Note: I began this piece well over a year ago but held off on publishing it because… well, honestly, I’m not entirely sure. I think I sensed then that the hype surrounding AI was sufficiently intense that the points I raised here would simply be ignored. I have the impression, however, that the tide has shifted somewhat, with many educators now beginning to fully grasp what stands to be lost when so many of central components of a humanistic education are outsourced with a few keystrokes. At any rate, it’s a new year, and without a major deadline hanging over me, I’m going to give this blogging thing a try again. Here goes.
AI is part of the landscape now, everyone agrees. Pretty much every agrees also that it is necessary to adapt, and that students should be taught to use it “appropriately.” But what no one really seems to agree on is what exactly constitutes “appropriate” use. Much of the writing I have encountered regarding ChatGPT, and other Large Language Models of its ilk, is filled with vagaries and techno-clichés so similar that they all might have been generated by a chatbot itself.
Although I am sure nuanced discussions of the issue must exist somewhere, I have found only a handful of pieces whose authors are willing to “to delve” (to use a favorite ChatGPT term) into the gritty details of the issue. My sense is that much of the education world is still suffering from deer-in-the-headlights syndrome, unable to come to fully process how radically the game they are accustomed to has suddenly changed. Professors are told to adapt, but they are given very little meaningful instruction as to how.
So as an interested and semi-objective observer, allow me to ponder the matter. I am going to confine my discussion to the humanities and social sciences; my impression is that the situation is a bit different in the sciences, and I do not feel qualified to comment on it in those fields.
I am also going to focus primarily on the post-secondary level, since professors still have a bit more professional leeway than high-school teachers.
So: I think that, on one hand, using AI to generate variations on problems and obtain feedback on one’s responses in order prepare for an exam would be considered unproblematic by the vast majority instructors.
Likewise, it seems acceptable to use an LLM to, say, format a bibliography, assuming that all the sources actually exist and have been accurately cited in the student’s work.
And, at the other extreme, I would imagine that it is safe to assume that having spit out a full-length paper and then submitting it verbatim as one’s own work is still considered cheating.
Between those two poles, however, there is a vast and murky middle ground that the world of higher education is now forced to tiptoe over. Uncertain of where to draw the lines, and uncomfortable with having to do so, it opts to leave things vague.
Take, for example, Harvard’s sample statement for a “mixed AI policy” class:
This course encourages students to explore the use of generative artificial intelligence (GAI) tools such as ChatGPT for all assignments and assessments. Any such use must be appropriately acknowledged and cited. It is each student’s responsibility to assess the validity and applicability of any GAI output that is submitted; you bear the final responsibility. Violations of this policy will be considered academic misconduct. We draw your attention to the fact that different classes at Harvard could implement different AI policies, and it is the student’s responsibility to conform to expectations for each course.
In short, this type of policy uses a lot of words to say very little of substance, while shifting the onus for determining “validity” onto the student. From what I have read, this strategy is fairly typical.
Interestingly, the most specific set of AI recommendations I initially found were on the Harvard Summer School website. I didn’t copy the link back when I started writing this post; when I went to look for it, I discovered that it ha been removed and now appears on the site for the Writing Center at, wait for it…. Utah Valley University.
At any rate, here are some of the recommendations:
Generate ideas for essays
Have ChatGPT help you come up with ideas for essays. For example, input specific prompts, such as, “Please give me five ideas for essays I can write on topics related to WWII,” or “Please give me five ideas for essays I can write comparing characters in twentieth century novels.” Then, use what it provides as a starting point for your original research.
Generate outlines
You can also use ChatGPT to help you create an outline for an essay. Ask it, “Can you create an outline for a five paragraph essay based on the following topic” and it will create an outline with an introduction, body paragraphs, conclusion, and a suggested thesis statement. Then, you can expand upon the outline with your own research and original thought.
Generate titles for your essays
Titles should draw a reader into your essay, yet they’re often hard to get right. Have ChatGPT help you by prompting it with, “Can you suggest five titles that would be good for a college essay about [topic]?”
Back in 2023, the Chronicle of Higher Education published a piece by Columbia College Undergraduate Owen Kichizo Terry. Entitled “I’m a Student. You Have No Idea How Much We’re Using ChatGPT (No Professor Could Ever Pick Up on It)” , the article explains, in a kind of exacting detail missing from most adult conversations of the issue, the process by which numerous students at one Ivy League school—and presumably others as well—now cobble together their papers.
