Development of a Generative AI-Assisted Digital Language-Cognitive Rehabilitation Content Prototype for Older Adults with Mild Dementia

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

This study aimed to develop a generative AI-assisted, web-based digital language-cognitive rehabilitation content prototype for older adults with mild dementia and to describe its development process, structure, and initial revisions. ChatGPT was not used as an intervention system that automatically analyzed user performance or adjusted task difficulty in real time. Instead, it served as a development support tool for generating initial drafts of linguistic stimuli, task items, instructions, and feedback messages. The researcher selected and revised the generated outputs according to clinical appropriateness, linguistic difficulty, cultural familiarity, potential emotional burden, and feasibility of screen-based implementation. The development process comprised six stages: data collection and needs analysis, design of content domains and modules, draft generation using ChatGPT and clinical revision by the researcher, implementation of a web-based prototype, expert content-validity review, and pilot application with five older adults with Long-Term Care Grade 4 attending a day care center followed by final refinement. The prototype consisted of seven modules: word recognition, active-passive verb comprehension, basic verb production, story retelling, rhythm- and melody-supported verb and sentence production, superordinate-category vocabulary, and Semantic Feature Analysis. Each module contained 15 tasks, resulting in 105 tasks. The final product was an interactive browser-based prototype implemented using HTML, CSS, and JavaScript, integrating text, images, voice guidance, mouse clicks, and touch responses. Task difficulty was not adjusted automatically by AI; instead, an operator selected modules, number of choices, stimulus presentation time, repetition frequency, and cue levels based on the user’s performance and response characteristics. This development study presents a clinician-generative AI collaborative content-development process and an operator-controlled modular application model. Clinical effectiveness and objective usability remain to be evaluated in future research.

Links and sources

Comments