Technology and Dementia Preconference

Area of Science:

• Artificial Intelligence in Healthcare
• Gerontology and Dementia Care
• Human-Computer Interaction

Background:

• Caregivers of people living with dementia (PLwD) experience significant stress, with digital reminder systems offering limited task verification.
• Generative AI (e.g., GPT) presents a novel approach to enhance task verification and support caregiver decision-making in dementia care.
• This study explores the feasibility of an AI-powered system for task verification within a digital reminder framework for PLwD.

Purpose of the Study:

• To assess GPT's ability to generate high-quality, tailored follow-up questions for PLwD using few-shot prompting.
• To evaluate the accuracy of an AI system in identifying concerning responses from PLwD.
• To determine optimal strategies for balancing AI automation with essential caregiver control.

Main Methods:

• Simulated interactions using an anonymized dataset of 64 reminders involving caregivers, PLwD, and an AI system.
• Evaluation of GPT-generated follow-up questions for quality, with and without contextual information.
• Development of a flagging mechanism to categorize responses by concern level (High, Medium, Low), prioritizing safety and hygiene.

Main Results:

• Contextual information and caregiver feedback significantly improved the clarity, specificity, and relevance of AI-generated questions.
• The AI flagging mechanism demonstrated high accuracy for safety-critical tasks (e.g., medication, fall prevention) but showed subjectivity in non-urgent areas.
• Simulated caregiver feedback and input from members of the Engagement of People with Lived Experience of Dementia (EPLED) were crucial for system adaptation and evaluating stress reduction.

Conclusions:

• Integration of generative AI into dementia care is feasible, enhancing task verification and decision support.
• Contextual data, caregiver input, and lived experience perspectives are vital for optimizing AI system performance.
• AI-assisted verification holds potential for reducing caregiver stress and improving support for PLwD, warranting further real-world validation.