In a philosophy seminar—“my favorite class I’ve taken so far,” she said—Gwen used AI to write almost all her essays just to avoid late submissions.
That’s “Gwen”, a student at Yale, interviewed for an article by Alex Moore in The New Journal about how students and professors at Yale are grappling with the use of AI.
Here’s some context for the above quote:
In the fall of her sophomore year, Gwen began falling far behind in all of her classes. She struggled to prepare for tests, and forgot deadlines until the last minute or missed them altogether. It was a “sophomore slump,” she said. Too embarrassed to seek out her professors or peers for help, Gwen turned to ChatGPT. (She asked to be referred to by a nickname to speak openly about her artificial intelligence use, out of fear of disciplinary action.) By the end of the semester, she was using it to help complete many of her assignments.
By last spring, her struggles had only worsened. In a philosophy seminar—“my favorite class I’ve taken so far,” she said—Gwen used AI to write almost all her essays just to avoid late submissions. With AI, Gwen could hide her struggles from friends and professors. She never told them she was using AI. On rare occasions, when her friends saw she was falling behind, Gwen said they would joke: “Oh, your essay’s due tomorrow? Just use ChatGPT.”
AI helped Gwen turn her philosophy papers in on time. But guilt, she said, “was just eating me alive.” She felt that she was deceiving her professors and betraying the purpose of her education. “If I’m using AI instead of doing the work myself, why am I here?”
When her professor emailed her feedback on an AI-written paper, she refused to open it. She still hasn’t. “It’s wasting his time because he’s editing a fucking machine,” Gwen said. “Whatever he has to say is meaningless to me anyway, because it’s not my writing.”
Meanwhile, some professors are trying to understand what their students are up to. Math professor John Hall is quoted as saying, “I don’t think people are trying to get by and not learn. But I think they don’t realize in the moment what they’re learning or not learning.”
Others “expressed sadness or frustration,” such as philosophy professor Shelly Kagan: “Yale offers you this astonishing opportunity to stretch your brain. If you want to just waste your time [by using AI], then you’re an idiot.”
Maybe professors conveying that message can make a difference at the margins to certain kinds of students (at least for a time—who knows what educating the kids who grew up with AI will be like?).
Moore writes:
By the end of the spring semester, Gwen reached a breaking point with AI. “I felt like I didn’t deserve to be a Yale student,” she said. She thought that she was throwing away her education and the scholarship paying to support it.
She decided to quit using ChatGPT altogether. She didn’t ask AI to help her sentences flow or to strike the right tone. She didn’t even use it to brainstorm. “I think that’s a skill you need to have on your own. That’s the core of being original, anyway.”
Gwen recalled completing her final assignments without AI, sitting in the middle of a crowded library so that she felt watched. “I struggled. And they weren’t very good.”
It was only then, she said, that she started to understand just how much learning she had missed out on.
Of course, Yale students are hardly representative of the average college student (and Yale is hardly representative of the average university). Good for Gwen, but her revelation doesn’t make for much of a feel good story.
When the takeaway is finally, a gifted and morally sensitive student at a top university realizes that cheating is bad for her, what hope should we have for everyone else?
The whole article is here.
(via Shen-yi Liao)
Tomorrow, we’ll look at an interesting way of approaching the problem of preserving trust and judgment in a world with AI.
What can be done to turn this ship around? I genuinely fear for the future.
Just stop assigning take-home essays. Find other ways to cultivate and test learning.
I have done exactly that. However, writing short essays in class and administering oral exams (if possible given one’s teaching load) don’t cultivate the same intellectual skills and virtues as slow, multiple-draft, essay writing.
I agree, but neither does giving assignments that allow students to use an AI tool to do their work for them. If the ideal version of a thing is no longer available and the current version is unacceptable then one has to abandon the thing, unless something radical changes.
Get proctored computer labs without general internet access, where students can sign up for working times, and log in to access class files and their drafts, and write at various points in the week or term.
That resource simply does not exist at my cash-strapped University.
Justin asks: “When the takeaway is finally, a gifted and morally sensitive student at a top university realizes that cheating is bad for her, what hope should we have for everyone else?”
One answer: engage with students on these questions, *in class*, early and often. Show them how bad (by our standards) AI answers to questions are; and make it clear that learning to write bad answers is both epistemically *and* practically unhelpful to them. If you hide the right answer to this question, don’t be surprised when some (many) people come to the wrong one.
