Large Language Model Agents for Improving Engagement with Behavior Change Interventions: Application to Digital...

Area of Science:

• Digital health interventions
• Behavioral science
• Artificial intelligence in healthcare

Background:

• Self-directed wellness engagement often declines over time.
• Traditional social support for behavior change is often inaccessible.
• Large Language Models (LLMs) show potential for emulating social support.

Purpose of the Study:

• To assess the impact of LLM agents on user engagement with mindfulness exercises.
• To explore the effectiveness of information-providing vs. self-reflection-facilitating LLM agents.
• To investigate LLM-driven social support for sustained behavior change.

Main Methods:

• Two randomized experiments were conducted.
• Study 1: A single-session study with 502 crowdworkers.
• Study 2: A three-week study with 54 participants, comparing two LLM agent types.

Main Results:

• Both LLM agents increased users' intentions to practice mindfulness.
• The information-providing LLM agent with a friendly persona significantly improved exercise engagement.
• LLM agents show promise in bridging the social support gap for digital health.

Conclusions:

• Specific LLM agent designs can enhance engagement in digital health interventions.
• LLM-powered social support may be a scalable solution for behavior change.
• Further research into LLM applications for in situ behavior change is warranted.