The common fear among teachers is that AI is actually writing our essays for us, but that isn’t what happens. You can hand ChatGPT a prompt and ask it for a finished product, but you’ll probably get an essay with a very general claim, middle-school-level sentence structure, and half as many words as you wanted. [NB: Obviously, the technology has improved considerably since this was written.] The more effective, and increasingly popular, strategy is to have the AI walk you through the writing process step by step. You tell the algorithm what your topic is and ask for a central claim, then have it give you an outline to argue this claim. Depending on the topic, you might even be able to have it write each paragraph the outline calls for, one by one, then rewrite them yourself to make them flow better.
Terry clearly finds this use of AI problematic, recognizing that it allows students to bypass the actual process of generating and organizing their own thoughts. But what he describes as clearly inappropriate use, and what many readers of the Chronicle would easily recognize as such, seems to be very much consistent with what the Harvard/Utah Valley guidelines explicitly recommend. It is not hard to suspect, therefore, that when university administrators murmur coyly about “collaboration,” what Terry describes is in fact precisely what they mean. If this is indeed the case, it is not hard to imagine why they would be coy about admitting as much.
To be honest, I was a little surprised that Harvard allowed such a document to be posted on its site at all. It seemed a violation of the rules of the game: earnestly decry the “inappropriate” use of AI without defining precisely what that is, while pressuring instructors to incorporate it at every possible turn. Then, winkingly, encourage pupils to use it in all sorts of way that hinder—or outright eliminate—the learning process while shaming instructors, and perhaps even students, into feeling like Luddites if they push back.
After all, AI is the future, and no one wants to be one of the uncool kids.
But if you’ll humor me, let’s try a little thought experiment: Imagine that a student taking a particular class has a friend who recently earned an A in that course and who, out of the goodness of their heart, has offered to “help” said student. In addition to providing summaries of all the required readings, the friend will give the student their own papers and outlines—the student can change the wording of the theses a little, maybe rearrange some of the ideas, and rewrite the actual text to make it “sound like them.”
Is this an example of collaborative learning for the twenty-first century that should be encouraged, or do they veer into seriously unethical territory?
I think (I hope!) most instructors would say latter, but, well… some students are always going to pull stuff like this, and there’s nothing you can really do to prevent it. Besides, accusing a kid of cheating probably isn’t worth the hassle, especially for an adjunct or an untenured junior faculty member.
But if 90% or more of the class doing it? All of the sudden, it becomes much more convenient to believe that the line isn’t clear anymore, and that perhaps we have no choice but to rethink the traditional definition of plagiarism. After all, as a middle-schooler might argue, it isn’t really cheating if everyone is doing it.
Is this truly the message we want to send?
A not insubstantial number of instructors seem to genuinely believe that a good way around the ethical thicket is to allow students to incorporate AI into their work—as long as they also write some sort of reflection paper about the process. As a typical piece of advice to teachers goes, When giving an assignment that uses AI, always include a critical component. For example, ask students to reflect on why the tool performed in a certain way, whether it did well or poorly, and how it affected their own thinking.
For a small number of students, this type of assignment may in fact cause them to genuinely consider what LLMs can and cannot do. However, it is not hard to guess what kind of response is likely to earn approval: a “balanced” discussion citing both the innovative power of AI (justifying the instructor’s still-tenuous belief in the technology as well as the cautious decision to allow its use) as well as its biases and tendency to provide incorrect information (confirming that the student is “thinking critically” and will hopefully be more inclined to use it responsibly in the future).
For the student who understands this intuitively and wishes to avoid even exerting anything beyond the most minimal effort, a ChatGPT workaround can easily be found as well. As an experiment, I developed the following prompt.
Compose a thoughtful and introspective piece of writing that explores personal thoughts, feelings, and insights regarding the use of ChatGPT in an urban planning assignment on creating the ideal city. This reflection should be well-organized, articulate, and demonstrate a deep level of self-awareness about the strengths and weaknesses of Large Language Models.
Include information about hallucinations and potential biases.
