When
Where
Text-to-Image AI and User Generated Social Media Content
Abstract: Text-to-image AI tools enable social media users to transform their imaginative ideas into visual content in a wide range of visual styles, adding a new source of novelty for social media platforms. We argue that the AI tools also pose a spillover effect on the novelty of user-generated regular content on social media (i.e., content generated without an AI tool) because using AI tools could potentially change the reputation dynamics on social media platforms (i.e., reputation mechanism) and/or fundamentally reshape human cognitive processes and innovative thinking (i.e., AI-usage mechanism). Collaborating with an image-sharing social media platform, we employ a difference-in-differences (DID) approach to empirically demonstrate that using a text-to-image AI tool significantly reduces the novelty of user-generated regular images. That is, after using the text-to-image AI tool, AI users’ regular images become more similar to the platform-wide recent images than those of non-AI users. We do not find evidence supporting the reputation mechanism. However, empirical evidence consistently validates the AI-usage mechanism. Specifically, novelty reduction worsens with more frequent AI usage, and frequent AI generation induces a fixation effect evidenced by users’ conformity to AI, i.e., regular images from AI users become more similar to the AI-generated images than those from non-AI users. Delving into more granular image-level generation records, we conduct an instrumental variable analysis and find that repeatedly generating AI images in a specific style reinforces user perception of a strong prototype and intensifies the fixation effect, leading to greater conformity of their regular images to that AI style. Finally, prompt input plays a key role in shaping the AI-induced fixation effect on the novelty of regular content. AI users who consistently input prompts with broader informational scope and greater semantic variation are less subject to novelty reduction of regular content.
Bio: Jingchuan Pu is an Associate Professor of Information Systems at the University of Florida, where he also received his Ph.D. in Information Systems. Prior to returning to his alma mater, Jingchuan was an Assistant Professor at the Pennsylvania State University for three years. Jingchuan’s research focuses on Platform Economy (e.g., Social Media, E-Commerce), as well as AI and Human. His research has frequently appeared in leading journals and been funded from various agencies (e.g., NSF). Jingchuan currently serves as senior editor at Production and Operations Management. He is also a member of the Editorial Review Board at Information Systems Research.