Proceedings of the Annual Meeting of the Cognitive Science Society

Human memory exhibits vulnerability in cognitive tasks; comparing visual working memory with new perceptual input can cause unintended distortions. Prior studies report systematic memory distortions post-comparison, but understanding their impact on real-world objects and identifying contributing visual features remains challenging. We propose an AI-driven framework generating naturalistic stimuli based on behavioral object dimensions to elicit similarity-induced memory biases. Using two stimuli types—image wheels (dimension-edited) and dimension wheels (activation-based)—we conducted three visual working memory experiments under conditions: no perceptual comparison, image wheel comparison, and dimension wheel comparison. Results show that both similar images and dimensions induce memory distortions. Visual dimensions (e.g., shape/texture) are more distortion-prone than semantic ones (e.g., category), indicating that naturalistic stimuli's object dimensions critically influence memory vulnerability.