The End of the Take-Home Essay?

Until this summer, I had avoided getting worked up about ChatGPT. My position, prompted by this Columbia undergraduate’s article in The Chronicle, was that if students know enough about writing essays to make ChatGPT work for them without detection by a minimally alert instructor, they have probably mastered the essentials of the art of essay-writing more than the author of that piece seems to realize. I could rest easy in the knowledge that, at least, I wasn’t not teaching my students what they needed to learn how to do.

But a more recent Chronicle article, by a Harvard undergraduate, made me think again. It showed that GPT-4 — the latest iteration of OpenAI’s GPT software, as yet available only to paid subscribers but likely to become more broadly accessible soon — could write plausible papers on a range of subjects, from Latin American history to Proust’s In Search of Lost Time, that earned A’s and B’s from qualified instructors.

Rattled, I asked my daughter, a rising sophomore in high school, to run through ChatGPT several take-home essay questions that I had assigned to students this past year in a course I taught on politics and literature.

The initial results ChatGPT produced didn’t alarm me. They were well written and structured. There were none of the usual flaws one finds in student writing: Each sentence logically followed the other, paragraphs had points, transitions were purposeful and clarifying. But all the answers lacked a thesis, provided needless exposition, referred to texts we didn’t read for class, and made basic errors about those we did. One answer identified Ralph Ellison’s “The World and the Jug” as a short story rather than an essay. Another — in response to a question asking students to apply a specific statement from Bertolt Brecht’s “Short Organum for the Theatre” to Aeschylus’ Oresteia — took me through a long detour about the alienation effect before depositing me in a dust storm about religion and society. If a student produced such an essay on her own, she’d receive a grade of C or lower purely on the basis of content and execution, leaving her with little incentive to use ChatGPT.

Then my daughter began refining her inputs, putting in more parameters and prompts. In response to the question, “How might the person you imagine yourself to be in 30 years respond differently to The Bacchae from the person you are now?” the chatbot had initially produced some vague, inconsequential statements about future selves. My daughter now instructed it to imagine the author as “someone who struggled in their youth with their gender identity but now, as an older adult, does not,” and to cite specific examples from the play. ChatGPT now generated a series of close studies of how a younger, less settled self might take an absolutist stance, pro or con, on the gender-bending of Dionysus. An older person might treat the desperate gender absolutism of Pentheus with compassion, yet still see in him a cautionary tale about the anxious intolerance of youth and inexperience. My daughter pressed on, asking ChatGPT to formulate a thesis from these examples, which led to a statement about the relationship between aging, gender identity, and the experience of reading.

With each round and just a little nudging, the machine’s essay got more pointed and specific. By its final iteration, the essay read like the well-wrought response of a concrete self, with a particular biography, to the ongoing reading of an ancient text across the course of a lifetime. It was as good as what many of my students produce after a semester of effort.

Where I had initially thought that students would have had to master many of the skills essential to essay-writing in order to use ChatGPT effectively, it’s clear to me now that that’s not the case. They need only to be able to identify the difference between good work and not-as-good work, which even students who struggle academically can do. I’ve often been surprised that students who have a difficult time writing a thesis statement can spot it a mile away in another student’s essay. That doesn’t mean they can write one themselves.

And now, thanks to ChatGPT, they don’t have to.

Why do we ask students to write papers, in the first place? Why do we grade them?

A text — of political theory, history, fiction — presents us with a created, often alien, world. Writing about that world offers us an opportunity to struggle through it and render it meaningful, to us. Students might take such a world apart, or pitch a tent in it. They might do something else entirely. One way or another, they make sense of that other world — and learn, I hope, that that meaning is also a creation: their creation.

None of this is easy. Through requiring students to write multiple drafts, intensive comments on each draft, ongoing revision, and conversation, I teach them that all writing is rewriting, and good work is just that: work.

Academic writing has never simply been about producing good papers. It’s about ordering one’s world, taking the confusion that confronts us and turning it into something intelligible, wresting coherence from chaos. And knowing that that doesn’t happen spontaneously or instinctively.

That’s not a skill for college only. It’s a lifelong practice. Being able to see a situation, picking out those elements that matter and lend it significance, bringing clarity to obscurity: These are what good readers and good writers do. They’re what good friends, good parents, and good citizens do, too.

But how we perform those tasks as writers is distinctive. Where the expression of thought in speech is ongoing, allowing us to modify and modulate our point in response to that of our interlocutor, writing requires objectification. Even a rough draft freezes our thought in time. Objectifying our thoughts, putting them on paper or on a screen, enables us to look at them from a distance, as if they were not our own. Distance allows us to see what in our thoughts is unformed, not yet whole, in need of further revision.

In conversation, we’re forever slipping the noose of that permanence. We say to our interlocutors, “That’s not what I meant, I mean this.” On paper, we’re denied that escape. That’s what makes writing, and rewriting, such a distinctive experience — and opportunity.

The only thing, in my life, that has come close to the kind of mental objectification that writing forces me to do is psychoanalysis: the old-fashioned kind, five days a week on the couch, with your analyst behind you saying almost nothing. Only on the couch have I been led to externalize myself, to throw my thoughts and feelings onto a screen, to see them as other, from a distance, the way I do when I write.

