Rememo: A Research-through-Design Inquiry Towards an AI-in-the-loop Therapist’s Tool for Dementia Reminiscence

Paper Title

Rememo: A Research-through-Design Inquiry Towards an AI-in-the-loop Therapist’s Tool for Dementia Reminiscence

Publication Info

• Topic area: AI-assisted tools for reminiscence therapy in dementia care.
• Keywords: Reminiscence therapy, dementia care, generative AI, human-AI collaboration, therapist tools, memory recall, person-centered care, AI-in-the-loop, cultural specificity, research-through-design.

Background and Problem

• Problem / challenge: Existing technology-mediated reminiscence therapy (RT) tools often focus on replacing human facilitators, overlooking the relational and affective labor critical for effective RT. Additionally, sourcing culturally and personally resonant materials remains labor-intensive for therapy staff, especially in multilingual, multicultural contexts like Singapore.
• Significance: Dementia care requires person-centered interventions to support memory recall, emotional well-being, and identity. Addressing the structural challenges faced by therapy staff, such as understaffing and cultural barriers, can improve the quality of care and therapy outcomes.
• Motivation and related work: Prior work has explored conversational agents, recommender systems, and generative personalization for RT. However, these approaches often neglect the role of human facilitators and the relational dynamics of care. This paper builds on these efforts by focusing on tools that empower therapy staff rather than replacing them.

Solution

• Proposed approach: Rememo, a therapist-oriented tool integrating generative AI to support personalized reminiscence therapy while maintaining therapist oversight and resident agency.
• Novelty:
  1. Development of an AI-in-the-loop model for therapist–AI collaboration, emphasizing human-led facilitation.
  2. Reframing synthetic imagery as a therapeutic support for reconstructive memory work rather than a substitute for authentic photographs.
  3. Design of a hybrid tangible–digital system combining generative AI with physical prompt cards and printed images.
  4. Inclusion of culturally specific and multilingual features tailored to Singapore’s context.
• Procedure and key techniques:
  1. Iterative research-through-design (RtD) process involving therapy staff as co-design partners.
  2. Development of themed prompt cards with multilingual translations and dialect support.
  3. Integration of generative AI engines (e.g., Imagen, Flux.1) for personalized image generation.
  4. Deployment of a mobile web app with optical character recognition (OCR) for card scanning and image generation.
  5. Two-week field study involving 5 therapy staff, 21 residents, and 151 generated images.

Results

• Concrete findings:
  • 151 images generated across 26 sessions, with 33.1% printed as physical media.
  • Imagen was the most used engine (55.0%), followed by Flux (35.1%) and SDXL (9.9%).
  • Average image generation latency: 27 seconds for card-based prompts and 19 seconds for free-text prompts.
  • Residents reported 70–80% accuracy of generated images in evoking memories.
• Advantage over baselines:
  • Reduced logistical preparation time for therapy staff.
  • Enhanced resident engagement and recall through personalized, culturally resonant stimuli.
  • Bridged communication barriers with multilingual and visual aids.
• Experiments / evaluation:
  • Conducted at two nursing homes in Singapore with 5 therapy staff and 21 residents.
  • Sessions included individual (n=19) and group (n=7) formats, lasting 21–55 minutes.
  • Thematic analysis of session logs, surveys, and focus group discussions.
• Limitations and future work:
  • Small sample size and single national context limit generalizability.
  • Latency in image generation disrupted session flow.
  • Future work should explore longitudinal studies, faster AI models, and integration of multimedia stimuli.

Summary

This paper introduces Rememo, a therapist-centered tool leveraging generative AI to support reminiscence therapy for dementia care. Developed through a two-year research-through-design process, the system combines physical prompt cards, a mobile web app, and AI-generated images to ease session preparation, bridge communication barriers, and enhance resident engagement. A two-week field study demonstrated its utility while highlighting challenges such as latency and cost constraints. By advocating an AI-in-the-loop model, the study emphasizes the irreplaceable role of human facilitators in relational and affective care. Future research should focus on improving system efficiency, expanding personalization, and exploring broader applications in dementia care.