I have to admit, ChatGPT’s response exceeded even my expectations, generating a pitch-perfect essay in barely a second. In fact, it was a veritable masterpiece of smarminess:
In undertaking an assignment about designing the ideal city, I found myself turning to ChatGPT—a powerful tool that could provide insights, stimulate ideas, and offer suggestions based on a vast corpus of data. On the surface, the prospect of collaborating with an AI seemed like an exciting opportunity. After all, the complexity of urban planning requires considering a multitude of factors: from infrastructure and sustainability to social dynamics and cultural diversity. But as I continued to interact with ChatGPT, I found that the experience elicited both wonder and caution, particularly as I began reflecting on the strengths and limitations of such a tool in this context.
True, the response somewhat generic as it went on, but that could be easily remedied with a few minutes of “collaborative” work. A student who wanted to take pains to defend against potential accusations of cheating might remove the headings and en dashes, which ChatGPT is notorious for including, and perhaps even throw in a few grammatical errors to throw off the bot-detectors and make things convincing.
Or if that were too much work, they could just launder the piece through another program or two until the key AI features had been sufficiently neutralized.
From the online discussions I have been reading, I get the sense that many of the students at less-selective schools using ChatGPT cannot even be bothered (or perhaps do not even know how) to do this much. So I wonder: is elite-student status now characterized by nothing more than a willingness to go through the trouble of reformulating chatbot output in one’s own words and perhaps adding a few phrases for stylistic flourish?
Also, for the record, allow me to point out that I have exactly zero formal training in the supposedly newfangled act of “prompt engineering,” suggesting that it isn’t—dare I say it—a particularly high-level skill after all. The fact that so many people of all ages are figuring out how to use ChatGPT on their own would seem to suggest this as well.
In any case, even supposing that students take these kinds of reflection assignments seriously, then the focus of the work is shifted to the technology itself rather than learning the subject matter. (Although to be fair, if one truly believes that school is just about the generic skills of “learning how to learn, or learning to “think critically,” then it is not overly concerning if students do not acquire much actual knowledge.)
Given the pressure that instructors are under to include AI, it is not difficult to imagine that a student might be given this type of assignment in multiple classes. But how many times can someone be expected to discover that AI is “biased,” or cliché, that it sometimes spits out false information? Who could even blame students for turning to AI itself? In a sense, it almost does not matter whether a reflection is generated by a human or a machine. Even after just a few years, the ideas are so well trodden and the points so cliché that the distinction is nearly moot.
I have come across a few strategies for teaching students to manage the technology in what seems like a genuinely responsible way e.g., having students generate a text and then requiring them to research and validate every claim and reference it contains; however, these tend to be done by professors who have an unusually clear understanding of what LLMs can and cannot do (one was a cognitive scientist who helped develop chatbots), and who have the time and energy to assess responses adequately.
I also wonder whether students who are continually asked to reflect on their AI use will develop an inflated sense of their ability to manage it responsibly, and thus have difficulty perceiving when their actions do in fact cross the line—which can, in any case, always be blurred as circumstances demand.
Considering all this, some educators have posed the seemingly reasonable question of why so many assignments can be completed by AI in the first place. Does this not imply, they wonder, that teachers should focus on even higher-level skills, ones that are beyond the reach of a prompt box?
Daisy Christodoulo, author of Teachers vs Tech, offers a succinct and lucid response:
Of course we want our students to develop the higher order skills of being able to critique writing produced by AI chatbots and to direct the outputs of new technologies. But those skills depend on more fundamental skills and there is no way we can jump ahead to the more advanced skills without acquiring the more basic skills first. In order for students to successfully grapple with problems computers cannot do, they must work through problems that computers can do. If schools could only teach maths using problems that computers cannot solve, we would have to teach six-year-olds maths using problems even top mathematicians find difficult!
So it doesn’t matter if we set our students tasks that can be easily solved by computers. It doesn’t matter if they produce writing that is weaker than that of ChatGPT. The easy problems and the weak writing are milestones on their journey to mastery which cannot be skipped or outsourced.
Reading a text carefully to understand its literal meaning and the author’s intention; summarizing it succinctly; and formulating one’s response it into an outline with a logical structure of ideas—these are the nuts and bolts learning of to think and write analytically. No technology in the world can alter this reality, and pretending otherwise reduces education to a game of charades. Yet these are the exact skills that students are being encouraged to outsource to a machine. The problem is that they require sustained thought and attention and practice, not to mention expert instruction, and are not easily gamified—anathema to a system that prizes instant gratification.
