Zaynab Rehman^1,2 (), Julia Thomas³, Wilma A. Bainbridge¹; ¹University of Chicago, ²University of Illinois Urbana-Champaign, ³Cardiff University

When reading a story, one often conjures up mental images of the scenes and events unfolding in the text. Mental imagery generated by individuals after reading a given text may differ depending on which experiences and biases that individual is drawing from. How similar are people in what they visualize, and is there a way to access these mental images by using generative artificial intelligence methods? To test this question, we presented participants (N = 124) with brief synopses of news stories and asked them to mentally visualize what they had read. Participants were then presented with four images (the “ground-truth” image used in the original news article, as well as three randomly intermixed AI-generated images) and rated how well these images matched their own personal mental imagery by providing a concordance score. The participants also described what specific details about the images led to higher or lower concordance with their own mental imagery. Of the 20 articles selected, participants gave the AI-generated images higher concordance scores than the ground-truth image used by the news article 80% of the time. This suggests AI better matches mental visualizations than the images chosen by news editorial teams and that AI platforms may serve as a powerful tool for uncovering these mental images. Consistent patterns in concordance were also observed across participants for the news articles, indicating frequent agreement in which images matched participant mental visualization the most and which matched the least. We are now assessing whether it is possible to improve concordance scores across participants by re-generating the images according to participants’ qualitative comments about the AI-generated photographs. Leveraging AI-generated images and ratings of concordance could present a novel way of extracting information regarding mental imagery, memories, or individual biases.