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. Charles Pigden's avatar

    Is LLM -based cheating a massive problem for higher-education? Yes. LLMs create an environment where students are incentivised not to…

  2. Jason Stanley's avatar
  3. Chris Letheby's avatar

    Generative AI is (in my view, undeniably) amazing and impressive: the idea of machines doing the sorts of things these…

  4. AG Tanyi's avatar
  5. Craig Duncan's avatar

    I teach at a relatively non-selective private college of around 4000 students (we accept around 2 out of 3 applicants).…

  6. Jonathan's avatar

    At my elite, private R1, most students in the humanities still write their own papers (judging from my scanning with…

  7. Mr. Blue's avatar

Is AI devastating higher education?

Brian Leiter

A report from Australia:

As an academic, you encounter the process at the opposite end. When you open Speed Grader to confront another bland but competent assignment, suspiciously devoid of grammatical or spelling errors, your heart sinks. You know that, in all probability, the document you’re reading was composed not by your student (whose contribution likely extended no further than dropping the essay question into Claude’s interface and adding “make it sound human”), but by a legion of absent (and, quite probably, deceased) authors, whose forgotten texts enabled an algorithm to stitch together an essay that is at least passable and often reasonably good.

Grading induces a sense of utter futility: what could be less meaningful than offering feedback to dead people?….

One study of 8,000 people found 40% admitting they had cheated via algorithm. But that was in 2024 – aeons ago, given the rapidity of technological improvement.

A 2025 survey by the ARC Centre of Excellence for Automated Decision-Making and Society at RMIT identified students as particularly heavy users of AI. Nearly 80% employed the technology – though the study did not specify what for. Research on American undergraduates conducted in 2024 by UC Berkeley’s Center for Studies in Higher Education suggested two-thirds used AI. The following year, a UK analysis suggested generative AI was “near universal” among British students….

If you work at a university today, you talk a lot with the dead. I don’t mean that as a metaphor – or, at least, not entirely….

No one knows the proportion of Australian students reliant on AI for university assessments. One study of 8,000 people found 40% admitting they had cheated via algorithm. But that was in 2024 – aeons ago, given the rapidity of technological improvement.

A 2025 survey by the ARC Centre of Excellence for Automated Decision-Making and Society at RMIT identified students as particularly heavy users of AI. Nearly 80% employed the technology – though the study did not specify what for. Research on American undergraduates conducted in 2024 by UC Berkeley’s Center for Studies in Higher Education suggested two-thirds used AI. The following year, a UK analysis suggested generative AI was “near universal” among British students….

“Why memorise facts or do equations or slog it out over an essay when ChatGPT or Gemini can do it in seconds?” a Macquarie University student identified as “Hayden” asked rhetorically in a recent report by the Australian’s Ros Thomas:

You get everything in the blink of an eye – any equation, any answer, any essay. It’s all being hacked by AI and the sooner people accept that the better. Stop worrying about what’s right and wrong, because everyone’s using it and you’d be dumb not to.

Thomas presses Hayden as to whether he knows any students who complete their assignments without AI. “Nup,” he replies. “Not a single person.”

How are things where you are? Please identify the kind of institution (e.g., R1 university, SLAC, etc.) you teach at.

I have yet to read something written by AI by any of my students here (and I have checked with Pangram to confirm my impression).

,

Leave a Reply

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

13 responses to “Is AI devastating higher education?”

  1. UK academic

    In my department, we ran all postgraduate dissertations from last year through Pangram and found quite a few had generative AI signatures, ranging between 10% and 80% of the text. It’s difficult to know how seriously to take that, because text which is only very lightly edited (e.g., cleaning up the grammar and minor edits for readability) can get flagged. Personally, I wouldn’t care too much if the Pangram detection was only for light editing, but I suspect that in some cases (I have no idea how many) it was also for the argumentative content.

    I’m increasingly of the view that we’re going to have to drop all restrictions on AI use for dissertations. Why? Because it takes way too much time to enforce that policy and to run academic misconduct cases when we do find clear evidence to support such a claim. (It also exposes the academic tasked with prosecuting the academic misconduct case to allegations of bullying, by a student who follows the best-defense-is-a-good-offense strategy.)

