Integrating Behavior Change Techniques into an AI-Powered Virtual Coach for Mobile Applications

Sammanfattning

This thesis highlights the importance of fall prevention, as falls are a leading cause of injuries. Previous work led to the development of a mobile application that offers personalized training programs focused on fall-prevention exercises, based on the individual’s physical abilities. The application includes a virtual coach, featuring a conversational AI-powered system that enables users to engage in more human-like interactions. The system uses conversational properties from large language models to provide personalized information and guidance. The application is developed using Flutter and the backend is built using Python, which manages the AI model integration and data handling through frameworks like LangChain and FastAPI. While this version of the application does not account for user behavior changes over time, the proposed solution aims to address this limitation by enhancing personalization and implementing motivational approaches to sustain long term commitment. To achieve this, the thesis investigates the integration of behavior change techniques, grounded in the self-determination theory and the persuasive systems design model. These have shown potential in enhancing user motivation, and the thesis will explore how these can be applied to further improve user engagement and fall prevention outcomes. The results indicate that providing choice and tailoring have a statistically significant positive impact on behavior change at a 95% confidence level. Particularly through the implementation of a notification system that allows users to select the timing of reminders and other notifications. While other techniques did not demonstrate statistical significance within this study, their theoretical relevance suggests they may still offer value under different conditions or with alternative implementations. To fully realize the potential of these techniques, further investigation and optimization are necessary.