Generation Gap or Diffusion Trap? How Age Affects the Detection of Personalized AI-Generated Images

René Lüdemann, Alexander Schulz, Ulrike Kuhl

Abstract: AI-generated content, particularly artificially generated image data, presents a fascinating challenge as it blurs the line between reality and fabrication in digital media. This study explores the ability of different age groups to distinguish between real and AI-generated images, using a custom model based on Stable Diffusion to create personalized synthetic images of the same individual. Participants (N = 112) showed an overall accuracy of 87.01%, with younger individuals outperforming older ones in both detection accuracy and confidence. Older participants also took longer to make their decisions, indicating either an age-related decline in processing speed, or a more careful and deliberate analysis of the presented content. These findings highlight the importance of improving digital literacy across age groups and developing robust detection tools to better equip users to navigate an increasingly AI-driven digital landscape. In the interest of reproducible research, the entire code, images, and data is available at: https://github.com/ukuhl/GenGapCHIRA2024.