    If we allow AI use on dissertation, the grade would then likely be based on an hour-long oral examination where we ask really challenging and probing questions about details of the argument, the literature cited, and responses to objections. That’s unfortunate, because oral examinations are biased against non-native speakers, some people who are neurodiverse, and other people who are just not natural public speakers, for whatever reason. But at least the oral examination would allow us to measure that there was some actual knowledge and understanding acquired at the end of the dissertation process, however it was acquired.

    Reply
    1. Top 50 R1

      I myself am not so much interested in whether there was some knowledge and understanding acquired at the end of the dissertation process, however it was acquired. I am more interested in whether in whether a student has learned through sacrifice and devotion how to—himself or herself or themself—draw blood from the stone of truth.

      If AI can produce artifacts that used to be indicators of what we want from PhD graduates, then that part of higher education is hardly being devastated. Instead, perhaps we face the challenge of intensively weeding out students who aren’t passionate in their bones about the actual work including the originality part.

      Reply
    2. Paul Canis

      “That’s unfortunate, because oral examinations are biased against non-native speakers, some people who are neurodiverse, and other people who are just not natural public speakers, for whatever reason.”

      The era of apologizing for oral examination needs to be over. Where is the bias, right now? In favor of those who will use a machine. The grid of how bias works in our pedagogy, is not the same as it was just 5 years ago. The canards about oracy especially, must be mercilessly weeded out from our repertoire.

      In democratized higher education, writing became a necessary means to the required end. Large classes solved many problems and the result was, many have been educated.

      In order to save this same project, we must figure out how to return to grounding education in oracy. Our students do not have the luxury of time here, nor do we. And I refer specifically to undergraduate education, without which the graduate schools do not exist.

      (Or at least, they will no longer exist in any recognizable democratic form whatsoever.)

      I am not seeing how private writing survives in pedagogy moving forward. I think it is sunk cost thinking. Technology just threw us the biggest curve ever.

      Reply
    3. Mr. Blue

      I teach at a large R1. Much of my teaching is for a large (at one point 300 students) required ethics course for students studying data science.

      At one point, cheating via generative AI was rampant. Like many, I could not prove it. It soured my teaching (one of my favorite activities) and led to a lot of soul searching of how to solve the problem.

      Made some changes and am very happy with where I am now.

      I will not describe all the changes I made, but here are two of the biggest with respect to the cheating.

      1. I started using blue books, but not in a traditional manner.

      I still wanted to teach writing (and so did not want to have one-shot blue book exams). I thus have (closed note) writing sessions where students draft essays from scratch, by hand.* We collect those and redistribute them in sections to do peer review. We then have students write their final draft from scratch, but with access to their latest draft. During these writing sessions (including the final), students are free to consult with the teaching staff about their essays. This often promotes visits to office hours (the good kind of visits!) and has made worries about integrity vanish (a huge weight off my shoulders—phew!).**

      2. Mastery-based grading

      I hold these essay to high standards. But I combine these high standards and the blue book method with the opportunity to rewrite papers for a higher grade. But the rewriting is done just like the “final” draft: from scratch, with access to the previous commented upon draft. This introduces necessary friction; students do not overuse the system. And due to its presence, writing becomes lower stakes (making room for students to have fun and be creative with their ideas). It also makes cheating irrational (I tell them this: why cheat, which can get you expelled, when you can rewrite the essay?) Most beautifully, it mops up all grade grubbing: since instituting this system, I have had ZERO grade challenges (and, instead, good faith conversations on either how to improve the essay in a rewrite or how to do better on the next one).

      I have been extremely happy with this intervention, and, honestly, wish I had done it before. As the post notes, algorithmic cheating did not begin with LLMs. One last thing: this system does eat into time that could be spent lecturing, but I think it’s worth it. So much more learning is happening in my classes than before, and if I am not mistaken that is the point of having classes in the first place.

      *Objection: what about accommodations. Reply: I of course accommodate them, and this really is not hard to do. Many classes still have handwritten exams, universities have resources for accommodating students that can’t take these exams by hand. These students have a lot of experience navigating those systems. A quick conversation with those students early in the semester reveals what needs to be done, and what needs to be done has never been a heavy lift (neither for me nor the student).
      **Objection: they can consult with LLMs at home to prepare! Reply: Please see 2. below. Also, if some students are doing this, it honestly doesn’t bother me that much. They at least still have to be able to explain ideas for themselves, and that’s not nothing. Also, I feel confident that much fewer students are cheating now than before (in part because of what I say in 2.).

