发表机构
Stellenbosch University（斯泰伦博斯大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
探讨人工智能辅助研究对科学认证机构的影响，指出其改变了成本结构，影响了认证意愿和费用，还研究了固定审查能力等情况下的认证变化，将人工智能与科学生产经济学和相关机制相联系，转移了稀缺性关注点。
AI 中文摘要
本文研究了成本更低的人工智能辅助研究如何改变科学认证机构。人工智能降低生成完善手稿的成本比判断潜在贡献是否有价值的成本更快。因此，完善性失去信息，进入门槛扩大，未经认证的研究平均质量可能下降。在固定标准下，由于外部选择变差，对可靠认证的支付意愿上升。具有市场支配力的认证者可以获取这一溢价；竞争和替代披露规则可能产生不同费用。在固定审查能力和弱承诺情况下，认证会稀释。在此扩展中，对提交的部分庇古税和审查能力的影子价值都随人工智能辅助进入而上升。贡献在于将人工智能与科学生产的经济学与信号传递、认证和同行评审联系起来：成本更低的生产将稀缺性转移到下游，从使研究看起来可信到验证哪些研究可信。
英文摘要
Journals cannot referee every paper they receive. When submissions outrun the time reviewers can give, an editor must decide what to do with the papers no one reads: reject them, or check them with a faster but less accurate tool, of the kind artificial intelligence now supplies. A paper the tool passes carries the same public stamp as one an expert approved, and readers cannot tell a screened paper from a reviewed one. This paper asks which choice makes the stamp more valuable. Rejecting the unread papers keeps the approved pool cleaner, because the faster tool sometimes passes weak work. Yet rejection can make the stamp worth less: good papers turned away unread join the pool of unstamped papers and raise the quality against which a stamped paper is judged. Which effect wins depends on how common good work is, and the model gives the exact switching point when there is one. The choice also shifts how much a strong affiliation is worth; and once authors can run the screening tool themselves, wherever some game it and some do not, the stamp's value is set by what getting past the screen costs, not by what the screen detects, while expert judgment allows no such gaming.
Comments30 pages, 2 figures, JEL codes D82, D83, I23, L15, O33 AI use disclosed in the paper
