Leiter Reports: A Philosophy Blog

News and views about philosophy, the academic profession, academic freedom, intellectual culture, and other topics. The world’s most popular philosophy blog, since 2003.

  1. TW's avatar
  2. Cian Dorr's avatar
  3. Charles Anthony Bakker's avatar

    It is worth noting that prospective graduate students, like undergraduate students, cannot take their own judgment. In submitting anything, whether…

  4. John Pillette's avatar
  5. Brian Leiter's avatar
  6. John Rapko's avatar
  7. Duane Long's avatar

Graduate school admissions in the age of AI

Brian Leiter

This is going to become a very serious issue. PhD and MA admissions depend very importantly on the writing sample. What do programs do when they later come to suspect the admitted student used AI to produce the writing sample? I think all graduate programs need to adopt an absolutely draconian rule, namely, automatic expulsion if it turns out the writing sample was written with AI. Of course there should be due process: evidence should be presented to the student, who should have an opportunity to explain or rebut the evidence. But absent such a rule, I would expect this to become a recurring problem.

I have heard from some journal editors that they are also increasingly getting articles they suspect are AI generated in significant part (or entirely). So journals will have to think about imposing similar rules (e.g., a lifetime ban of the author from submitting?).

What do readers think? Other proposals welcome.

, , ,

Leave a Reply to Cian Dorr Cancel reply

Your email address will not be published. Required fields are marked *

11 responses to “Graduate school admissions in the age of AI”

  1. early career scholar

    Things might have changed since I last checked, but out of curiosity, about a year ago, I fed various bits of my writing, including portions of chapters and article abstracts, into some AI-detection software. It consistently told me that my human writing was AI-generated — sometimes stating the likelihood of AI-generation was as high as 97-99%.

    The problem is that it’s difficult to tease out human-generated writing from AI-generated writing, especially if one is not already familiar with the style and writing habits of the human writer in question.

    Also, I was recently on a selection committee for a travel grant award, and I was certain one of the applications was AI-generated. (It had several tells that I’ve picked up from teaching undergraduates, including excessive dashes and an annoying tendency to list things in threes.) Ultimately, however, the committee rejected the application for other reasons.

    Reply
    1. John Rapko

      If the annoying tendency to list things in threes is a tell-tale sign, then surely Hegel’s Encyclopedia Logic and much of Charles Sanders Peirce was written by AI.

      Reply
      1. Brian Leiter

        Surely no LLM writes as badly as Hegel!

  2. Mark Taylor

    There are no good arguments, in an educational environment, for distinguishing between student submissions which are AI-generated and those which are plagiarized with the author’s permission. Bans on AI-generated application materials should therefore be a simple extension of our current policies against plagiarism. Further, although graduate writing samples should not be allowed to use AI, we should impress upon them that using AI is unlikely to advantage them anyway. In the MIT preprint study that made the rounds earlier this year (which, admittedly, still hasn’t finished peer-review), they remark: “We found that the Brain-only group exhibited strong variability in how participants approached essay writing across most topics. In contrast, the LLM group produced statistically homogeneous essays within each topic, showing significantly less deviation compared to the other groups” (Kosmyna et al. 2025). Graduate applicants already struggle to distinguish themselves among a crowd; using AI will actively make them sound more average and uninteresting.

    Regarding journals, I have recently been a referee for a couple of articles which were obviously AI-generated. They take a lot more time to review. The intellectual mistakes are well-hidden because they do not have the same red flags accompanying those human mistakes have, and they are smothered in a tone of confident professionalism. I would be unlikely to accept a referee request from a journal which allows authors to use AI if for no other reason than that it consumes too much time for something that is very unlikely to contribute anything to the discipline.

    Reply
  3. s. wallerstein

    When I entered graduate school in literature in 1969, we had to sit down with a dictionary and a pen and translate from two languages into English within a determined period of time. I passed so I have no idea what occurred if one did not pass.

    Why not require entering graduate students to sit down pen in hand (to avoid cheating) and require them to write a page or two about a subject, which is not known before-hand?

    Reply
    1. Duane Long

      I don’t think we’d want to do just any subject in philosophy for such a test, as many incoming undergrads may not have exposure to things faculty might think of as “fundamental things every philosophy student should know”. However, we might ask them to write two essays. For one, it’s on the topic their writing sample was on, and the other, they get a list of, say, 8 topics, of which they have to pick 4, from which the examiners then pick one. Any one topic might have been missed in undergrad, but if they can’t write competently on at least 4 of 8 basic topics, they probably aren’t a good pick for the program anyway.

