Edward Dunsworth
Two approaches dominate discussion about how professors should handle generative “artificial intelligence” in the classroom: give up or give in.
Give up. Faced with a powerful new technology custom-cut for cheating, many professors are throwing up their hands in despair. This was the dominant mood of last month’s widely shared New York Magazine article. “Everyone is cheating their way through college: ChatGPT has unraveled the entire academic project,” its doomsday headline proclaimed. The article paints a depressing picture: students using AI to cheat, constantly and without compunction; professors out of ideas for how to deal with it. “Every time I talk to a colleague about this, the same thing comes up: retirement,” one professor told author James D. Walsh. “When can I retire? When can I get out of this? That’s what we’re all thinking now.”
Give in. A second response has been to surrender to the techno-hype of ChatGPT, to embrace generative AI as a teaching tool. “It’s an opportunity to open the door of creativity in the classroom,” gushed historian Jo Guldi in a 2024 interview, “and simultaneously raise the bar for the quality of the work we expect from our students.” Professors are encouraging students to use AI software not just for rote tasks like transcription and data compilation, but for more cerebral activities like brainstorming, analysis, and even writing. Mark Humphries, who has led the pro-AI charge among Canadian historians, boldly declared in a February article that, with increasing AI use among students, “poorly crafted theses, unsupported arguments, and narrative papers without an argument should become a thing of the past.”
I reject both approaches.[1] Not because I don’t appreciate the revolutionary challenge that generative AI poses to humanities and social sciences education, and to our society at large, but precisely because of it.[2] At this worrying juncture, as multitudes – on campus and off – cede ever more of their thinking and writing to computer programs, historians and other humanistic intellectuals should not be shying away from the challenge, but rising to it. We know (or should know) the value of deep thinking, of labouring through complex research and writing projects. We have (or should have) an inkling of what students are losing when they skip over these tasks. Rather than giving up or giving in, we should be standing up and speaking up. For our students, for our craft, and for quaint human practices like thinking and writing.
To do so first requires some precision as to what exactly we are talking about when we talk about “AI,” and which activities under that vast umbrella I am signalling for concern. “To put it bluntly,” write Emily M. Bender and Alex Hanna in their new book, The AI Con, “AI is a marketing term. It doesn’t refer to a coherent set of technologies. Instead, the phrase ‘artificial intelligence’ is deployed when the people building or selling a particular set of technologies will profit from getting others to believe that their technology is similar to humans.” All sorts of computerized tasks are called “AI” nowadays: from transcription and translation to facial recognition to text- and image-generation. Part of the challenge of confronting the AI boosters, in academia or elsewhere, is that they have co-opted seemingly the entire world of computing under the banner of AI. This allows them to paint critics of ChatGPT and the like as paranoid technophobes, robe-clad and sequestered, studying ancient manuscripts by candlelight. (Actually, that sounds pretty nice. But I digress). So let me be precise. What I wish to critique here is not the use of software (you can call it “AI” if you like) for tasks like transcription, or keyword searches, or spell-check, or the extrapolation of numbers from a text to a spreadsheet (assuming it works). There certainly are critiques to be made of these technologies, and there has indeed been robust debate and discussion about them for many years. But those are not my target. Instead, I take aim specifically at the use of generative AI tools like ChatGPT to replace the labour of thinking and writing.
*****
Hip professors everywhere are clamouring onto the gen-AI bandwagon. Jo Guldi, the professor excited about ChatGPT’s turbocharging of student creativity, was one of seven historians interviewed in a December 2024 episode of the American Historical Association’s podcast, History in Focus, on generative AI and the history classroom. Three of them were boosterish. Thinking about students “who are struggling to articulate how they want to do that creative project,” who “have ideas” that aren’t “fully formed,” historian and digital humanities scholar Kelani Craig averred that “having a chat partner that isn’t judgmental is a really valuable way for them to play with those ideas.” Then, they can “take them out of ChatGPT and really apply their own thinking.”
It’s not just the profs. Seemingly everybody is now using generative AI in their day-to-day work: lawyers, school teachers, administrators, trade unionists, non-profit staffers – these are just examples from personal interactions. And, of course, students.
But I am convinced that much is being lost in this rush to outsource our writing – much more than we can even know.
