Exploring the Role of Generative AI in Dementia Resilience Building Activities: Uncovering Opportunities and Challenges
Sushant Kot, Margi Engineer, Elizabeth Gilman, Christopher Flathmann, Alisha Pradhan, Emma Dixon · 2026 · ACM Transactions on Computer-Human Interaction · doi:10.1145/3773029
Summary
This qualitative HCI study examines how people with Mild Cognitive Impairment (MCI) and mild-to-moderate dementia envision using commercially available generative AI applications in everyday life. Rather than designing for this population from the outside, the researchers centered first-person perspectives—a deliberate counter to the dominant medical and caregiver-focused framing of AI in dementia contexts. Data was collected across three phases between September 2023 and March 2024. Phase I comprised two technology workshops introducing 17 participants to ChatGPT (version 3.0). Phase II consisted of one-on-one contextual interviews with six participants to explore ChatGPT use more deeply. Phase III expanded to four additional workshops with 45 participants covering a broader range of generative AI tools—including Fireflies.AI (meeting transcription), Ohai (scheduling and reminders), Microsoft Copilot (text and image generation), DALL-E, and Scribble Diffusion. In total, 39 unique individuals participated, all members of remote dementia peer-support groups. The study used a TechShop methodology—scaffolded, collaborative technology exploration workshops—followed by reflexive thematic analysis. The research was structured around the American Psychological Association's four resilience-building strategies: fostering wellness, building social connections, finding purpose, and embracing healthy thinking. Findings show that participants envisioned generative AI applications supporting the first three strategies in meaningful ways, but each use case surfaced tensions with the fourth strategy—raising important questions about the mental health implications of AI adoption for people with dementia.
Key findings
Participants identified generative AI use cases across three resilience-building domains: Fostering wellness: LLM-based chatbots were valued for simplifying complex medical information—reformatting dense research abstracts into plain language and offering more accessible health information than traditional search engines. However, participants noted that AI responses sometimes used stigmatizing, disease-centric language (e.g., describing dementia through a tragedy narrative), which conflicted with their efforts to maintain a hopeful outlook. Participants explicitly called for chatbots to be educated to generate more positive, person-centred responses. Finding purpose: Participants saw generative AI as extending their engagement in meaningful activities—writing dementia advocacy materials, drafting memoirs, taking meeting notes as volunteers, and continuing in the workforce longer. A key tension was overexertion: participants recognized that abilities fluctuate over time, and AI tools not calibrated to current capacity could inadvertently encourage overextension. One participant proposed integrating health-tracking apps (e.g., Oura rings) with AI tools to regulate engagement. Building social connections: Participants wanted AI to mediate human-to-human interactions rather than act as social companions. Use cases included preparing questions for medical appointments, drafting messages to disclose a diagnosis to family members, and supporting intergenerational creative activities. A significant tension emerged around disclosure: participants were concerned about whether others would know they were using AI and what that knowledge would mean for how their abilities were perceived—both protecting their self-image and risking withdrawal of care support.
Relevance
This paper directly challenges the paternalistic design assumptions that have dominated AI development for the dementia context. By centering first-person perspectives from people with MCI and dementia—rather than caregivers or clinicians—it surfaces use cases and tensions that external stakeholders are unlikely to anticipate or prioritize. For accessibility practitioners, the key takeaways are: AI tools must avoid stigmatizing or disease-centric representations of disability in their outputs; the social disclosure of AI use carries stakes that designers must explicitly account for; and adaptive AI that responds to fluctuating ability levels is both desired and technically feasible. The study's framework—mapping technology use to the APA resilience-building model—offers a useful evaluative lens for whether AI tools empower or undermine the mental health and self-determination of people with cognitive disabilities. A key limitation is sample homogeneity: all participants were English-speaking, US-based members of dementia advocacy organizations with moderate technological literacy, which limits generalizability to broader dementia populations.
Tags: dementia · generative AI · large language models · resilience · cognitive accessibility · mild cognitive impairment · qualitative research · HCI · dementia advocacy · co-design