Great, AI can write essays. Can we stop asking students to do useless writing assignments?

We were never that good at teaching the fundamentals of writing to begin with; this is an opportunity to rethink them.

Oct 13, 2023

For most of educational history, essays have been a big part of evaluating student learning. But now, with technologies like ChatGPT, for the first time those essays can be manufactured automatically at little or no cost to the student. This has thrown the entire system of writing and grading into chaos. Yet, the question most educators and writers seem to be asking is how we can preserve the original system in the face of this new technology. Another way to look at it is that ChatGPT has revealed the flaws in the system. For years, we’ve been asking students to write essays that elicit generic, unreflective responses. No one seems to think ChatGPT produces the kind of writing anyone wants to read. And yet that seems to be the standard we ask students to meet. Maybe the problem isn’t the new tech; it’s the old assignments.

Mere coherence as the sole measure of competent writing

I went to a high school that prided itself on teaching its students to write well. At the very least, they would tell students and parents, when you graduate from here you’ll know how to write. I always found this claim amusing. When I graduated high school what I had learned was not how to write an essay. It was how to write an essay that teachers would give an A. These were not at all the same thing.

It was clear to me that these were different things because I never did the assigned readings. I would skim the relevant SparkNotes, pull a few key quotes, and make connections between plot points, quotes, and general human themes. Teachers seemed satisfied with this approach, and for the most part I got As. But it was clear to me I wasn’t learning to write. I was learning to bullshit.

If a resource like ChatGPT had been available when I was a student, I don’t honestly know if I would have used it. Maybe. But regardless of what my personal choices would have been, the fact is that ChatGPT has made the cost of composing bullshit—the kind of externally coherent but intellectual hollow work I specialized in as a youth—fall to zero.

And as a result, the educational pipeline is now facing a system failure. Teachers and professors are concerned they can no longer assign essays in the way they used to. For example, in a December 2022 article from The Atlantic, the author presents a ChatGPT passage which she had been sent by a professor friend. The professor considered the passage “graduate level.” The author herself gave it a B+; she writes, “The passage reads like filler, but so do most student essays.”

The author goes on to cite another professor, one who teaches at a business school: “The OpenAI chat is frankly better than the average MBA at this point.”

The truth is, most student essays do read like filler. And yet, when teachers encounter a computer program capable of producing filler, their instinct is to question the technology and not what they’ve been asking their students to do this whole time. I don’t blame anyone personally for feeling this as a gut reaction. This is exactly the kind of dramatic revision of the status quo which is hardest for people to deal with. But at some point we need to pose the uncomfortable question. If the prompts we’ve been assigning produce writing that’s indistinguishable from filler, doesn’t that mean we should be rethinking the prompts?

As it currently stands (and this is definitely subject to change), what defines the prose of ChatGPT and its peers is that it is generic. It gives generic answers. It writes in a generic style. It is a statistical model of baseline, generic human writing—the idealized Wikipedia voice from nowhere in particular. When we say that writing sounds like “filler” what we mean is that it meets a baseline standard of coherence without going beyond these kind of generic, exactly-middle-of-the-road insights.

This is just a more technologically sophisticated version of what I was doing in high school. And reflecting on my own experience, it makes me suspect that, perhaps, the system isn’t worth preserving. Certainly not at all costs. Instead of asking how we can perpetuate the existing system for assigning and evaluating essays in light of this new technology, we should use this as an opportunity rethink what exactly it is we are trying to teach when we teach writing.

But students have to walk before they can run—don’t they?

The number one argument I can imagine people having against this perspective is that the current writing system is about teaching the fundamentals of writing. Sure, a lot of what students produce is generic and asks them to do little more than produce a grammatical, logically coherent essay. But students need to get that under their belt before their can tackle more advanced aspects of writing. You need to walk before you can run.

But while I agree that some fundamentals are important, I don’t agree that they’re the ones we’re currently teaching.

The point we have to grapple with is that we’re not currently minting strong writers. Academics are a case in point. Anyone with a PhD has gone through pretty much the full educational pipeline. They have successfully completed uncountably many writing assignments. But when is the last time you tried to read an article written by a PhD for a group of their peers? In most cases, it would be a blessing if it had been written by ChatGPT.

And it’s not just academics. Written communication in government or business is rarely as effective as it should be. This is understandable. It is genuinely hard to say what you want to say in written form. It doesn’t come naturally to us as humans. But I think we need to be honest about that and admit that there’s probably a lot of room for improvement.

So what are the fundamentals we’re currently attempting to teach? In light of ChatGPT is it worth teaching them in the same way we’ve taught them so far?