I believe that this is very important, but I also think that AI is going to get a lot better. AI can’t handle my assignments properly yet, and I encourage my students to play around with it to learn its limitations, but as AI improves, I believe that I’m going to have to completely reinvent the way that I teach.
Yep, this seems like at best a short term fix. It’s true that in 2025 if you just cut and paste the essay prompt into an LLM, and cut and paste the response into your essay, the result is not great. But (a) most students are a bit better at using LLMs than that, and (b) the LLMs are getting better, so by the time these students graduate who knows.
We can see this already in more formal classes. It can just do a whole bunch of problems, and can do more every year. Convincing the students that the results are bad isn’t going to cut it. At some point this kind of argument is like saying “You should actually run the 5K training run rather than drive it because the car might break down.”
But of course they should run the 5K training run. It’s good training. The trick is to convince them of that when all we can do is stand at the finish line and see who made it to the end.
We’re not supposed to talk about the fact that AI gets better over time. Remember, any discussion of future harms is just a distraction from current harms, and we ostracize anyone even attempting to point them out. Otherwise our research lab’s funding from Google or whoever will get cut
It’s also important, especially for lower-division courses, to help students understand that bad philosophical writing is part of the learning process.
Being wrong is half the fun!
Being wrong is half the fun – but unfortunately, struggling to write a paragraph that doesn’t quite express your idea as clearly as you like isn’t nearly as fun (though it’s also an extremely important step in the process).
The challenge with that approach is twofold: first, it reinforces the idea that the only reason not to use AI is that it’s not good enough yet; second, for a lot of assignments, and at places outside of Yale, it is perfectly sufficient. If you’re a student taking philosophy as a gen ed, and you don’t care about it, and feeding it to AI gets you a B-, which fits in well enough with your GPA, then why wouldn’t you, if the lesson you hear is ‘it’s not that good yet.’
I tried this approach with all my classes this semester. As far as I can tell, it caused no meaningful shift in chatbot usage.
been doing this since llm chatbots came on the scene. it doesn’t work.
Surely there is an answer here. It’s not great, and it’s otherwise undesirable in many ways but it will go a long way to disincentivize AI-based cheating. We simply revert to written exams for a major part of the grade. In my lower-level classes there will be a mark for a presentation, marks for two take-home essays but 50% will go on the final exam. Of course some will cheat with their take-home essays and I won’t be able to stop that. But such cheaters will be unlikely to prosper when it comes to the final exam. I will design the exam questions to be sufficiently different from the corresponding essay prompts that it won’t be possible to memorise an AI-generated essay and successfully regurgitate, especially as I will mark hard for relevance. They will have to answer the specific question I asked in the exam not the similar question that I asked as an essay prompt or they will lose points big-time. Furthermore they will have to answer one exam question on a topic that they have not written an essay on. Those who relied on AI to help write their take-home essays will be likely to come a cropper when asked write and think for themselves in the exam. This will gradually become obvious to students, making them less likely to cheat wrt their take-home assignments.
I would much prefer a system based entirely on class-presentations and long take-home essays (much better pedagogically IMHO), but AI has effectively wrecked this as an option.
It should not be forgotten that one of the functions of universities (and an entirely legitimate one in my eyes) is as gatekeepers to the lower slopes of the elite. A university degree should say a) the you know a fair bit about some worthwhile stuff, and b) that you have a range of cognitive skills and capacities which make you useful both to society and to potential employees. Unless we disincentivize AI-based cheating it won’t say either of these things and people will be getting their entry-ticket to the elite under false pretences. But the graduates of universities that DO disincentivize such cheating (and do so very publicly) are likely to be at a premium in the job market.
A far better answer is to continue with take-home essays but combine them with a short follow up interview where the instructor asks the students questions that test their understanding of their own work and then assigns them a mark based on how well they are able to take intellectual responsibility for the work they submitted.
This is a very good idea.
I really love this but many (most?) of us teach 75-100+ students across 3-4+ classes per semester. I’m not sure how to do this with such a load, but maybe others have?
I discuss the oral exam method in more detail here. Ultimately, I argue that universities need to respond to AI by investing in more intensive forms of instruction. Their standard business model is now under threat and people are increasing asking: If AI can complete the classwork set for students, what value do universities provide? The answer is to invest in more intensive, high-quality forms of education.
