For generations of aspiring academics, looking intelligent required considerable effort. A carefully written personal statement, polished research proposal, and a considerable research output in strong academic publications communicated quality. These criteria suggested diligence and industry, intellectual ability, and most importantly, a capacity to contribute something meaningful to established academia. Producing these signals took years and considerable effort. This high cost is what made them credible to reviewers, and is why academia has been able to continue; those who seek careers in academia are willing to pay the cost to prove their worth.
The newly–prevalent use of artificial intelligence in academic writing, however, changes this calculation for the reviewer. Students use AI to turn applications that once would be outright rejected into candidates for further consideration; researchers produce literature reviews and write drafts in hours or days instead of weeks. This technology has not destroyed the value of original research. Rather, it has made the appearance of competence considerably cheaper.
The economics of looking intelligent
Academia, like many markets, operates with significant signalling and screening. Signalling occurs when someone uses an observable action to communicate information about a quality that others cannot directly see. In his 1973 paper, “Job Market Signaling”, Michael Spence showed how education could serve this purpose. Employers cannot immediately observe a candidate’s productivity, so applicants use qualifications to indicate their underlying ability. The signal is credible largely because obtaining it requires effort; it has a cost. The willingness to bear this cost is a measure of an applicant’s dedication to the field.
Academic life is full of similar exchanges. Applicants signal their potential through essays, research proposals and extracurricular achievements. Researchers signal their competence through publications, conference presentations and carefully constructed arguments. Universities and journals, meanwhile, act as screeners: they use these visible indicators to distinguish promising candidates and credible research from weaker alternatives.
AI upsets both sides of this exchange. If a polished paper once required considerable writing ability, but can now be produced effortlessly with a few prompts, its relationship with the applicant’s underlying ability becomes far weaker; it is now a poor signal of competence. The same applies to academic prose. A well-written paper may still reflect excellent research, but its fluency and grammatical accuracy is no longer reliable evidence of the author’s ability to write well, how deeply they understand the material, or even willingness to meticulously proofread.
A market for academic lemons
The fundamental problem created by the new prevalence of AI in academia is the same as that in George Akerlof’s seminal 1970 paper, “The Market for Lemons.” Akerlof modeled the fundamental case of seller-heavy asymmetric information: sellers know more about the quality of what they offer than buyers do, so buyers must use imperfect methods of screening to separate the peaches (high-quality, desirable candidates) from the lemons (poor-quality, but difficult to distinguish alternatives). His original examples included used cars, insurance, employment and credit markets in developing countries, where uncertainty about reliability can damage transactions.
Let us take this academic example: buyers (reviewers) struggle to reliably distinguish a peach (high-quality applicant) from a lemon (applicant that has the appearance of competence due to AI, but that is actually incompetent). Previously, academia represented a stable lemon market with a separating equilibrium—sellers can signal reliably, and buyers can use these signals to distinguish lemons from peaches. However, with the flood of lemons caused by AI, academia has been forced into an unstable and suboptimal pooling equilibrium that forces sellers to treat all buyers as an average of the lemons and peaches in the market. When the market is full of lemons, that average is rather poor.
In Akerlof’s model, a common result of this pooling is adverse selection. When buyers cannot tell peaches from lemons, the price they offer reflects an average quality based on the estimated breakdown of lemons and peaches in the market. In academia, adverse selection causes buyers to institute submission caps, lowering a researcher’s expected value of submitting to a high-quality journal. This bottleneck leads peaches to exit the academic market, choosing to settle for lower-tier journals that fail to develop them as academics as well as top-quality publications do.
Alisher Aldashev’s paper from January of this year finds a new, worrying optimal move for sellers. The average academic, faced with the necessary increased scrutiny caused by a flood of artificially intelligent lemons, exerts minimal effort. Pal and Park explain this new optimality: high-potential applicants that may previously have been peaches end up using AI at comparable rates to low-potential lemons. Since high-potential applicants will produce many papers regardless of whether they use AI, the fact that it is cheaper to produce AI-written papers leads former peaches to do so. This decreases the quality of the average academic; reliable peaches struggle to hone their signalling skills in a rigorous market, while unreliable lemons increase in number and slip through the net. Aldashev notes a worrying possibility: even as evaluation policies adapt, the shock of an AI-driven deviation from academic hysteresis—a flood of lemons—could mean that the drop in average academic quality will prevent a separating equilibrium from returning.
Will academia collapse?
In short: I don’t think so. While Aldashev’s concern is valid, the incentives to participate remain powerful. Researchers seek publication for career progression, universities value prestige and applicants continue to compete for an ever-scarce number of places. The rewards of academic participation have not disappeared simply because its signals have become less reliable. As the market adjusts, we will continue to experience an excess supply of PhD graduates and difficulties in screening journal submissions until the market becomes rigorous enough to return to reliable screening.
Bringing back costly signals
How should universities and journals adapt, then? If polished prose is now cheap, greater weight may be placed on evidence that is harder to fake: a candidate's ability to explain their research, defend a methodological choice or respond to an unexpected question. We may see an increased reliance on defenses of applications in the academic process. Technical interviews, like those already in place for finance- and computer science-heavy fields, may rise as well. Reviewers may continue to impose limits on submissions to stem the flow of lemons. The market will adjust, given some time.
Can a typo save your career?
Probably not. While the market adjusts, however, many reviewers treat typos and basic grammatical mistakes—relatively minor and common issues that do not necessarily disqualify the content of one’s work—as signals of authenticity. Until review methods learn to filter out AI and remove the lemons, a tiny typo could be the distinguisher between you (hopefully a peach), and the flood of lemons you compete against.
Cover image from Red Maple Economics Group.
Rahul Sundaresan is Deputy Editor-in-Chief of the Consilium and a first-year Economics and Politics student. He writes weekly, focusing on political economy, international economics, and topics in game theory.
