Listening to Voices in AI: Ethical and Implementation Frameworks for Large Language Models (LLM) in Dementia Care and Research

Rachel Sava, PhD – Program Director and Instructor at Harvard Medical School, McLean Institute for Technology in Psychiatry, McLean Hospital, Mass General Brigham; Benjamin Silverman, MD – Senior IRB Chair and Assistant Professor at Harvard Medical School, Mass General Brigham; Ipsit Vahia, MD – Chief of the Division of Geriatric Psychiatry and Corrigan-Minehan Endowed Chair in Psychiatry at Harvard Medical School, McLean Hospital, Mass General Brigham

Abstract: As the global population ages, a critical shortage of geriatric and dementia care professionals has emerged. Large language models (LLMs) and AI-powered tools offer promising solutions to address this shortage, including extending the reach of professionals, facilitating caregiving, and aiding with tasks from medication tracking to behavioral management. Yet, their integration raises complex ethical, legal, and practical challenges - particularly around informed consent, data stewardship, and decision-making authority. Unlike existing AI ethics research, which often focuses on clinicians or policymakers, this study centers caregivers, primary decision-makers for persons with dementia (PWD), through scenario-based qualitative interviews and surveys to empirically explore these topics. This presentation incorporates qualitative research findings and addresses critical questions including: What constitutes meaningful informed consent for AI use in dementia care and clinical research? What do patients and their caregivers need and want to know? How do caregivers as surrogate decision-makers make decisions about AI use in the treatment of PWD? What are their preferences for AI-driven care? Through our interviews, we examine how caregivers engage with AI tools in real-world settings, highlighting the ethical and practical questions that arise with using AI in clinical and research environments. Research findings are incorporated into the development of a novel ethical framework addressing informed consent, accountability, and data stewardship, amongst many other areas, to guide the responsible implementation of AI in dementia research and clinical care. By centering on caregiver voices, the proposed implementation framework enhances patient-centered care and ensures AI deployment aligns with the realities of dementia caregiving.

Keywords: Artificial Intelligence (AI) and Large Language Models (LLMs), Informed Consent, Dementia Care