Memory Recall Support through Externalization of Remembered Scenes Using Generative AI

Evaluating the Effects of Iterative Image Modification Process

Abstract

This study proposes an approach to memory recall support through the visualization of remembered scenesusing generative AI. Previous memory recall support systems had limitations, requiring existing records such asphotographs or other media as triggers for recollection. To address memories without such records, our approachutilizes generative AI to transform verbal descriptions of memories into visual representations, enabling users toiteratively refine these visualizations through feedback. We introduce a process that combines two key elements: aniterative description of memories and detailed modification of generated images, focusing on the differences betweengenerated images and recalled scenes to enhance detailed recollection. To evaluate the effectiveness of these elements,we conducted experiments with three conditions examining their effects on recall quality: single description, iterativedescription, and iterative description with detailed modification. While no significant differences were observed inobjective recall quantity measures, subjective evaluations indicated that participants in the condition incorporatingboth iterative description and detailed modification reported a significantly stronger sense of recalling memories ingreater detail and with higher accuracy compared to the single description condition, suggesting the effectiveness ofour proposed method. This research demonstrates the potential of interactively visualizing remembered scenes usinggenerative AI as a new approach to memory recall support.