      Reply
  2. A disappointed undergraduate student

    It might be alarming that students are using AI to write their essays, but students can also be frustrated when teachers try to add AI-related content into their lessons and curricula, not as a philosophical topic, but as a way of studying and doing philosophy.

    I’m an undergraduate student in the philosophy department at a university in China. In a course on the history of Chinese philosophy that I took, students were required to select a topic in Chinese philosophy, engage in dialogue with a conversational AI, then compile a transcript of that dialogue to submit as an assignment. The intention is said to develop students’ ability to use AI to assist them in their studies.

    Furthermore, when students seek advice on studying a particular philosophical topic or writing a paper, some teachers suggest that they consult AI first before consulting them. I’ve even heard that when providing feedback for students’ papers, some teachers directly use AI-generated comments or just slightly modify them.

    I’m not denying that AI can be a useful tool to help students with their studies. However, it becomes dangerous when AI *replaces* teachers in guiding students, or even cuts off communication between students and teachers.

    All of these make students—especially *undergraduate* students—more *vulnerable* during their academic studies. Teachers may feel that undergraduate students, due to their lack of academic training, are more likely to have obvious flaws in their arguments that AI can point out, thereby rendering teachers unnecessary. But for students, this makes academic study disappointing, because feedback from AI can be obtained by anyone, anywhere, not just in college. It also makes it harder for students, especially undergraduates, to improve themselves, because AI doesn’t teach how to develop your own philosophical taste and instincts about what is interesting or valuable in philosophy. Undergraduates can hardly develop these skills alone.

    Sometimes I feel myself choked by AI and an environment full of AI. That’s not to say that there are no teachers who offer sincere guidance, but it’s still frustrating when things mentioned above frequently happen.

    Reply
  3. Bennett Gilbert

    One of the challenges the use of AI pose is that a charge of using it is very difficult for the instructor to prove and very difficult for the student to disprove. Most instructors rightly dislike the atmosphere of resentment this situation causes and therefore err on the side of cautious silence.
    I ask my students to sign a pledge not to use AI. By “sign” I mean write their signature in pen on a printed copy of this agreement. Embodying this action is a critical step. Before I introduce this document on day 1, I discuss the matter with all the earnestness at my command, carefully giving my reasons in a logical, sincere, and personal way.
    In my experience, this solves about 90-95% of the problems. There will always be some cheaters and liars; there always have been student plagiarists. Every term I find one, and I can never know how many I did not see. But, to the best of my knowledge, personalizing the commitment and the reasons for the commitment appeals to most students. They sense the agency this gives them that AI takes away from them. They appreciate personal concern for their growth and well-being. This is because most are human beings who actually have the desire for integrity and responsible personhood that the people pushing AI into “ed tech” know nothing of.

    Reply
  4. John Bussanich

    This is an informative discussion:

    https://www.ppfideas.com/episodes/can-universities-survive-ai-vs-higher-education

    Reply
  5. CB Work at a CC

    I work at a community college in a rural area. On any given assignment I average 70-80% AI usage. I fail these assignments, and I make it explicitly clear that AI is not to be used in my courses, and students can come to me for outside sources and assistance. None do, ever. The literacy rates in my class are also plummeting in tandem with AI usage, and the reticence of the student body is frightening. Mass disconnection and alienation constitute the campus ‘spirit’ of my students.

    And I’ve been *very disappointed* for the past year when I read this blog, because the number of AI acolytes and individuals actually ‘impressed’ by the technology is alarmingly too high. I thought academic philosophy would have a healthy skepticism towards AI, especially after Chomsky (et al.,) exposed just how myopic the technology inherently is.* Yet every time Leiter allows comments I see professors from prestigious institutions bragging about how ‘convenient’ and ‘efficient’ it is, and how ‘promising’ future advancements are. Mirroring the siren calls of the Nerd Reich.

    *https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html

    Reply
    1. Chris Letheby

      Generative AI is (in my view, undeniably) amazing and impressive: the idea of machines doing the sorts of things these systems do remained squarely in the realm of science fiction a mere decade or two ago. It also has substantial limitations and downsides that warrant at least a certain kind and degree of ‘healthy scepticism’.

      These claims are manifestly compatible.