However, I understand that many universities are not going to do this and that many instructors are stuck with a heavy teaching load. If you are in this situation, one thing you can try is readjusting your syllabus to make room for oral exams. For example, you might reduce the content in your course and leave the last two weeks of class time free for conducting oral exams. Or, if you used to assign two essays, you might replace that with one essay plus an oral exam (presumably conducting a short oral exam and marking a second essay take about the same time). Obviously these options involve compromises that might be less than ideal, but in difficult circumstances we must be willing to make such compromises.
‘Ultimately, I argue that universities need to respond to AI by investing in more intensive forms of instruction.’ Perhaps this would be a good idea Mark , though there are many other good causes to spend public money on, but it isn’t going to happen in the short term and probably not in the long term either. In the meantime we need to find solutions that do not add unduly to academic workloads and do not cut in too much to teaching time. Most of the work-arounds designed to eliminate or mitigate AI-cheating wrt the long-form take-home essay require policing activities that either add to academic workloads, cut into teaching-time or both. This is certainly true of your solutions. ’If [you have heavy teaching load, one thing you can try is readjusting your syllabus to make room for oral exams’ – that is, teach less, police more. ‘Or, if you used to assign two essays, you might replace that with one essay plus an oral exam (presumably conducting a short oral exam and marking a second essay take about the same time)’ Again this involves less time on education and more on police work since writing an essay is itself an educational process whereas the chief point of an oral exam is to flush out AI cheats. There is also the problem that the mark for an oral exam is going to be a lot more subjective – or, at least, a lot less intersubjective – than the mark for a written exam. If a student disputes the grade for a written exam, the mark can be checked by a colleague. This is not the case for an oral exam unless the sessions are recorded which, I suspect, is legally problematic. I will be replacing one of three take-home essays with an exam (reversing the previous process in which I replaced an exam with a third essay) which will be sat outside the teaching term. It will add a bit to my workload since toiling through hand-written scripts a lot more onerous than marking typed essays) but it won’t cut in to classroom time and won’t poison my relationships with my students by forcing me into the role of an AI policeman. Of course, I enjoy the advantage of a pre-existing centrally administered examinations framework so I won’t have to cut into class-time by invigilating the exams myself. But for those who are not in such a position, I think a written examination is still superior to an oral examination a) because it is more intersubjectively checkable and b) because getting them to sit an in-class examination need not take more time than one or two teaching session. (My classes are all two hours each).
Way too much work for a 40-member class.
Especially for, say, eight sections of that a year (but even just four or five sections that big, really), or with more than just one assignment.
There are a fair few options that sound great, but just don’t scale to the kinds of teaching loads some of us have.
It seems to me that a solution could be to reduce the content covered by a week or two and dedicate that to this activity. That does mean the class will cover less material but they might learn more overall. And something like this isn’t unprecedented. There are foreign language classes that include an assignment that is basically an interview with the professor in that language.
In my institution, over 50% of credits are being “earned” in asynchronous online sections. I can’t require students to come in for exams on my own, and if we required it for all philosophy courses, students wouldn’t register for our courses.
There are some students who are doing the work on their own, but this is a community college setting, and those students’ work is far inferior to those who are using AI to do all or part of their work.
I know that, in this environment, asynchronous online courses are utterly devoid of meaning. But the administration and the state don’t agree with me. They keep telling me that I need to find ways to integrate training in AI use into my Intro to Philosophy course.
What I need from my colleagues at universities are some policies that say that credits earned in asynchronous online courses in the humanities won’t count for transfer credit.
Sorry M K, but I am afraid I have no helpful suggestions to make, as your situation is too remote form mine. I am lucky enough to work at a fairly prestigious university – somewhere in the top couple of hundred in the QSM rankings – with a moderate degree of autonomy and a ‘legacy’ apparatus for conducting centrally administered exams. Solutions for the AI crisis that might work or us won’t work for other people.
I suspect that in the future the top graduates – and maybe even the middle-ranking graduates – of institutions that have a name for being tough on AI cheating will be at a premium in the job market, but this isn’t going to solve anybody’s short term problem.
When grading, I can tell when the paper is written by AI, but if it’s not beyond reasonable doubt, I don’t feel comfortable marking it against the student. I hope students are aware that their reputations, not just grades, are on the line.