Allow me lay my cards down fully: If an educated adult wants to use AI for a work presentation, or to generate a vacation itinerary, or to send mass announcements, or to avoid composing a thank you note to their mother-and-law, fine. And if a group of scholars want to collaborate with a theater company to try to recreate the creative process of a seventeenth-century playwright, as happened recently in France, then sure… why not? But a kid who spends a good part of their school years practicing generating what a recent New York Times article characterized as “dead but high-flown managerial bureaucratese” (more colloquially known, I believe, as “slop”) in order to be “career-ready” might find that they cannot get a job at all. There is currently a good deal of hand wringing over the scale at which workers will be replaced by AI; however, an equally serious concern is whether an overreliance on LLMs early on will make many adults unqualified for aspects of jobs that cannot be outsourced to technology.
Or worse, those adults will be hired because there is no one else available.
Or even worse, because organizations genuinely do not care whether they are disseminating nonsense, or are happy to disseminate it for the sake of profit, and are happy to hire workers who are themselves utterly indifferent to that state of affairs.
If this is what universities mean by “career readiness,” then we are in very deep trouble indeed.
It seems just as likely to me, though, that the people who are least dependent on AI on as students will ultimately be the most well equipped to navigate new technologies successfully in their professional lives.
Unfortunately, much of the K-12 American education system has, for many years now, been based on the search for a holy grail capable of circumventing the inconvenient fact that the ability to think critically requires access to bodies of factual knowledge stored coherently in one’s long-term memory. (If you don’t know what’s true, you can’t recognize what’s false, and you won’t even know when to be on guard.)
Cited in a recent New York Magazine article, the tech ethicist Brian Patrick Green of Santa Clara University offered the following warning:
“Massive numbers of students are going to emerge from university with degrees, and into the workforce, who are essentially illiterate,” he said. “Both in the literal sense and in the sense of being historically illiterate and having no knowledge of their own culture, much less anyone else’s.” That future may arrive sooner than expected when you consider what a short window college really is. Already, roughly half of all undergrads have never experienced college without easy access to generative AI. “We’re talking about an entire generation of learning perhaps significantly undermined here” […] “It’s short-circuiting the learning process, and it’s happening fast.”
But despite the ever more destructive consequences, ChatGPT and its ilk are, in for a certain type of educator, a godsend: they allow for an even deeper illusion that students can do high-level work when in fact they increasingly lack the most elementary abilities. And even if teachers do recognize how far behind their students are, they may find themselves under intense pressure from administrators to accept what is clearly bot-generated work.
To be frank, I cannot fathom trying to remain a holdout when one’s colleagues are jumping on the bandwagon, and administrators are constantly demanding to know what one is doing to incorporate AI into one’s teaching, regardless of whether it is needed or provides any genuine advantage over older tools. In the humanities at least, AI strikes me as a solution in search of a problem (the problem, of course, being that students are doing their own writing).
Perhaps I’m missing something, but the overwhelming argument seems to be “We have to let kids use AI, no matter the consequences, because it exists and can do cool stuff, and because it might be able to do some kind of (always unspecified) even cooler stuff in the future,” which is—not to put too fine a spin on it—basically the dumbest justification imaginable. Nor is the assertion that the genie can’t be stuffed back in the bottle, at least partially, particularly convincing; regulations have been retroactively imposed on plenty of harmful inventions.
Writing in the Chronicle of Higher Education, Dan Sarofian-Butin summarizes the impossible situation into which the professoriate has been placed:
[W]e could sheepishly accept a shared capitulation to a logic of convenience, embracing the performative spectacle of “entertainment.” The problem, of course, is that we all know where that road leads, as higher education lurches ever more steadily toward the transactional horizon, and what we do — the difficult yet rewarding friction of teaching and learning — becomes yet another far-fetched story grandparents tell youngsters: I walked uphill to school, both ways, in a snowstorm … and then, yes, I did, I had to read a book and write a six-page paper about it.
Or we could accept, after two long years, that we are in the midst of a crisis of purpose as ChatGPT strips away — efficiently, effortlessly, instantaneously — the heart and soul of what makes teaching and learning real…
There is no happy ending here. A disengagement spiral is upon us as AI supercharges students’ disinterest in learning and faculty disinvestment from teaching.
As if to underline Sarofian-Butin’s point, the Chronicle of Higher ad site is cluttered with every sort of ad for AI imaginable.
I am not normally a fan of slippery slope arguments, but in this case I think one is justified.
