Chorus of the Past: Toward Designing a Multi-agent Conversational Reminiscence System with Digital Artifacts for Older Adults

Research Background and Issues

• What problems or challenges did the authors identify?

  • Older adults can improve their mental health, emotions, and social interactions through reminiscing about the past. However, this practice is hindered by obstacles such as difficulties in memory recall, lack of conversational opportunities, and the absence of suitable conversational partners.
  • Traditional reminiscence practices rely on personal photos as memory cues and interactions with family or friends, but these conversational partners may not always be available.
  • In the field of technology-supported reminiscence, the potential of AI-driven conversational systems to simulate human dialogue for memory support has not been fully explored.
  • There are significant research gaps regarding the roles of various memory cues (e.g., personal photos and general nostalgic objects) and how to design AI agents to optimize this experience.

• Why is this issue important?

  • Reminiscing about the past has significant benefits for the mental health of older adults, but these benefits are not widely accessible due to the lack of suitable reminiscence environments in real life.
  • AI-driven conversational systems can overcome interpersonal limitations and provide older adults with opportunities to reminisce anytime and anywhere.
  • Understanding how to effectively design such systems is of great value for supporting the reminiscence practices of older adults.

• Research Motivation and Related Work

  • With the development of large language model (LLM) technology, AI agents can interact dynamically and naturally with users through conversation.
  • Existing research indicates that memory cues (e.g., photos or general nostalgic objects) can influence different types of memory processes, such as triggering personal memories or collective memories.
  • By using the technological probe "ReminiBuddy," the study explores how the roles of AI agents and memory cues influence the reminiscence behaviors of older adults.

Solution

• What methods or solutions did the authors propose?

  • Developed a multi-agent conversational reminiscence system called "ReminiBuddy."
  • Two conversational agents were designed with distinct identities: one representing an "older adult" and the other a "younger person," providing differentiated conversational perspectives.
  • The system uses personal photos and 3D models of nostalgic objects as memory cues to facilitate reminiscence through dialogue with the agents.

• What are the innovative aspects of this solution?

  • The dual-agent identity setup (older adult and younger person agents) is used for the first time to provide diverse perspectives in reminiscence interactions.
  • Combines personal photos and general nostalgic objects as cues, creating an integrated experience of personal and collective memories.
  • Implements a dynamic topic-switching mechanism (Artifact Switching Agent) to enhance the natural flow of conversations.

• What key technologies were used in the implementation steps?

  • Graphics and Interaction Design: Utilized glasses-free 3D screens to display narrated objects and photos, enhancing visual effects.
  • Multi-Agent Roles: Designed two types of agents (older and younger), with language styles and avatar designs enhancing their credibility.
  • Dynamic Cue Switching: Employed a semantic similarity search mechanism to switch between different memory cues.
  • Eye-Tracking: Used eye-tracking data to identify the agent the user was focusing on, thereby triggering natural conversational responses.

Research Outcomes

• What specific outcomes were achieved?

  • The system was proven to successfully enhance the reminiscence experience of older adults.
  • The older adult agent was more effective in eliciting deep reminiscence content, while the younger agent had stronger emotional appeal.
  • Personal photos primarily triggered personal memories, while nostalgic objects were more likely to evoke collective memories and reflections on the past and present.

• What advantages does it have compared to existing solutions?

  • Unlike traditional solutions that rely on single memory cues, ReminiBuddy combines personal photos and nostalgic objects to deepen reminiscence.
  • The use of multi-agent dialogue provides dual perspectives, enhancing the overall experience.
  • Dynamic topic-switching and eye-tracking response mechanisms make the system more natural and engaging for users.

• What were the experimental or evaluation results?

  • Participants reported improved emotions after using the system and expressed willingness to use it in their daily lives.
  • The younger agent was more effective in sustaining long conversations, while the older agent was preferred for in-depth discussions.
  • The 3D display of nostalgic objects enabled participants to recall more comprehensive and vivid memory content.

• Limitations and Future Directions

  • Limitations:
    • The participant group was predominantly well-educated and familiar with digital technology, so the system's applicability to users with lower education levels or more conservative backgrounds needs further study.
    • The potential for personal photos to be converted into 3D formats to enhance interaction has not yet been explored.
    • The experimental environment limited the time and scenarios for free-form interaction, which may differ from real-life home applications.
    • The system's adaptability to different cultural contexts remains to be validated.
  • Future Directions:
    • Explore more diverse agent identity settings, such as cultural backgrounds and personality traits.
    • Develop a broader and customizable library of nostalgic objects and AI models for dynamically generating memory cues.
    • Enhance conversational dynamics to enable the system to respond more flexibly to users' immediate emotions and feedback.
    • Expand research on the system's applicability in cross-cultural environments.