*****
Writing is thinking. It’s a writer’s cliché, but a good one. When you first conceive a lecture, dissertation chapter, a book, even an article for Active History, everything seems so straightforward. “This one will be a cinch,” you think. “Two days, tops.” Once you actually sit down to write, that boundless optimism meets an unceremonious death. Writing is hard. It’s painful. To write is to submit yourself to seasons of self-doubt. The ideas seemed so natural and free-flowing in your head. Now you get to the page. And what comes out is jilted, ham-fisted, and awkward. You are again and again confronted with nagging questions. What am I even trying to say? Who even cares? Why even bother?[3]
Given the nature of this experience, it is not surprising that people would jump at the opportunity to skip the pain and get right to the end. Or at least to an end. Because any writer can tell you that the end is determined by the path taken to get there. Rare is the piece that gets written exactly how it was outlined. Why? Because writing is thinking.
On the page, you are confronted with the shortcomings in your thinking, the gaps in your evidence, the flaws in your argument. You become aware of important matters that you had not considered. The process of writing is the process of working through those problems, of figuring out the best way to express whatever you are trying to express. Onerous as it may be, the result is gratifying, even if fleetingly so.
It’s gratifying because you produced a finished product, yes – and hopefully one that you are proud of. But a deeper gratification surely comes from having solved a problem (or, more accurately, dozens of interlocking ones). The pleasure of solving the Rubik’s Cube is not so much about the six shining, monochromatic sides. It’s the pride in having worked your way there. (I’m surmising; I have tried and failed to solve a Rubik’s Cube. I’d better stick to writing.)
If writing is thinking, it follows that outsourcing writing is outsourcing thinking. And this outsourcing is accelerating at a staggering pace. This should concern us all: as educators, but also as citizens.
The point of our teaching is not to produce the end results of essays and podcasts and scrawled-in exam booklets. “We do not ask students to write a ten-page essay on the Peace of Westphalia because there’s a worldwide shortage of such essays,” Irina Dumitrescu sagely observed in a Walrus essay about ChatGPT and the teaching of writing. The point of these assignments is the process of completing them.
The skills one develops through the complex research, thinking, and writing that essays entail are skills that carry over to countless aspects of life. They are skills of problem solving. Life, as it turns out, is one goddamn problem after another.
Where should I go to university? How should I vote in the election? What can I do about my negligent landlord? How can my team develop a better communications strategy for our company? How should our union approach the coming bargaining session? These are all questions that ChatGPT would happily answer. But will the answers set the asker up for success? When the union reps with ChatGPT-derived bargaining goals face the employer’s shark-toothed lawyers in the boardroom, how will they fare?
Part of our task in the face of generative AI is to make an argument for the value of thinking – laboured, painful, frustrating thinking. It is not an easy sell. But to give up on this is to give up on our students, most of whom are at an age where they can be easily seduced by techno-sirens promising instantaneous essays for minimal effort and with little chance of getting caught. They deserve better from us.
Is it the end of the world if a student has not fully thought through the outcomes of the Peace of Westphalia, or the Haitian Revolution, or Canadian Confederation? No, of course not. But the student skipping those steps is costing themselves experience and expertise in problem solving, crucial abilities for living, not to mention working or studying.
*****
I am amazed by how eager some of us historians seem to chuck out the skills and knowledge we have honed individually and collectively over decades. Do we hold ourselves and our abilities in such low esteem? Mark Humphries declares that advanced models of ChatGPT are “about as competent on tasks like document analysis, historical interpretation, and literature reviews as a good PhD candidate.” This sounds impressive until you remember that this software does not “analyze” or “interpret” or “review” anything. It mines a lot of text and responds to a prompt by predicting what word should come next. It is able to produce a convincing historiographical essay because it is combing through dozens of such texts and mimicking their tone, structure, and content. In other words, it is able to pass as a PhD student because it is regurgitating the past labour of (real) PhD students and others. Humphries is similarly willing to hand over editorial labour to the computers. He recommends that students use ChatGPT to edit their assignments, praising its ability to explain the reasons behind edits. “To my mind, this is no different than encouraging students to go to the writing centre.” What Humphries and others on his wavelength miss is that editing is not a mechanical process. It is not all about the end results. The relationship between editor and writer is a human one, a social one. When done well, the relationship results not only in a better text, but in learning. Discussing one’s writing with another human being opens one up to criticism and feedback. It is difficult and uncomfortable. These are important, formative experiences for students to have.
Perhaps we’ve been beaten down by the relentless neoliberalization of academia. How many of us have bought into the drive to produce, the CV-building and annual reports, the institutional culture that cares infinitely more about the accumulation of accomplishments – articles, books, grants – than about the quality or content of those endeavours?
No wonder we are so easily intoxicated by a writing machine. Imagine how many more unread articles we can produce now!
But just as the value of undergraduate assignments is not in the alleviation of essay shortages, the value of our scholarship lies not in its outputs. We are not widget producers. Or at least, we should not be. We are producers of knowledge, of analysis, of ways of understanding the world. And we are teachers. We teach history, politics, sociology – sure. But at a more elemental level, we teach our students how to think and how to write.
