Challenges of diet planning for children using artificial intelligence.

Area of Science:

• Pediatric Nutrition
• Artificial Intelligence in Healthcare
• Dietary Planning Systems

Background:

• Diet planning for childcare centers is complex due to nutritional and developmental knowledge requirements.
• Artificial intelligence (AI) offers potential solutions for optimizing diet design in complex scenarios.
• This study evaluates AI-generated diets for children aged 3-5 years.

Purpose of the Study:

• To develop and evaluate AI-driven solutions for children's diet planning.
• To compare the nutritional adequacy and composition of AI-generated diets against human-designed diets.
• To explore the utility of AI in addressing the complexities of optimal diet design for young children.

Main Methods:

• Developed two AI solutions using a generative adversarial network (GAN) and a reinforcement learning (RL) framework.
• Trained AI models to generate daily diet plans for children aged 3-5 years.
• Experts evaluated human- and AI-generated diets based on nutritional adequacy and diet composition.

Main Results:

• Reinforcement learning (RL)-generated diets received higher expert ratings for nutritional adequacy when only nutrient information was provided (P < 0.001).
• Human-designed diets were rated more favorably for diet composition when food names (composition information) were available to experts.
• AI demonstrates potential in nutritional adequacy, while human expertise remains crucial for compositional aspects.

Conclusions:

• This is the first study demonstrating AI development and evaluation for pediatric diet planning.
• AI-assisted diet planning for children is feasible, emphasizing the importance of composition compliance.
• Further interdisciplinary collaboration is essential to enhance AI solutions for children's dietary well-being.