From Production to Selection: How Academic Gatekeeping Misreads the AI Era

Academic debates about AI remain too focused on whether models can produce publishable prose. This paper argues that the more consequential shift is not writing alone, but the large-scale cheapening of weak-but-reviewable academic output. Under these conditions, the core bottleneck moves from production to selection. The paper examines how institutions respond differently to this pressure through entry tightening, manageable filtering, and compliance enforcement, using arXiv, IAMCR, and ICML as illustrative governance moves rather than isolated controversies. It argues that current responses often misdiagnose the source of overload by stretching recognizability, affiliation, and compliance into proxies for contribution and trustworthiness. The result is not only the sorting of texts, but the sorting of participants by their ability to endure rejection, delay, and repeated proof demands. The paper concludes by distinguishing first-stage filtering from exclusion and argues for a layered model in which AI handles obvious structural failure while human judgment is reserved for borderline, unusual, and potentially important cases.