*****
I understand the frustration of the profs dreaming of retirement. I don’t wish to shame them. What professor doesn’t dream of retirement in March and April, as we limp into the home stretch of an eight-month marathon, a mountain of essays and exams left to climb before the finish line?
But to give up is also to give in, to buy into the hype that ChatGPT and its ilk could ever replace writing or thinking. It cannot. If anyone should know this, it’s scholars in the arts and humanities.
But the “give up” position is simply unacceptable. Generative AI presents a serious challenge to education – perhaps especially in writing disciplines like history, a challenge I blithely dismissed two years ago on this site. I recognize the reality now.
Educators and historians are perfectly positioned to intervene in this moment of profound challenge. We can do so by standing up for writing and thinking. And by standing up for our students, who deserve the chance to reach their full potential as thinkers, writers, and problem solvers. The moment calls not for retirement, retreat, surrender. The moment calls for action.
What does action mean? In brief, we need to take the fight to many fronts. Undoubtedly, we need to change our methods of evaluation, a subject that many thoughtful teachers are addressing.[4] But we also need to hold our institutions accountable. Just as 19-year-olds are easily lured by automatic essays, university administrators are highly susceptible to the temptations of technology-driven downsizing, big tech donations, and the appearance of being on the cutting edge. Scholars in the humanities must serve as the adults in the room to both groups. And we need to take our arguments out of the academy, into the public sphere and on to the halls of government, whose occupants are also caught up in the AI hype. It is a tall order. It would be easy to give up or give in. We need to stand up instead.
Edward Dunsworth teaches history at McGill University and is a member of Active History’s editorial collective. He thanks Mack Penner and Tom Fraser for comments on this piece.
[1] “Give up” and “give in” are by no means the only public responses by professors, even if they are the most prominent. There are plenty of critical voices, some of whom are cited here. And many scholars have articulated nuanced positions.
[2] This is a challenge I grossly underestimated in a 2023 article on this site. I stand by most of what I said in that article, but I was obviously dead wrong about the technology’s potential for cheating, which I downplayed.
[3] My comments on writing are indebted to William Zinsser’s book On Writing Well and John McPhee’s New Yorker column, “The Writing Life.”
[4] See, for example, comments by Katharina Matro, Johann Neem, Kevin Gannon in the same American Historical Association podcast cited previously. There are countless articles and resources online about creating AI-proof assignments and other questions of student evaluation.
I posted this to Bluesky the other day:
Klein has a nearly insufferable self regard in this kind of interview…. But I think he articulates what we need to try and teach our students. Short cuts with AI will not lead to original thinking. If you want to stand out, you need to do the work. Prohibition isn’t going to work and a large number of first years student will choose the easy option. But over four years, we need to build a culture of striving for original writing with a distinct voice that will set our students apart from anyone with a high school degree and an AI subscription. https://youtu.be/smb7hy6KufQ
So I think we’re converging in our thinking on Gen AI or maybe always shared a fundamental agreement. I’m a little less critical of Jo and Mark, as I think there is still room to use Gen AI tools as effective copy editors. My experience as a dyslexic makes this option really valuable. And they probably have a role in brainstorming ideas. But the second you ask them to do the reading or drafting the core prose, you’re giving up on the hard part where we learn. So the question remains, how do we convince students to do the hard part when the temptation will be ever present and our efforts at prohibition will fail (the new models don’t make up fake citations nearly as frequently as they did in 2024).
Tools like ChatGPT are privately owned and have no clear path to profitability unless they substantially increase subscription prices. We cannot assume they will always be available or affordable. Perhaps there is value in learning how to prompt large language models well, but we should develop those skills in parallel with research, writing, and analysis so we aren’t left with nothing when AI companies close up shop or demand astronomical fees.
In my experience as a high school history teacher (who taught university history in the past), AI is mainly being used to cheat. The cheaters are in a minority at my school, and I’m fortunate that I don’t have classes of several hundred or more, so I can take the time to catch the miscreants. Having my students work in Google Docs (for essays) is very helpful because I can follow their edit history with the help of AI detection programs like PassedAI (which runs in the background of a Google Doc). Another helpful program is Google Draftback, although the creator now charges for this. It shows the number of revisions (or keystrokes) a person made to create a document. I’ve had students hand in 5- and 6-page essays that were completed with only 250 keystrokes (quite an accomplishment). I’ve had students write five-page essays in five minutes—a truly remarkable accomplishment that the student typically cannot account for when I speak with them.