I think peer review could be a very powerful deterrent in this case. I’ve been in classes where reviewing and grading other students work was part of the grade. Peer pressure can be used for good, as no one wants to read an AI paper, not even the students.
I had a student come to me this semester, LIVID, because her peer reviewer had used ChatGPT for the review.
students are policing each other more than we are policing them in my experience. this should give us pause on the “cop shit” rhetoric recently linked here, i think.
fwiw, i don’t like cops and the last thing i want is to do “cop shit” as a teacher. but students who care about college education are starting to react to prohibited ai use by peers. perhaps we too ought to care.
Agreed. I also hate cop shit and believe in being in community with students. But part of being in community is that we hold each other accountable.
It’s remarkable that she was able to get away with this at all. I’ve found AI to be so exceedingly bland and generic at philosophical analysis that it’s difficult to believe it could produce a convincingly high-quality undergraduate essay that would fool a Yale professor. But perhaps she is craftier, relying on it as an undetectable crutch rather than a wholesale ghostwriter.
One takeaway from this anecdote is that we’ve filled top colleges with strivers who seek the prestige of the name-brand degree. They approach education instrumentally, and so they think of cheating as a high-risk, high-reward strategy in an adversarial game, not a deplorable and self-destructive moral failure. So, university instructors should be spending time in classrooms urging these people to care about their own education as such, showing them how it makes them better and more versatile people overall. And without lame moralizing platitudes that just sound like we’re trying to justify the academic job market.
It is also clear that education will need to change in response to AI. It is already ubiquitous and won’t go away, so we will need to be creative, not just to prevent cheating but to make sure students can’t be tempted to offload their formative thinking to AI systems. Scaffolded essays based on in-class writing exercises. Oral exams in senior seminars. Holding students (if administrators will allow it…) accountable for completing their assigned readings and being able to intelligently discuss them. AI might be good for society, but until institutions adjust their policies to accommodate it ethically and epistemically, it’s just going to make the pre-AI systems less and less tenable.
I was in one of my school’s libraries today which is mostly full of undergraduate students. The seats are arranged in rows so you can see other screens quite easily. I couldn’t help but notice that almost every single screen had chatGPT open with its distinctive interface, with the other side of the screen a word document or a reading or something of that sort. It is quite depressing to realize students, many of whom you teach, are using AI for not just their final papers but for much of the process that makes education worthwhile.
I really do wonder if the solution is either a) something like what others have motioned to, by way of tests, quizzes, in-class writing, requiring draft process, etc., or more radically, b) grade abolition. I’ve been thinking about b) a lot, even before chatGPT, and the case for doing away with grades seems stronger than ever.
I’m in a philosophy master’s program and the use of ChatGPT is even prevalent in the classroom. There’s been times when I ask a classmate a question during class discussion and they’ve typed it into Chat and then read it back out to me. Not much lateral exchange going on.
Everyone is responding to incentives, so the solution is to reshape the incentives: https://certifiedaifreeskillsandknowledge.org/
I agree with reshaping incentives. I went to an undergrad institution that did not tell people their grades and created social pressure not to know one’s grades. It had a good effect. (And now that I think of it my grad school didn’t grade us either.)
But it took students who got the point of that to have those good effects.
Can we replicate that at scale like at the large behemoth at which I currently teach?
I suspect we should just get rid of grading as in doing the thing we put in transcripts. (But I’ve always thought that which is why I went to the place I got my BA from after dropping out of a place much like I teach at now.) The thing is to get students to do things for their educational value and not for the credential. But even there, the example of the person who liked philosophy but had completion issues, shows that even that isn’t enough. This is a hard problem. I’m glad I’m old.
Hey Marc, I’m very sympathetic to this effort, however I think the ‘auditor’ requirement is going to sink it. Since we’re on the topic of incentives, I bet that 95% of people who would otherwise be interested will turn away upon learning that they have to arrange an outside visit to their class. I certainly did.
I’m sure you’ve thought about this, but someone who was really determined to “fake” this credential could probably fake an auditor. In addition, there’s nothing stopping that person from having a single AI-free day. Verification here seems like the wrong approach, because you need large numbers to sign up and you just won’t get it with this requirement.
Thanks for making the attempt, and I hope it takes off.
“When the takeaway is finally, a gifted and morally sensitive student at a top university realizes that cheating is bad for her, what hope should we have for everyone else?”