Even if students are not cheating in the sense of submitting 100% chatbot-produced texts, the gray area into which much LLM usage falls is vast, easily interpreted to extend beyond its intended confines, and impossible to control. Regardless of how many conversations about AI students are corralled into, how many reflection papers they are asked to write, or how conflicted they feel, the vast majority will willingly turn themselves over to the technology because it is readily available; because it makes their lives easier; and because there are unlikely to be any real consequences for doing so.
Yes, some students who are already working at a high level will find ways to use LLMs to further deepen their understanding, and a small percentage of less accomplished ones remain determined to do the work on their own; however, they will be the exception, not the rule. Meanwhile, weaker students grow more and more dependent on technology to hide their lack of skills. is this what is meant by “leveling the playing field”?
To be clear, however, I am not in any way suggesting that the existence of LLMs should be ignored.
By all means, explain how the algorithms work and how statistical prediction models differ from human thought; demonstrate when they are and are not useful for academic work; and discuss the myriad ethical, social, economic, political, and environmental issues involved.
But to embrace what is essentially a therapeutic model, whereby students are merely “encouraged” to develop the correct feelings toward AI usage in the hope that their behavior will follow (thereby sparing the adults of the responsibility for setting limits), is to court academic disaster.
As things currently stand, it is—circumstances permitting—absurd to assess students on anything other than responses they have written by hand or given orally, on questions they have never seen before, in the presence of an attentive supervising scholar. If they find that AI helps them prepare effectively, they can use it at home. Of course there will be cheating attempts, but the scale of the problem will be immensely reduced. If this is the only way to ensure that students are doing their own writing and thinking, then the fact that they may not “recommend” the experience is utterly beside the point. At the risk of sounding like an old fogey, 20-year olds, even very bright ones, do not have the perspective or life experience to grasp what they may be sacrificing otherwise.
Perhaps, as Carla Arnell of Lake Forest College proposes, universities will eventually create dedicated internet-free writing labs where students can develop their skills over the longer term without the constant temptation of digital shortcuts. For the time being, however, trying to stay ahead of the technology is a game of whack-a-mole, with instructors responding by creating ever more elaborate assignments and rubrics designed to either incorporate LLMs in ways that they hope might aid learning, or penalize their likely use without having to explicitly call students out for plagiarism.
How exhausting this must all be! The goal should be to ensure that students acquire a solid baseline level of knowledge in the necessary material, not for the professoriate to twist itself in knots trying to hold onto forms of assessment that still contain enough loopholes to let the learning process get short-circuited. There is, after all, no commandment that states, Thou must assign at-home essays!
I cannot speak for the entire world, but plenty of European universities have continued to base grades either primarily or exclusively on in-class exams all along, for the simple reason that it has never been possible to know how much of a student’s independent work is truly theirs. In contrast, for the past fifty years at least, U.S. universities, with their much heavier embrace of papers and projects, have willingly allowed for a certain amount of exploitation of trust, in exchange for the privilege of boasting that they—unlike institutions elsewhere, which just teach “rote learning”—develop students’ creativity and “critical thinking” skills.
That bargain is clearly broken now.
The problem with a straightforward and relatively easy-to-implement response—beyond the fact that it requires instructors to go against inculcated pedagogical doctrine and accept a type of authority that many of them have been taught to feel uncomfortable with—is that attempts to impose hard guardrails against reliance on AI are likely to run up against the corporate-university-customer-is-always-right model. (NB: I am not even going to attempt to address the issue of cheating in online classes, which is a problem of an entirely different magnitude.)
If students enter higher education with severely underdeveloped academic skills and are already reliant on LLMs for basic tasks, then requiring them to work without technological crutches is likely to produce intense backlash. And with yearly costs approaching $100K at some private institutions, students—and their parents—expect to be given a good amount of leeway to do as they wish. Given that, it is easier for the bureaucratic university to simply appoint committees that engage in endless, unresolved debate while dodging the uncomfortable question of whether college students do in fact need to write their own papers.
In trying to salvage their bottom lines, however, universities may find themselves sacrificing whatever shreds remain of their authority, not to mention their souls. No matter how unjustified the accusation may be, it is already a running joke that a humanities degree qualifies one for nothing more than asking, “And will that oat milk latte be a grande or a venti?” What will a $400K degree be worth when university graduates can no longer be assumed to possess even functional literacy? Not everyone can go be an influencer. Predictions that the higher education bubble is about to burst have been coming for a long time. And when it finally does, the “pop” may be very loud indeed.