"It Seems to Understand My Heart": An Empirical Study of Persona-Driven Persuasive AI Agent for Aging-in-Place in Singapore

Paper Title

"It Seems to Understand My Heart": An Empirical Study of Persona-Driven Persuasive AI Agent for Aging-in-Place in Singapore

Publication Info

• Topic area: Persona-driven AI for aging-in-place support
• Keywords: Persona-driven AI, aging-in-place, empathetic agents, health behavior change, cultural adaptivity, GMMAR framework, Dual-Persona framework, Singapore, IoT sensors, conversational agents

Background and Problem

• Problem / challenge: Existing conversational agents for aging-in-place lack cultural responsiveness, dynamic personalization, and sustained empathetic engagement. They often fail to address the diverse needs of older adults, leading to disengagement and limited long-term impact.
• Significance: Addressing these gaps is critical for supporting the independence, health, and well-being of aging populations, particularly in multicultural contexts like Singapore.
• Motivation and related work: Prior research highlights the limitations of rule-based and static persona systems, as well as the need for culturally aligned, empathetic AI. While advances in large language models (LLMs) enable more natural interactions, they often overlook emotional and cultural nuances. This paper builds on persona-based design and computational empathy to propose a novel solution.

Solution

• Proposed approach: PersonaBot, a persona-driven persuasive AI agent, integrates a Dual-Persona Framework for matching user and agent personas with the GMMAR Framework for structuring empathetic conversations.
• Novelty:
  1. Introduction of the Dual-Persona Framework, dynamically linking user personas with culturally adaptive agent personas.
  2. Development of the GMMAR Framework (Greeting → Mirror → Match → Ask → Recommend) for sustained empathetic engagement.
  3. Integration of IoT sensor data for real-time, context-aware personalization.
  4. Empirical evaluation in a multicultural aging-in-place context in Singapore.
• Procedure and key techniques:
  • User personas are constructed using interviews and psychological models (e.g., HAPA, OCEAN framework).
  • Agent personas are generated with LLMs and AI tools, incorporating cultural, linguistic, and emotional traits.
  • Persona matching and dynamic adaptation refine interactions based on user feedback and sensor data.
  • The GMMAR Framework structures conversations into five phases to build rapport and deliver personalized recommendations.
  • The system operates within ARSmartHouse, a smart home platform with IoT sensors for activity tracking.

Results

• Concrete findings:
  • Perceived empathy increased by up to 48% across dimensions such as emotional support and trustworthiness.
  • Engagement decline was mitigated in PersonaBot Mode, with steadier participation and a 71% increase in participant message word count.
  • Participants valued the agent as a "co-pilot" rather than a caregiver replacement, emphasizing collaborative autonomy.
• Advantage over baselines:
  • PersonaBot outperformed the Non-Persona baseline in engagement, empathy perception, and conversational depth.
  • Daily active user counts were higher and more stable in PersonaBot Mode compared to the baseline.
• Experiments / evaluation:
  • Eight-week field deployment with 8 participants (aged 68–86) in Singapore.
  • Mixed-method evaluation included quantitative metrics (e.g., PETS scores, engagement trends) and qualitative feedback (e.g., interviews, open-ended responses).
  • Study phases included a Non-Persona baseline (Weeks 1–4) followed by PersonaBot Mode (Weeks 5–8).
• Limitations and future work:
  • Small sample size and limited diversity (primarily Chinese participants).
  • Short study duration (8 weeks) limits assessment of long-term engagement.
  • Challenges with sensor accuracy and LLM reliability.
  • Future work includes expanding multilingual capabilities, targeting isolated populations, and conducting longitudinal studies.

Summary

This study introduces PersonaBot, a persona-driven persuasive AI agent designed to support aging-in-place for older adults in Singapore. By integrating the Dual-Persona Framework and the GMMAR Framework, the system dynamically adapts to users' cultural, emotional, and autonomy preferences, fostering empathetic and engaging interactions. An eight-week field study demonstrated significant improvements in perceived empathy, engagement, and health behavior change. Participants valued the agent's collaborative autonomy and cultural adaptivity, though challenges remain in maintaining long-term engagement and addressing ethical concerns around "artificial intimacy." The findings highlight the potential of persona-driven AI in aging-in-place contexts and propose design principles for future systems.