LLM-Driven Daily Conversation for Cognitive Health in Older Adults

Cognitive decline is an important health concern in aging populations. Early cognitivestimulation that is meaningful and related to daily life may help older adults maintaincognitive and psychosocial function. However, many digital cognitive programs userepetitive exercises and provide limited opportunities for natural social interaction.This study aims to develop and evaluate a large language model‐driven daily conversationcognitive interaction platform for older adults at high risk of dementia. The platformuses voice‐based, semi‐structured conversations based on topics that interest eachparticipant. The conversations are designed to stimulate five cognitive areas: memory,attention, language, executive function, and social cognition.During each session, participants speak with the platform using natural conversation. Thesystem converts speech into text and analyzes selected features of the participant'sresponses, such as response accuracy, language fluency, organization of ideas, attentionto the conversation, and understanding of social situations. The system then generates acognitive profile report. This report is intended to provide understandable informationabout performance across the five cognitive areas and is not intended to diagnosedementia.The study will be conducted in two stages.Stage 1 will evaluate the initial feasibility and usability of the platform. Fifteenparticipants will be enrolled, including five healthy younger adults, five older adultswith normal cognitive function, and five older adults with cognitive difficulties. Eachparticipant will complete two 20‐minute conversation sessions. Participants will thencomplete questionnaires about system usability and intervention acceptability. Theircomments and experiences will be used to improve the platform before the randomizedtrial.Stage 2 will use a single‐blind, randomized controlled design. A total of 60 older adultsat high risk of dementia will be randomly assigned in equal numbers to one of threegroups:Large language model conversation group: Participants will complete voice‐based cognitiveconversations with the platform. The activities will include exercises targeting memory,attention, language, executive function, and social cognition.Interest‐based news reading group: Participants will read news content selected accordingto their interests and complete questions related to the content.Waitlist control group: Participants will receive weekly health education materialsduring the initial study period.The interventions will continue for 12 weeks. Cognitive and psychosocial outcomes will beassessed before and after the intervention. The cognitive assessments will examineoverall cognitive function and specific abilities such as memory, attention, language,processing speed, executive function, and social cognition. Psychosocial assessments willexamine depressive symptoms, loneliness, and social connection.The study will also evaluate intervention completion, system usability, and interventionacceptability. The findings will be used to determine whether the platform is feasiblefor home‐based use and whether it shows preliminary benefits for cognitive andpsychosocial health in older adults at high risk of dementia.