Evolving the narrative utilization ecosystem with life story interpretation and generative AI tools

Open Access
Article
Conference Proceedings
Authors: Masayuki IharaHiroko TokunagaTomomi NakashimaHiroki GotoYuuki UmezakiYoko EgawaShinya HisanoTakashi MinatoYutaka NakamuraShinpei Saruwatari

Abstract: In nursing care, providing high-quality, individualized care requires a deep understanding of each care recipient’s narratives (Guendouzi et al. 2015). These narratives are rarely recorded or organized for practical use, making it challenging for novice care workers to effectively incorporate them into caregiving. To address these issues, we propose a “narrative utilization ecosystem” aimed at improving the quality of nursing care services (Ihara et al. 2025). This paper outlines the proposed ecosystem and presents case studies for collecting and analyzing narrative fragments by revisiting familiar locations from the care recipient’s past. Additionally, we define the requirements for prompt engineering of a generative AI-based care advisory tool that provides advice grounded in person-centered care principles (Kitwood et al. 1992).The proposed narrative utilization ecosystem consists of three core components: collecting narratives, analyzing them, and applying them in caregiving settings. Narrative fragments are often gathered during initial assessments when a care recipient begins using nursing services. Daily conversations are another common method for collecting narratives. For individuals with dementia, opportunities to recall and share old memories are especially valuable. To facilitate this, we conducted case studies in which an individual with dementia walked with an accompanying care worker around familiar locations, such as former homes, schools or playgrounds. During these walks, we recorded their conversations and the scenery they observed. In one-hour walking sessions, the care recipient spoke for approximately 10 minutes. Out of 165 recorded utterances, 84 were brief responses like “Yeah, that’s right,” and 33 were expressions of uncertainty, such as “I don’t know,” referring to directions or places. However, we also gathered meaningful narrative fragments, including 11 statements related to location recognition (e.g., “There was company housing beyond here”) and 9 observations about vacant homes (e.g., “No one lives there anymore”). This approach encouraged storytelling by prompting the care recipient with questions linked to shared visual cues, enabling the construction of richer narratives from the collected fragments. Analysis revealed that asking questions to reaffirm the past, rather than replaying it, was particularly effective in eliciting narratives from individuals with dementia.When designing generative AI tools for handling narratives, it is essential to prioritize the emotional sensitivity of individuals with dementia and minimize burden on care workers. Through discussions with dementia care and AI experts, we identified key design requirements, including the ability to adapt to the recipient’s daily mood and fatigue levels, considering the current rapport between care workers and recipients, and avoiding both labeling care recipients with dementia and blaming failures in care on dementia.Our future work will focus on developing AI tools that meet these requirements and testing them in real-world care settings. We also plan to continue collecting and structuring additional narrative fragments to enhance the ecosystem's effectiveness.Guendouzi, J. et al. (2015). Expanding Expectations for Narrative Styles in the Context of Dementia. Topics in Language Disorders. 35. 237-257. 10.1097/TLD.0000000000000061.Ihara, M. et al. (2025). Narrative Utilization Ecosystem for Person-Centered Care. Proc. of Intelligent Human Systems Integration: Integrating People and Intelligent Systems (IHSI 2025). (to appear)Kitwood, T. et al. (1992). Towards a theory of dementia care: Personhood and well-being, Ageing and Society, 12(3), 269-287.

Keywords: narrative, dementia, ecosystem, nursing care, generative AI

DOI: 10.54941/ahfe1006293

Cite this paper:

Downloads
19
Visits
75
Download