Academic researchers are already submitting AI-generated manuscripts for publication, sometimes without acknowledging the role of the AI. This raises a host of moral and practical questions, including questions about the overall moral status of using AI for any purpose and about whether such submissions constitute research misconduct. However, relatively little attention has been devoted to analyzing and predicting the large-scale effects, on the academic literature, of a scholarly paradigm in which many or most manuscripts are AI-generated. In this paper, I argue that at least in the case of the humanities, pervasive AI-generated articles will likely produce a kind of systemic epistemic degradation or knowledge-base erosion of an institution or system. Briefly put, the resulting academic literature will become even more dominated by articles that are safe, technical, and insular, thereby crowding out a valuable form of radically innovative scholarship. These problems will occur because of the current incentives in the academic-publishing system in the humanities and the near-future capabilities of LLMs. I conclude by suggesting a few ways by which to mitigate these impending problems.
Thomas Metcalf (2026) studied this question.