One of the most obvious fundamental skills is grammar and basic sentence composition. I would agree that a baseline ability to use words and form sentences is important for writing. But I don’t see technology having any impact on this, any more than SparkNotes did back when I was a student. First of all, a lot of this gets taught in elementary and middle school, before students are going to be tempted to turn in counterfeit assignments. Even if that’s not entirely true, the technology to aid student’s grammar and spelling has been around for a long time. Students are still able to form sentences.

Second of all, grammar is more of a social signal than the objective, right-or-wrong skill we credit it to be. The ability to use “correct” grammar reflects whether someone has received an elite education—not whether they can reason logically or write prose worth reading. As a tool of logic, people naturally learn syntactic structures which allow them to get their point across to other members of their community. For example, in one of my favorite ever linguistics papers, William Labov (the founder of sociolinguistics, and one of the first linguistics to truly mount a countermovement to Chomsky) argues that sophisticated grammar can actually obscure the speaker’s point, and that what we think of as “non-standard” grammatical structures can actually be more direct.1

Another fundamental skill is how to structure an argument. The conventional wisdom is that students can’t learn to compose more sophisticated forms of essays until they have mastered the basic format. So we teach students to write the ubiquitous, hyper-standardized five paragraph essay: an introduction, three body paragraphs supporting the claim made in the introduction, and a conclusion.

This assumption more reflects what is easier for teachers to grade than what is meaningful for students to write. The inescapable fact is that almost no writing in the real world conforms to the five paragraph template. I’m not necessarily against the five paragraph essay, as I do think it’s a good baseline structure for making an argument. But relying on this as a kind of gold standard of coherent writing seems that it encourages students to learn the rules of the five paragraph essay while excusing them from having to do the more difficult work of thinking through the best way to say what they’re trying to say.

I think it’s worth questioning whether we’re doing a good enough job teaching the principles of argument structure as it currently stands. It seems to me a little like calculus: we tell students that this is the standard to which they must aspire, and then once they’ve learned it we set them free into a world which will pretty much never ask them to deploy it. Is technology really what stands in the way of students learning how to make an argument?

Overall, my response to the walk-before-run argument is that we’re just okay at teaching students to walk and to be frank most of them never end up running all that fast anyway.

Ultimately, in order to figure out what fundamentals we want to teach, we need to figure out what we want to students to be able to do in the long-term. At the very least, one consideration should be matching what we’re requiring of students to the problems they are going to face after school. This is not the only point of education, but it is an important one. And there’s no rule in real life that says you can’t use technological solutions. For most students, they need to be learning to write with the aid of technologies like ChatGPT, not to learn to be scared of them because using them is tantamount to cheating.

In cognitive science, an individual’s ability to perform an action is a function of two things: their capacity to perform the action, and their motivation to do so. A lot of how we think about education is about in terms of increasing students’ capacity. The need to be better at whatever it is we are trying to teach them—in this case, writing. But they also need to want to be better. And to me, this is the problem that more often than not gets overlooked. Therefore, the bigger gains to be made are not in the effectiveness of our increasing student capacity, but in increasing student motivation.

And here’s the thing. What we’ve been asking students to learn is the most boring form of writing possible, the kind that ChatGPT can now produce ad infinitum. Is it any wonder students don’t like writing? From this perspective, the introduction of ChatGPT is great. We just automated the most boring part of writing. Now students don’t have to bother with writing filler to please their teacher! If we play our cards right, they might discover that no, they do not in fact despise writing. They might actually love it.

Honestly. This is great news.

Good writing starts with good reading

My proposal is that the way to improve how we teach writing is to improve how we teach reading. As Stephen King once wrote, “If you want to be a writer, you must do two things above all others: read a lot and write a lot.”

The problem is that we start off by giving students literary masterworks before they’re ready, and because they don’t have the skills to grapple with them they conclude that so-called great books are a lot less great than teachers say they are.

Imagine if we taught math this way. Students would walk into the classroom in ninth grade, and the teacher would write out the fundamental theorem of calculus on the blackboard (or, you know, the digital screen thing). “This is one of the greatest inventions of human intellect,” the teacher would say. “If you really dig into it, you’ll uncover the secret wisdom of the ages. Trust me.” But no one teaches math this way. It would be madness. Obviously, we all know that you have to start with the basics and build up to the most advanced, most sophisticated concepts. This is even how we try to teach basic writing skills, as discussed above. But somehow, this logic is completely lost on us when we teach reading.

