FULL STORY

Economist Novosad's Refusal to Review AI-Written Papers

Economist Paul Novosad first called for zero tolerance of AI-generated academic writing in September, then announced he would refuse to review papers that read like AI. The debate over journal policy continues.

2026-09-24 ~ 2026-10-05 · 2 episodes · 19 posts

Episode 1 · Economist Calls for Zero Tolerance of AI Writing in Journals, Sparking Academic Debate (2026-09-24, 12 posts)

On September 24, development economist Paul Novosad, writing as "a disillusioned reviewer," posted a series of threads arguing that academic journals should adopt a zero-tolerance policy toward AI-generated writing, sparking wide discussion in academia. His concrete proposal is to screen submissions with AI-detection tools such as Pangram and ban violators from submitting for two years; philosopher Seth Lazar endorsed the idea and proposed even harsher penalties. In exchanges with critics, Novosad clarified that his target is the scale of AI slop, not ghostwriting itself.

Confirmed

  • Novosad's core claim is "writing is thinking": journals have every reason to require that at least one scholar has genuinely thought deeply about the topic and written several pages in their own hand.
  • Concretely, he proposes screening submissions with detection tools like Pangram, with a two-year submission ban for violators.
  • Seth Lazar proposed stricter rules: if a paper is found to be substantially AI-written, the submission is rejected, published papers are retracted, and the author is permanently banned from the journal.
  • He stressed that the new problem of the AI era is volume: large volumes of low-quality submissions require substantial human reviewer time to identify as junk—something that never happened in the era of human assistants. He added that in the social sciences, writing itself is the thinking process, and the understanding generated while writing has value worth sharing, possibly unlike in STEM fields.
  • He acknowledged some defensible uses of AI but argued these exceptions will be drowned by a flood of low-quality AI-generated text, demoralizing reviewers and ultimately killing journals—a "bad money drives out good" dynamic.
  • In response to Franco Calabres's point that human ghostwriting and editing for senior scientists was already common 40 years ago, Novosad replied that his objection is not to ghostwriting per se but to the massive scale of AI slop.
  • He further argued that even if AI generating and evaluating economics research has value, once it mixes into the existing journal system under a "don't ask, don't tell" equilibrium about AI involvement, peer review and the journal system will collapse; AI-generated research needs a separate publishing venue.
  • In the discussion, ipeirotis offered an alternative: low-quality papers have always existed, AI will not go away, and institutions should adapt rather than block. He said that if "technical correctness" could be handled first in review (e.g., AI reviewers filtering technical errors), he would gladly review only technically correct papers and focus on judging interest; for papers written with AI but with conclusions that hold up, he does not care about AI involvement—implying skepticism about the reliability of tools like Pangram. lpachter also joined via a repost commenting "Kill journals?"

Why it matters

  • "Bad money drives out good" is the mechanism at the core of the argument: without restrictions, diligent authors competing with AI-written slop will gradually exit, reviewer morale erodes, and journal quality declines systematically.
  • The exchange with critics moved the debate from "can AI be detected" to "what exactly is being banned—ghostwriting or volume," with Novosad's answer showing his target is the scaled flood of AI-generated content.
  • The position touches on the survival preconditions of peer review—whether journals may require authors to prove hands-on writing—and proposes bifurcating academic publishing in the AI era (human-written papers and AI-generated research in separate venues), while ipeirotis's response represents the "fight AI with AI" institutional-reform route; the tension between these two paths will continue to shape academic publishing.

Episode 2 · Economist Novosad Says He Will Refuse to Review GenAI Papers, Calls for Separating AI and Human Writing (2026-10-03, 7 posts)

Paul Novosad, an economics professor at Dartmouth, publicly declared that he will no longer review papers that read like they were written by AI, after accepting a review invitation from a top journal in his field and finding the submission疑似 AI-generated. He followed up with a lengthy thread of tweets laying out his critique of GenAI papers, with the core argument that AI writing and human writing should be kept in separate spaces. Alex Tabarrok echoed his views.

Confirmed

  • On October 3, Novosad said he had received a review invitation from a top journal, found the paper疑似 AI-generated, and announced he would reject any paper that reads like AI wrote it; he noted that many people in the comments actually enjoy reading AI-generated text, but said he can't stand it.
  • In a long thread on October 5, Novosad argued that academic success comes from two things—doing great, important work, and/or packaging that work in a way that maximizes success; some academic "superstars" are clearly far stronger at packaging than at research quality itself.
  • He criticized that most GenAI papers don't start from an author with a good idea to express; instead, lacking good ideas, they are dressed up in overblown academic jargon and an "AI-style precision" prose to make them look like good ideas, calling reviewing GenAI papers the most frustrating task and vowing to refuse such reviews going forward.
  • He compared GenAI papers to the critical-theory paper at the heart of the 1996 Sokal affair: dense in rhetoric and jargon, hollow at the core; AI happens to be better at packaging than substance.
  • He argued that AI writing and human writing cannot coexist in the same space—wherever AI writing is allowed, it drives out human writing—so a dedicated, segregated space should be created for AI writing to develop, while keeping AI writing out of spaces for human writing.
  • In discussion with Tabarrok, Novosad went further: even if economics research were eventually done entirely by AI, its best form might still be in AI's own format, kept separate from the space where human scholars write.
  • Tabarrok responded that the problem with writing in the AI era is not good-idea authors using AI to express themselves, but that most authors lack good ideas and merely use AI-style, over-precise terminology and prose to package mediocre ideas as good ones.

Why it matters

  • The debate touches on foundational rules for academic publishing in the GenAI era: whether reviewers should reject papers suspected of being AI-ghostwritten, and how academia defines the boundary between human and AI writing.
  • Novosad's "separate spaces" proposal offers an alternative governance approach to AI's role in research—not banning AI research itself, but separating it from the space of human academic writing, which could shape future journal policies and debates over academic norms.