Sorry, this one made me laugh. Student is “gifted” because at “elite” university and is “morally sensitive” because she realized that cheating was bad.
Maybe this just shows my working-class minded childhood and non-elite BA institution, but both of these assumptions … ewww. How will the “rest of us” (and our students) ever have any hope if Yale students can’t get it right?!?
Also, I wasn’t actually convinced that this student was “conscientious”. I mean, if she is so “conscientious”, how is she “falling behind in all of her classes”? I don’t think that’s a deep moral failing. I have fallen behind in many thing in my life! But, I also accept that that means that I’m less conscientious than I’d like to be. Feeling bad about falling behind doesn’t make you conscientious, nor does cheating to catch up (and then feeling bad about that.)
On the occasions where I’ve caught students cheating, them feeling bad about it and saying that they cheated because they felt they were falling behind, didn’t suggest conscientiousness to me. I had sympathy for them because I had fallen behind myself many, many times. But merely knowing that one’s cheating is due to the feeling of falling behind never struck me as a reason to think of someone as conscientious. It suggested insight into why they did what they did. But conscientiousness? Probably a bridge too far.
When I encountered colleagues who admitted (usually after a few drinks) to having cheated (on high stakes assessment in their undergraduate years), I was even less inclined to appraise them as conscientious. My first thought was: “Why are they telling me this? This is making them look bad! How do I know they aren’t still cheating, but that cheating is taking a different form now? etc.”
Conscientious students fall behind for all kinds of reasons. Outside responsibilities may overwhelm them. Tragedy may distract them. They may become ill. They may even find that they lack the skills or talent for what is required of them.
It is true the even the conscientous may be overwhelmed by outside forces. There’s no indication of that, here, though. It’s to the student’s credit that she doesn’t make up the sort of sob story most of us have heard about why she couldn’t do what she should have done. And, of course, taking on more than one can do it not an excuse for not being consientious, but a way of not being consientious. So, in relation to this instance – all I was talking about – I’m happy to stick with my evaluation.
well, without a protracted analysis of conscientiousness, why not just go with a common definition, and call it a wish or desire to do what is right, say to do so thoroughly?
that’s compatible with not knowing what the good is, not knowing what is right, etc.
here conscientiousness won’t be proof against ethical failure, but its presence may indicate some possibility of helping conscientious students choose differently, or of our altering their situations so that a choice to do what is right would be easier for someone who falls short of virtuous (esp. for some unity-of-the-virtues reason) to make.
why not just go with a common definition, and call it a wish or desire to do what is right, say to do so thoroughly?
The thing about wishes is that they are free. A desire that doesn’t actually move the person is also not worth much. I am thinking of being conscientious as having a virtue, and having a virtue is necessarily success-connected. So, wishing, or even wanting, to do what’s right isn’t enough, any more than wishing to be brave is. I don’t think this is a moral failing (at least not on its own) but it is a defect of character.
I believe that you have to be gifted to get into Yale. If not, something is wrong with their admissions process. As for “moral sensitivity”, while the student hasn’t demonstrated that they are a paragon, they are demonstrating some moral sensitivity.
two words: legacy admissions
The problem is what college has become. College is no longer about learning. It’s no longer about scholarship. It’s about getting a job. And almost none of the stuff you learn in college are going to be used in the world when you get a job. So why try? Why bother? Why do anything in a world that only cares about the degree and the job you make and the starting salary you get.
I remember feeling like I cheated by reading a secondary source on Aristotle’s Metaphysics, lol. Those were the days.
As someone teaching online, asynchronous philosophy classes, any advice on what the hell to do?
I require my students to share their essays as Google docs with me, giving me the editing history. This helps somewhat. Any other ideas for assignments that would skirt the possibility of using AI? The class is intro to ethics, by the way.
admittedly cynical response: admin, departments, and programs have clearly
abdicated anything resembling a stance on these matters, broadly speaking (cf adjunct labor). there’s essentially no institutional support of any position on ai use, pro-, anti-, or centrist.
whatever you try may or may not work and it may or may not work the second, third, etc. time you try it. you can do everything, nothing, or something in between (like many of us) and basically the outcome will be more or less the same. it’ll depend on the students of the specific class.
our only hope for real consequences is social opprobrium from peers. we should all take a moment & let that sink in. our institutional decrepitude is such that our best hope is 18yos deciding that using ai is cringe.
they’re right of course.