Advances in Generative AI and Platform Moderation: Implications for Online Knowledge Sharing

Abstract

The release of ChatGPT demonstrated generative artificial intelligence (GenAI)'s potential to generate human-like content, transforming knowledge-based sectors such as content creation, education, and programming. While extant research has focused on the early impact of GenAI on knowledge sharing, there are gaps in our understanding of the role of improving AI (GPT-4 class LLMs) and the platform's AI moderation policy. Using post-GPT-4 release data from Stack Overflow, this study takes advantage of a large natural experiment (moderators' strike) to empirically address these gaps using a difference-indifferences approach with structural breaks. We find that while GPT-4's release further reduced user contributions, it improved overall content quality, unlike GPT-3.5, especially among less established users. Also, the replacement of Stack Overflow's initial strict AI-ban with a lenient moderation approach due to the moderators' strike, increased user participation but compromised content quality. Surprisingly, the quality of questions was preserved even with lenient moderation, and was associated with increased answers from more established users. Our findings suggest that AI quality and platform moderation policies interact in significant ways to reshape online knowledge sharing.

Keywords: generative AI, content moderation, knowledge sharing, online platforms