Remembering with Reminiscope: Codesigning with Generative AI for Reminiscence Among Older Adults
Lisha Zhu, Rui Qi, Siyuan Huang, Xueliang Li · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791390
Summary
This CHI 2026 paper presents two linked studies investigating how generative AI can support reminiscence among older adults, both individually and in group settings. In Study 1, the authors conducted individual co-design sessions with 16 residents of a care home in Shenzhen, China (aged 76-95, four with minor symptoms of memory loss). Each participant used a textile-collaging toolkit (wool yarn, plastic tube, felt, fabric scraps on a 20x20cm board) to craft an artwork representing a cherished memory, then collaborated with a multi-agent generative AI system (Qwen 2.5 and Llama 3.2-vision LLMs, Flux+IP-Adapter image generation, Runway Gen-4 video generation) to transform the collage and their verbal description into a 5-second memory scenario video, viewed through a Meta Quest 3 VR headset. Study 2 introduced Reminiscope, an interactive cabinet artifact with a Google Cardboard viewing window, rotating handle for height adjustment, and a tactile touchpad with vibration motors and force-sensitive resistors, designed to let 15 participants (14 returning from Study 1) share their AI-generated memory scenarios in group workshops. The researchers used think-aloud protocols, semi-structured interviews, Likert-scale ratings of human-AI collaboration, and reflexive thematic analysis across 1,793 minutes of audio data to explore older adults' attitudes toward GenAI as a creative collaborator and how tangible AI artifacts can facilitate group reminiscence and social bonding in care home settings.
Key findings
Textile-collaging as a slow, hands-on craft helped participants recall richer memory detail than verbal prompts alone, and the tactile process surfaced details (colours, spatial arrangement, sensory cues) that photos could not. Participants varied in how they perceived AI's role: some treated it as a creative partner contributing 50-60% of the final output, others as a pure execution tool scaffolding their own creative direction. Crucially, mismatches between AI output and lived memory - what the authors call 'echoes of the past' - often triggered DEEPER reminiscence as participants corrected the AI with additional detail (e.g., P11 elaborating the Hongshan Hotel grand hall when the AI got the setting wrong). Co-designing with AI reduced technology anxiety and countered fears of being 'left behind' by rapid AI advancement. In Study 2, Reminiscope's tangible form triggered curiosity and turn-taking; participants spontaneously guessed the authorship of others' memories, leading to untold stories surfacing (P12's first-choice university admission, P14's career as a primary-school teacher) and strengthening social bonds among care-home residents who were acquaintances but not close. Participants expressed bittersweetness when contrasting past joy with present circumstances (illness, loss of spouse), suggesting savoring - not just recall - as a design orientation.
Relevance
For accessibility practitioners working with older adults, dementia-adjacent populations, or anyone designing memory-support technologies, this paper offers concrete evidence that generative AI is most effective as a TANGIBLE design material woven into craft and social ritual, not as an autonomous content producer. The finding that AI inaccuracies can trigger richer recall challenges the usual HCI assumption that hallucination is always harmful - ambiguity becomes a design resource for reminiscence. The Reminiscope prototype also demonstrates accessibility-aware physical design: adjustable viewing height accommodates wheelchair users, multimodal vibrotactile feedback supports varied sensory preferences, and the cabinet form supports group use without imposing turn-taking rules. Limitations practitioners should weigh: the care-home sample is culturally specific (urban China), most participants had low prior AI literacy, generated video quality was technically limited, and the study does not assess use by people with moderate-to-severe cognitive impairment or sensory disabilities beyond minor memory loss.
Tags: Generative AI · older adults · reminiscence · human-AI co-design · reminiscence technology · tangible interaction · aging · memory · wellbeing · participatory design