Survey Experiment on News Sharing Behavior Generated by AI

Abstract

With the increase in malicious content created by generative AI, the reliability of information on social networks has decreased, while control over the spread of such content has become significantly more difficult. Among these malicious generative AI contents, many are being systematically and extensively spread as hoaxes or propaganda, causing serious social unrest. Therefore, the proposal of prompt methods for controlling this spread is desired. In this study, we quantitatively investigated which types of images in pseudo-news created by AI are more likely to be shared. The results showed that in political pseudo-news, images that evoke negative emotions have a significant impact on their sharing and spreading. In contrast, for pseudo-news related to entertainment, the emotional expression of the images does not affect sharing. Furthermore, we conducted an experiment by replacing the original image in the same pseudo-news with one of lower emotional value, revealing that this manipulation could potentially reduce the willingness to spread news generated by AI in both politics and entertainment. These research results form an important basis for social network operators to perform content moderation in the era of generative AI.