AI shows potential in providing nutritional information, researchers find

Key takeaways:

• AI chatbots had similar estimations vs. nutritionists for fat, carbohydrate and energy content.
• Although AI will not replace nutritionists, it could provide real-time food analysis, researchers said.

Artificial intelligence performed similarly to nutritionists in estimating energy and macronutrient contents, according to a study published in JAMA Network Open.

“In a digital world, people increasingly rely on the internet for food-related and nutrition-related information,” Yen Nhi Hoang, MSc, from the College of Nutrition at Taipei Medical University in Taiwan, and colleagues wrote. “However, a recent report showed that almost one-half of online, nutrition-related information was inaccurate (48.9%) or was of low quality (48.8%).”

They added that although artificial intelligence (AI) has shown potential in streamlining public information, little research has been done on how accurately chatbots can respond to nutrition-related questions.

Thus, the researchers compared the calorie and macronutrient content estimations for 222 food items across eight menus from two chatbots — ChatGPT-3.5 and ChatGPT-4 — to those of nutritionists.

“The consistency of AI responses was determined on the basis of the [coefficient variation (CV)] for each food item across five repeated measurements,” Hoang and colleagues wrote.

The accuracy of AI responses was determined if answers were within ±10% or ±20% of the ground truth level energy or macronutrients.”

Overall, there were no significant differences between the chatbots’ and nutritionists’ estimations for carbohydrate, energy and fat contents of the eight menus, although there was a significant difference in protein estimation. Although ChatGPT-4 performed better than ChatGPT-3.5, it overestimated protein.

In addition, both chatbots provided accurate energy contents for around 35% to 48% of the 222 food items within ±10%, with a CV of less than 10%.

The findings suggest that AI “can be a useful and convenient tool for people who want to know the energy and macronutrient information of their food,” Hoang and colleagues wrote.

“Although AI chatbots cannot replace nutritionists, they may provide real-time analysis of foods, and the capacity to harness AI technology in a supportive role may fundamentally transform the way nutritionists communicate with patients,” they wrote.

Still, ChatGPT is limited in its ability to provide personalized dietary advice, like specific portion sizes and nutrition guidelines, Hoang and colleagues pointed out.

“ChatGPT is also unable to provide accurate common household units to consumers,” they wrote. “Portion size and household units vary substantially depending on the food type, preparation method, and regional differences in measurement standards.”

These limitations are likely the result of AI’s general-purpose design “that is not specialized in the field of nutrition and dietetics,” the researchers wrote.

“Future improvements in providing more accurate and practical nutrition information to customers will be important,” they concluded.