Students are also using it to summarize their sources. I use the above programs to give students low marks or zeroes on their research notes if they can’t substantiate they did their work. I encourage students not to do this and explain how reading your sources allows them to make their connections. I use my research into HBC post records as an example. Also, showing Grade Nine students nineteenth-century cursive writing blows their minds.
Universities have, to some extent, dug their own hole in this problem. Classes have become ridiculously large, and grading and teaching are often handed off to teaching assistants and part-timers for these large classes. This problem stems from a lack of funding, which has worsened over the last few years. As a result, it is nearly impossible for anyone to follow up on student cheating in classes with hundreds of students.
When I hear people discuss AI’s potential to assist students, I agree. However, it has incredible potential to be misused, and there is currently no effective way to deal with this without a significant amount of time and effort as professors attempt to grade hundreds of essays that were largely submitted in the same week. We are, I think, at the same point we were twenty-five years ago, with students beginning to cut and paste from websites. Turnitin.com helped to reduce that problem. Hopefully, some very skilled programmers or coders will solve our current dilemma.
This is a thoughtful and provocative post, Edward. Thanks for writing it. The conversations around “AI” (great nod to Bender and Hanna, as AI is a marketing term that I suspect will give way again to “machine learning” or “natural language processing” as the hype cycle matures) are important ones. Posts like this help push us beyond the constant drumbeat of AI boosterism. If anything, I think we have set up a technology that will never meet the expectations of its boosters nor the nightmare-ish visions of its detractors.
That said, I think there’s a more productive path forward. This is one that involves taking seriously the perspectives of those you are quick to dismiss. A hand-wave dismissal of scholars like Jo Guldi and Kalani Craig as “hip professors” overlooks their substantial and long-standing contributions to digital history, much of it well before the rise of large language models, not to mention their own contributions to their critical fields of history.
Generative AI tools have clear strengths and serious limitations. Critical, informed engagement, rather than polemic, is what we need most. We all benefit from open and respectful dialogue that tries to understand each other.
Finally, lines like “scholars in the humanities must serve as the adults in the room” are not productive. If anything, they risk alienating those we need most to work with.
David Calverley, these tools are super attractive as they give us hope that we can hold on to tried and tested assignment. But they are not going to work. It is not hard to get a computer to mimic the key strokes and edits of a human in a Google Doc. There are start ups currently being created by undergraduate students to share tools to work around any AI blocking attempts we create. So we can catch the lazy students who panic and use AI 5 minutes before the deadline, but our success rate will be very low with students who do a little research. In fact, it took me 3 minutes with Google’s Gemini 2.5 pro to come up with a few options:
Here is a summary generated by a Google AI on why process-based AI detection is easily circumvented:
A student looking to bypass tools that track edit history, keystrokes, or timing (like Draftback) doesn’t need to be a technical genius. They only need to avoid the lazy method of direct copy-pasting.
The key is to simulate a human writing process. The primary methods are:
Manual Re-typing: A student generates the essay with an AI in a separate document, then manually types it into the Google Doc. This simple act defeats keystroke counters and edit history analysis by creating a seemingly authentic digital footprint.
Paced Work: Instead of typing it all at once, they spread the work over several days to mimic a natural writing schedule, thus neutralizing suspicion based on rapid completion times.
Laundering Sources: To fake research notes, a student can have one AI summarize a source, then use a second AI or paraphrasing tool to rewrite that summary. The final text appears original, even if the student’s understanding is superficial.
The conclusion is that these detection tools only catch the laziest cheaters. A slightly more dedicated student can easily fabricate a “perfect” writing process that is indistinguishable from honest work. This suggests that the arms race of detection is a losing battle, and the more effective solution is to evolve pedagogical strategies—such as oral defenses, in-class assignments, and prompts requiring personal reflection—that assess genuine student understanding, something an AI cannot fake.
[back to Jim: I tested getting AI to write a genuine student reflection my winter term course and it aced the assignment. The tools are improving so fast that even they can’t keep up.]
Jim and Sara, thanks for your comments.
David, thanks as well for sharing your perspective as a high school teacher and those tactics. That was informative.
Ian, thanks for your comments. Yes, my piece is a polemic. Diplomatic dialogue has its place. My opinion is that the situation, in which vocal critics like me are in a tiny minority (speaking on a societal level), calls for a stronger response.
Regarding university administrators: I work at a university whose uppermost administrators are openly musing about using “AI” to reach greater efficiencies (read: fire people). Other universities are embracing “AI” with open arms, with the California State University system being perhaps the most extreme example. Are all university administrators jumping on the AI hype train? Of course not. But people in power need to have their feet held to the fire from time to time (or perhaps more frequently than that). Those to whom the comment does not apply need not concern themselves with it.