      Reply
  6. Jonathan

    At my elite, private R1, most students in the humanities still write their own papers (judging from my scanning with Pangram). STEM students taking my classes often turn in AI assisted work on a first pass and then are told to do a rewrite. I discuss my AI policy and how enforcement will work openly with them once at the beginning of the semester (and of course on the syllabus).

    Reply
  7. Craig Duncan

    I teach at a relatively non-selective private college of around 4000 students (we accept around 2 out of 3 applicants). I’d say from 2023 onward I have seen more and more usage of AI to write essays and do other homework, to the point where this semester I felt forced to adjust by switching mainly to in-class graded exercises.

    I was reluctant to make this change. Philosophers have long relied on sustained writing done outside class as one of the central ways to teach students to formulate and defend arguments. FWIW, last spring I made a long reddit post voicing some of my concerns about a shift toward handwritten in-class writing: https://www.reddit.com/r/Professors/comments/1rpdicd/some_questions_about_inclass_essay_writing_as_a/

    Despite my misgivings, however, this semester I made the shift to in-class writings. I’d say my reservations have largely been borne out: less depth is possible, revising is made difficult, significant class time that would have been spent on new philosophical content is lost, and the time-pressure of the in-class writing turns the activity into essentially an exam (and thus, into a much different animal than the take-home essay, with attendant logistical headaches, like having to schedule make-up writing slots outside of class for students who missed the in-class session with a good excuse, etc.) So I grieve for what has been lost with the take-home essay.

    I’m somewhat tempted to go back to take-home essays, and be aggressive with programs like Pangram, etc. However, I fear that reverting to policing at-home essays would only be a temporary solution: surely the arms race between AI programs and AI dectors will continue to seesaw back and forth, so that we can’t always count on having reliable dectors.

    And honestly, it was nice when grading the in-class essays to know they were 100% student produced. (Like “UK Academic” above, I found it to be a dreary fate to have to grade essays I believed were significantly machine-written. What was even worse was giving some of those essays higher marks than essays that were clearly human-written, with the usual human flaws visible.) So I would guess I’m likely to stick with in-class writings in all but my upper-level seminars.

    An aside: I don’t really see oral exams as a great solution. With around 70 students a term for me, even ten-minute oral defenses of three essays would amount to roughly 35 hours of interviewing, before scheduling and preparation are counted. That’s not going to happen. And what is more, students could still have AI write their essays. They would merely need to understand and remember the essay that AI has churned out for them. So they still wouldn’t learn to write or think creatively all that much.

    I’ve seen the same tradeoff with reading quizzes. For years I assigned short online at-home quizzes due before class, partly to incentivize reading but also to use my questions to direct students toward important ideas in the texts. Once students could simply give AI a screenshot and get the answers, I saw quiz scores go up and up. As a result, I’ve now moved to two-question in-class quizzes at the start of class. These provide some incentive to read, but the questions necessarily have to be much simpler and rely on memory. So here too, I’ve gained confidence that the work is the student’s own but at the cost of losing something pedagogically useful.

    So in answer to Prof. Leiter’s question, I wouldn’t say my classes have been “devastated” by AI. They still function; students still learn philosophy. But AI has pushed me toward forms of assessment that are more AI-proof precisely because they are more restrictive. In many ways that has made the courses worse.

    Reply
  8. AG Tanyi

    I work in Italy. Less of an issue here perhaps, but it is an issue. I hold workshops for students at the end of the course with slight supervision prior. In the workshop they present their ongoing work, then they can submit. This helps. Obviously, it doesn’t work for larger courses. Otherwise, oral exams are standard here, but I don’t like them. So if I must I opt for sit-in exams on controlled desktops. That’s about it.

    I am not sure how long this Pangram solution will work (if it is a solution). Isn’t the whole point of AI progress that it will also hack this? Anyway, I don’t use it since it would expose me to all sorts of extra requirements, complaints etc. I anyway think that if students want to stay dumb, they will stay dumb, no matter what I do.