Writing is not therapy, but it pushes for a discovery of self in a similar way. It requires you to make your fleeting thought a hard fact in the world, and to make yourself answerable for that fact. If you can’t do that, you need to revise your thought and find another fact, one that you can live with as your own.

The work I’ve described is difficult. It doesn’t come naturally to most of us. It’s effortful. It’s frustrating. It’s disappointing. Failure looms large. We need incentives to do it. All of us are vulnerable to shortcuts and escape hatches.

When confronted with frustration or failure, I recoil. I procrastinate. I surf the web, I lie on the couch, I respond to emails, I stare out the window. I do everything but the work that I need to do. This seems normal to me. From the many conversations I’ve had with students, it’s normal to them, too.

How do we deal with that? Some part of it is working through, getting help, breaking things down into smaller parts, developing resilience. It requires a patience with your own crap, as Philip Roth once put it. And some part of it involves deadlines and expectations from other people and, yes, the sanction of their negative judgment.

Many of my fellow academics are uncomfortable exercising this sanction, or acknowledging that it’s part of their job. They wonder why we need grades at all. When I listen to them talk about these issues, I hear about a classroom that works solely through the pleasure of positive inducement, the inspiration of the instructor, the well-designed assignment. I hear about a form of education that elicits and responds only to curiosity and desire. There is even a strange self-flagellation among instructors who feel they have to be the consummate entertainer, the most understanding parent, the most empathetic friend, before they can ask anything of, or expect anything from, their students. If the students are cheating, it’s because we’ve failed. The solution must be that we need to work harder.

In my experience, that’s a fantasy that serves neither student nor instructor. There is nothing in the realm of work — no matter how interesting or exciting or desired — that does not entail, at some point, the experience of frustration, self-doubt, loneliness, and anxiety. Experiences that most of us (realistically, all of us) flee from, especially when we’re by ourselves, without the helping hand or reassurance or conversational ease of another. Some professors think that it’s the pressure and “high stakes” of our grading and assessment regimes that produce those feelings of discomfort in our students. I think it’s intrinsic to the work, if you’re doing it right.

Our goal shouldn’t be to eliminate this discomfort. We need to teach students that it’s part of the process, and develop strategies for coping with it. But for students to really get that — to believe it, to feel it — they have to do the work. They have to go through that process in order to learn that they can’t run from it, or outsource it to AI, and, more important, that they don’t need to run from it.

Sometimes — often, in my experience — it’s the shock of a bad grade that gets students to start doing the work in a way that leads to the pleasure of self-knowledge and discovery, of seeing inchoate thoughts turn into clear ideas.

If students know they can get a higher grade using GPT-4, how many of them will ever reach that moment?

What is to be done?

Some professors think we should carry on as before, assigning take-home essays but writing ever-more-artful questions that GPT-4 or some other form of AI can’t answer. I’m dubious whether that can be done. Even if it could, I’m not interested in making my questions elusive enough to outsmart a machine. At a certain point, it risks demanding too much of the students: expecting a superhuman effort on their part, just for the sake of proving their humanity.

Others say we need to redesign our writing assignments, scaffolding the effort from start to finish and assessing students on each step, from taking notes and making outlines to revising and polishing a finished piece. I’ve done that kind of work in previous classes, and my students tell me they’ve been doing it since middle school. Both the students and I found it artificial and alienating. Not to mention infantilizing.

Still others say it’s not our problem. We’re teachers, not cops. If students don’t want to do the work, that’s their decision. But ChatGPT has made the choice over whether to cheat more fraught. For this generation of students and the ones that will follow, the temptation is greater than ever. Moreover, if using ChatGPT to write essays becomes widespread, those students who elect not to use it, who prefer to do the work themselves, may suffer a penalty for doing so. We are on the verge of a global academic collective-action problem.

As for professors, we shouldn’t let our fear of being cops prevent us from being good teachers. The issue is not punishment but pedagogy. Unlike policing, teaching is a two-way street. To throw myself into my students’ work, I need to know that they’re willing to do the work. But neither of us can know that, for certain, until we’re doing the work, together. Simply leaving it up to students to decide whether they’re going to do the work, without further comment or intervention or negative sanction from me, is a failure of pedagogy. Classroom liberationists may style their laissez-faire policy as an embrace of student autonomy. To me, it looks like a refusal to see students as they are, as we all are: fallible people capable of growing, reflecting, and remaking themselves.

When I return to the classroom this fall, I’m going to do something I’ve never done in my 30 years of teaching. Instead of take-home essays, I’ll be requiring in-class writing, including midterms and finals.

I dread this decision. It decreases the amount of class time spent discussing the texts we’re reading, of which there is already never enough. It eliminates the elements of process and revision that are so important to writing. I hope to re-incorporate those elements, but I’m not yet confident or clear about how.

The truth is, I don’t know what is to be done. Some days, I feel like giving up. But as that’s not an option I leave for my students, it’s not an option I can take for myself.

So we’ll approach this situation the way we approach our writing: as a challenge, an experiment, a draft, knowing that we can, and must, eventually revise.