Exploring the Design of a LLM-Based AI Assistant for Mindfulness Practice With Older Adults

Paper Title

Exploring the Design of a LLM-Based AI Assistant for Mindfulness Practice With Older Adults

Publication Info

• Topic area: Human-Computer Interaction (HCI) and AI applications in mental health and well-being.
• Keywords: Large Language Models, mindfulness, older adults, participatory design, socioaffective alignment, AI-guided well-being, personalization, trust, emotional regulation, accessibility.

Background and Problem

• Problem / challenge: Existing mindfulness technologies often lack conversational engagement and personalization, which are critical for older adults. Current AI systems risk socioaffective misalignment, raising concerns about trust, autonomy, and human connection.
• Significance: Mindfulness technologies can promote healthy aging by addressing physical and mental health challenges. Designing AI systems that align with older adults' values and needs is essential for meaningful engagement and well-being support.
• Motivation and related work: Prior research has explored mindfulness technologies and conversational AI but has not sufficiently examined LLM-based mindfulness systems for older adults. This study addresses gaps in understanding how generative AI can support well-being while considering age-specific socioaffective dynamics.

Solution

• Proposed approach: Development and participatory evaluation of LugnAI, a prototype LLM-based AI assistant designed to guide mindfulness practices for older adults.
• Novelty:
  1. Investigation of older adults' perceptions of AI-guided mindfulness through participatory workshops.
  2. Identification of design preferences and resistances specific to older adults.
  3. Formulation of socioaffective alignment principles for LLM-based mindfulness technologies.
  4. Development of participant-informed design considerations addressing personalization, accessibility, and human connection.
• Procedure and key techniques:
  • Conducted two participatory workshops with 16 older adults (ages 65+).
  • Developed LugnAI using a retrieval-augmented generation (RAG) workflow and Swedish-language mindfulness materials.
  • Evaluated user interactions, preferences, and resistances through individual sessions, focus group discussions, and co-design activities.
  • Applied reflexive thematic analysis to identify themes and formulate design considerations.

Results

• Concrete findings:
  • Participants valued personalization but emphasized optional, user-controlled features to maintain autonomy.
  • Sensory preferences (e.g., voice tone, visuals, soundscapes) varied widely and required situational adaptivity.
  • Trust was relational, built incrementally through transparency, perceived usefulness, and comfort with technology.
  • Resistance to AI companionship highlighted the importance of preserving human connection.
  • Technical disruptions (e.g., speech recognition errors) reduced immersion and trust.
• Advantage over baselines:
  • LugnAI provided dialogic, personalized mindfulness guidance, contrasting with static, pre-recorded mindfulness apps.
  • Contextual responsiveness enhanced relevance compared to traditional systems.
• Experiments / evaluation:
  • Workshops included individual interactions with LugnAI, focus group discussions, and co-design activities.
  • Participants completed a Chatbot Usability Questionnaire (CUQ) and provided qualitative feedback.
  • The study involved 16 participants, with 15 attending Workshop 1 and 10 attending Workshop 2.
• Limitations and future work:
  • Small, demographically limited participant group; broader diversity needed in future studies.
  • Workshop environment may not fully capture natural, home-based mindfulness use.
  • Prototype reliance on a single LLM provider (OpenAI) and curated Swedish-language materials limits generalizability.
  • Future research should explore longitudinal use, multimodal interactions, and diverse participant groups.

Summary

This study explored older adults' perceptions of a LLM-based AI mindfulness assistant (LugnAI) through participatory workshops. Findings revealed that trust, personalization, and sensory adaptivity are critical for meaningful engagement, while preserving human connection and autonomy remains essential. Participants expressed diverse preferences and resistances, highlighting the importance of socioaffective alignment in AI design. The study proposed nine design considerations addressing personalization, accessibility, and emotional regulation practices. These insights contribute to HCI and AI research on well-being technologies and inform future development of LLM-based systems for older adults and beyond.