Creating customized chatbots with ChatGPT to promote physical activity: a mini review.

Area of Science:

• Health behavior change
• Artificial intelligence in healthcare
• Digital health interventions

Background:

• Large language models (LLMs) like ChatGPT show potential for personalized physical activity interventions.
• Concerns exist regarding AI chatbot accuracy, safety, privacy, and theoretical grounding for health applications.

Purpose of the Study:

• To review methods for customizing ChatGPT-based chatbots for physical activity promotion.
• To outline approaches for evaluating the performance of these AI-driven interventions.

Main Methods:

• Literature search across scientific databases, white papers, and technical reports.
• Identification and synthesis of AI chatbot customization strategies: retrieval-augmented generation (RAG), system prompt engineering, and fine-tuning.
• Review of intrinsic and extrinsic evaluation methods for AI chatbots.

Main Results:

• Three primary customization strategies (RAG, system prompts, fine-tuning) can be used individually or combined.
• RAG improves accuracy by grounding responses in evidence-based guidelines.
• Evaluation requires both model-based testing and human-centered assessment.

Conclusions:

• Customized ChatGPT chatbots offer significant potential for scalable physical activity promotion.
• Safe and effective implementation necessitates meticulous design, rigorous evaluation, and attention to data privacy and operational costs.
• Addressing concerns about AI in health behavior change is vital for widespread adoption.