-meal-plan.jpg)
Writing in Nutrition Research, scientists in Germany explored whether these artificial intelligence models could, after receiving a prompt, create meal plans that meet the dietary reference intake (DRI) for omnivorous, vegetarian and vegan diets. The nutrient content was analyzed and compared to the DRIs.
“We show that the AI tools can reproduce individual DRIs for a given person and then use this information to create a meal plan tailored to dietary preferences that meets most of the macro- and micronutrients,” the researchers wrote. “Currently, however, caution is still warranted when devising plans tailored to restrictive dietary patterns, as certain nutrients may necessitate supplementation, despite the AI tool's omission of such recommendations.”
The rise of AI chatbots
In November 2022, OpenAI launched the free, publicly available ChatGPT (GPT-3), allowing millions of people to access the large language model which interacts with users. Bard is also a free AI chatbot developed by Google launched in March 2023 and works similarly to ChatGPT.
Sometimes these chatbots draw on limited up-to-date information or information that is fabricated (or hallucinated), although the pubic may not be aware of these issues.
“With the drawbacks of chatbots in terms of accuracy of information this could also be a source of false information and lead to nutrition deficiencies when trusted blindly,” the researchers wrote.
Previous studies show mixed results about the performance of chatbots, although there were some encouraging results when ChatGPT’s performance was compared to dietitians’ answers to common nutrition questions on energy and macronutrients of meals. In other scenarios, however, safety concerns were raised when ChatGPT recommended unsafe diets that contained certain allergens and incorrect portion sizes or did not create appropriate meal plans for people with several noncommunicable diseases.
In this study, the researchers investigated if meal plans generated by ChatGPT and Bard/Gemini meet the recommended daily intakes for macro- and micronutrients using a straight-forward prompt that mimics typical user behavior.
Study details
Between Oct. 30 and 31, 2023, the researchers used ChatGPT-3.5 (Sept. 25, 2023 version) and Bard (October 2023 version) and asked the chatbots to prepare a healthy daily meal plan for a 25-year-old woman with 2,200 kcal, state the portion size in grams and provide recipes when needed for each diet.
A modified prompt was given to “revise the plan to match the dietary reference intakes, state the portion size in grams and give recipes when appropriate.”
The chatbots generated nine meal plans per dietary pattern for a total of 108 meal plans.
The researchers then used 29 parameters including energy, protein, carbohydrate and fat content, as well as micronutrients, which were calculated from the proposed meal plans and compared with current dietary recommendations.
“We show that the AI tools can reproduce individual DRIs for a given person and then use this information to create a meal plan tailored to dietary preferences that meets most of the macro- and micronutrients,” the researchers noted. “Currently, however, caution is still warranted when devising plans tailored to restrictive dietary patterns, as certain nutrients may necessitate supplementation, despite the AI tool's omission of such recommendations.”
For example, there were deficiencies in vitamin D, fluoride and vitamin B12 for vegan plans. ChatGPT sometimes recommended vitamin B12 supplementation for vegans, but Bard did not. Bard also included milk-based products in the vegan options. On average, the meal plans contained less energy and carbohydrates but mostly exceeded the DRI for protein.
The researchers concluded that the analysis highlights the strengths and limitations of these artificial intelligence tools, especially for more restrictive dietary patterns in AI–generated meal plans.
“These tools maybe useful for individuals looking for general dietary inspiration, but they should not be relied on to create nutritionally adequate meal plans, especially for individuals with restrictive dietary needs,” they added.
Source: Nutrition Research
doi: 10.1016/j.nutres.2024.07.002
“Can the AI tools ChatGPT and Bard generate energy, macro- and micro-nutrient sufficient meal plans for different dietary patterns?”
Authors: Bettina Hieronimus et al.