      Reply
  4. Amy Flowerree

    Sometime DGS of a terminal MA program here.
    This is something we have been thinking very carefully about. So far our only actionable item is to ask applicants to certify that their work is their own (and any use of gen ai or similar means the work is not their own). Lying on the application materials is grounds for withdrawing acceptance.

    I think the Draconian Proposal needs to interact with the nuance of what it means to “use” AI. Students use it to summarize papers, they put text in to restructure the paper, they use grammarly which suggests rewrites for “clarity” (though this can end up being substantive). One could imagine a student who was told how to “ethically” use these programs in the writing process, then what can they submit as a writing sample? GenAI is becoming more the fabric of education, its role becomes invisible to students. Very soon, we will have students that have relied on these technologies since high school.

    Now, an entire writing sample from genAI? That you can clearly say is plagiarism. But what about someone who uses various genAI tools to shape the project? I myself am very skeptical of the utility for students. I think they need to go through the hard work of understanding difficult texts. Using genAI as a tool, one could seemingly display a grasp of difficult philosophical issues, and then arrive at grad school unprepared.

    We haven’t implemented this yet, but in our department we have discussed interviews as part of the admission process. Obviously, some people don’t interview well. But the goal would be to see if they genuinely understand what they wrote in the writing sample.

    Reply
  5. John Pillette

    Grad students are aspiring professionals and so in California his sort of thing is inarguably within the ambit of California’s Unfair Competition Law (Bus. & Prof. Code s17200). The UCL is famous for its use of “predicate” violations—i.e., a violation of any other state or federal rule or regulation however minor … did you fart and then blame it on the dog? Prob a UCL violation—but the UCL was first set up to address exactly this sort of unfair competition. If a grad student cheats by using AI he or she is engaging in classic unfair competition.

    Obviously any other student competing against the cheater has a cause of action, as would any academic department or journal gulled by the cheater. Our research into this matter has begun but we have yet to talk with any grad students. We will be noting the interesting developments as they arise. Drop us a line!

    Reply
  6. Charles Anthony Bakker

    It is worth noting that prospective graduate students, like undergraduate students, cannot take their own judgment.

    In submitting anything, whether generated or written, the student demonstrates that they hoped that what they were submitting would accomplish some desired end. This end could be admission to a program, or a good mark on their test. Regardless, though, what is submitted still says something about the student’s (lack of) understanding of what is being asked for.

    A really good essay, which demonstrates comprehension, critical thinking, and creativity, is still a really good essay, no matter how it was constructed. If the student who submits that really good essay can verify through conversation that they understand what makes what they have submitted a good essay, along each of these dimensions, then surely this speaks at least to the actual capacity/potential of the student.

    As for demonstrating honesty, if we change how we evaluate essays – by allowing them to be generated at the student’s discretion – then we take away the opportunity to use AI *dishonestly.*

    We could even warn students, as I have, that LLMs don’t write essays, they write “sounds-like-essays,” and they do this on the basis of statistical averages. So if verifying that one knows what makes a *creative* essay creative is more likely to perform well in a bell-curve scenario than will verifying that one knows what makes a more average essay average, and if AI tends to generate average essays for average students who use it in average ways, then there remains a motivation to be as creative as one can, whether with AI or without.

    Of course, what I am suggesting requires more time spent in dialogue with students. That increases the odds of those students becoming better philosophers, but it also means that, contrary to what university/businesses would prefer, we would need to accept fewer students. And while I want as many people to learn philosophy as possible, reducing our cohort sizes might have an added benefit of making the admissions process too competitive to bother submitting an average essay, AI-generated or not.

    Reply
  7. Cian Dorr

    My department (NYU) is very concerned about the issue of applicants using AI to help with writing samples, which have traditionally been the most central element in the final stages of our admissions decisions. This year, we have made two significant changes, motivated largely by this concern:

    (i) We returned to requiring the GRE, which (like most departments) we had made optional at the time of the pandemic.

    (ii) We decided to add a round of video interviews as part of the last stage of our applications process.

    While of course neither of these steps completely solves the problem, we are hopeful that adding more sources of information will make the admissions process more robust against AI-aided exploitation. The interviews, in particular, will be tightly focused on discussion of the writing samples, which should help to assure us that they represent the applicant’s own work.

    Reply
  8. TW

    As a future Philosophy PhD applicant I am curious if it is plausible to assume that this will increase the importance of personal references? I would think that at the very lease personal references from people who provided feedback on the writing sample would rise in importance as this might be a reasonable way to distinguish writing samples that are essentially 100% generated by AI (or close to it) and those that are not. Obviously a student could simply have the professor provide feedback on a writing sample that was produced by AI but given that providing feedback would typically include at least some amount of meetings and discussion it seems plausible that this would weed out many (though not all) of the fraudulent writing samples.

    Reply

Designed with WordPress