    Reply
  9. Charles Pigden

    Is LLM -based cheating a massive problem for higher-education? Yes. LLMs create an environment where students are incentivised not to learn but to FAKE learning (since by doing so they tend to get higher grades). The robots help create an environment where cheaters tend to prosper and in which, accordingly, many people cheat. Are there reasonably effective solutions to this problem? Also yes. To re-incentivise genuine learning we have to disincentivise cheating by reconfiguring our assessment regimes, so that cheaters tend NOT to prosper. Some such regimes are, to my mind, rather dubious and a tad too labour-intensive. Some of my colleagues set take-home essays and then cross-examine the students on the results. This is a) rather subjective (penalising the merely shy and tongue-tied) and b) takes an inordinate amount of time. Some get their students to write their essays in-class either by hand or on computers that have been insulated from the internet (and therefor function as little more than word-processors). I have some sympathy with this approach but I think it takes too big a bite out of in-class teaching time. (If they are writing in-class essays about last week’s topic then we won’t be discussing this week’s topic.) My solution is to allow some of the mark to go on take-home essays (which are a valuable learning tool, though of course not cheat-proof) ) but for the bulk of the mark to go on the final exam (50% this year, 70% in the last two years of my teaching career). Students are likely come a cropper (and I tell them that they are likely to come a cropper) if they do not write their essays themselves because the exam questions do not correspond directly to the essay or the class discussion prompts but require an in-depth understanding of the issues. (The exam questions cover much the same issues as the essay prompts but they are designed to come at them from a slightly different angle. Thus memorising and regurgitating an AI-generated essay that is addressed to the essay prompts rather than the exam questions will probably lead to failure, especially as I mark hard for relevance.)

    So there is at least one viable solution to the problem of AI-based cheating, that is , reverting to an exam-based assessment regime. Is it otherwise undesirable to do this? Yes, because final exams are, in the odious jargon which we have to put up with, ‘summative’ rather than ‘formative’: you don’t learn by doing them but only manifest the learning that you have already acquired. In the absence of cheat-facilitating LLMs an assessment regime based on long-form take-home essays would definitely be preferable (at least in philosophy) since students learn about the topic in the course of writing their essays. So in minimising the share of the grade that goes on take-home essays (for example) and maximising the share that goes to cheat-proof mechanisms such as final exams, there is undoubtedly some loss. But it is absurd – simply bonkers – to suggest that nobody can acquire either a love of learning or a love for their subject in the context of a (relatively) cheat-proof assessment regime. After all, many of us have done precisely that. From the age of eleven until I graduated from Cambridge at the age of 22, all or most of the final mark for the vast majority of my courses went on a final exam. 

    For context, I teach at a University that is technically an R1 (since it is a research -focused university that grants doctorates). However to Americans this label may be a bit misleading since we don’t run a graduate school in the American sense and the vast bulk of our teaching is devoted to undergraduates. This semester was the first time for about a decade that that I have taught an exam-assessed undergraduate course. It was a split-level 200/300 paper for about thirty students with the 200-level and the 300-level students being subject to slightly different assessment procedures. The exam-based assessment regime had the effect of stretching out the grades and restoring something like a normal curve of distribution, which suggests that there was some AI-based cheating in the past and perhaps some in my student essays this time. However, it was not a case of ‘crash and burn’. Quite a lot of students got an exam mark that was roughly commensurate with their take-home essays. So the stiff talking-to I gave them at the beginning of term in which I explained that AI-based cheating wrt take-home essays would probably lead to disaster in the final exam seems to have worked.

    In the future, 70% of the assessment for all of our undergraduate courses at Otago must be ‘learning assured’ , that is based on cheat-proof mechanisms (though we will have a considerable degree of latitude as to how to manage this) . In the short term I expect this will lead to a drop-off in average student performance since AI-based cheaters will fail – sometimes spectacularly – to prosper But in the longer term OUR students (as opposed to the students at colleges and universities which don’t impose similar requirements) will be at a premium in the job-market since we will be able to say with our hands on our hearts that successful graduates will have genuinely mastered their subjects, acquiring a range of cognitive skills in the process.

    If there were no solutions to the problem of LLM-based cheating then wailing and gnashing of teeth might be appropriate. But since there are such solutions (suboptimal though they may be) the wails of sorrow are somewhat premature. 

    But though wailing and gnashing of teeth would be inappropriate, indignation is not. Those bastards who created these bloody things have forced us to reconfigure our teaching and assessment regimes, often for the worse. From a teaching point of view take-home essays are lot better than end-of term exams and it is really unfortunate that we have been forced to revert to the latter. Is there any prospect of SUING the big AI companies for the damage they have done and the time and effort that we have had to put in dealing with the (probably) unintended consequences of their actions?

    Reply

Designed with WordPress