We give students the best stuff, the highest achievements of humanity’s literary minds, way too soon. Students don’t have the skills or experience necessary to engage with these masterworks, just as they wouldn’t have the ability to handle calculus if they didn’t have a grasp of algebra first. As a consequence, they don’t get the point of reading. And as a consequence of that, they don’t get the point of writing.

So why not? Why not get ChatGPT to write your essay for you? What’s the use of reading the book when you can just get a decent grade on the assignment and move on with your life?

The reason this strikes me as a big problem is largely because there’s an easy and obvious solution. Simply put, there is SO much outstanding work written to appeal to young adults. And instead we insist on teaching… Shakespeare?

Shakespeare is great and all, but it kinda seems to me like the calculus of literature—an advanced skill rather than an entry-level one. There is just too much excellent writing out there specifically for middle school and high school kids for us not to teach them. The fact that something is a “literary classic” is not a good enough reason to teach it students without giving them the necessary preparation to appreciate why it is such a profound literary achievement.

For example, a couple months ago I read a work of Young Adult fiction called More Than This by Patrick Ness. I picked it up because the cover featured a blurb from John Green (another YA author whose work is incredibly good) which said, “Just read it.” So I did. It’s about a high school boy who wakes up one day to find that he’s in hell. For this young man, hell is a post-apocalyptic version of his childhood hometown, but with no one in it. He has to figure out what is actually going on. Is he really, like, in hell? Why? What’s he supposed to do here? Is there a way to get out—or is this it?

The plot of the book is incredibly well-structured. The imagery of this grayscale, ash-covered hellscape is as vivid and worthy as any literary work. It deals with big themes, but ones that teens care about. They are not dissimilar from the themes adults care about—family, purpose, trauma, death, friendship—but are presented in the sharp relief in which teenagers experience them, rather than the oblique, layered way that adults do. And, most important of all, the book is constructed to achieve this in simple, straightforward language. You get all the upside of a literary classic, without any of the barriers to entry. And for the emotional circumstances that most teenagers find themselves, the number of students who are going to get something out of an author like Patrick Ness versus Charles Dickens or Emily Brontë… it’s huge!

For that matter, what about graphic novels? We’re not talking about comic books here. We’re talking about serious literary works that make use of a visual medium in addition to a verbal one. For example, Persepolis by Marjane Satrapi. Or Maus by Art Spiegelman. Or my personal favorite: Logicomix, a graphical based on the life and work of Bertrand Russell, by Apostolos Doxiadis and Christos Papadimitriou.

My own educational philosophy is that you must meet a student where they’re at. You have to ask them to go slightly beyond what they’re currently comfortable with. You can’t undershoot it, or they get bored. And you can’t overshoot it, or they get discouraged. There’s no use in trying to force them to do something they’re not yet ready to do. It’s better to be patient, to play the long game; the teacher’s job is to big the student tools with which they can build their own future. In most cases, it’s really difficult to make this work at scale: how do you know where each student is at individually? To me, choosing the right books seems like the lowest-hanging fruit of meeting a student where they’re at.

The logic is simple. If they’re a young adult, ask them to read a book for a young adult. As much as you want them to be able to read Shakespeare, you can’t force them to do it now. They have to build up to it. And you have to trust that if you can get them hooked on Patrick Ness and Marjane Satrapi, and give them the tools to continue challenging themselves as a reader, that they’ll get to Shakespeare and Jane Austen and Henry James eventually.

As it currently stands, we structure our students’ literary curricula as if they are all going to be avid readers of literary fiction from a diverse group of international authors by the end of high school. And do you know what the result is? The vast majority of adults, even educated ones, grow up to think they either have no interest in literary fiction or wouldn’t get much out of it even if they tried.

Sure, a tiny fraction of students will find that kind of work—the Brontës and Dickens of this world—engaging before they start college. And those kids? They’re going to be alright. They’re going to find their way. They’re not going to be leaning on ChatGPT to complete their assignments.

The problem we need to solve is how to get students to care about reading. This is one big difference between math and literature. No one thinks that we’re going to educate a populous of life-long mathematicians. People aren’t going to sit around in their forties or fifties in an armchair doing problem sets. But there’s a different expectation for literature. In a very real sense, an educated populous should be sitting down, at least every once in a while, to read a book. The extent to which we’re able to instill that motivation early matters—even more for books than for mathematics.

We do a decent job teaching students to have the capacity to read and write. Could we do better? Absolutely. And we’re going to figure out how ChatGPT fits into that. But where we really fall short is equipping students with the motivation to care about reading and writing. Just because something is part of the “literary canon” isn’t a good enough justification for teaching it; in fact, it might have a diminishing effect on student motivation. And if we can get that right—creating students who want to be life-long readers and therefore have an intuitive sense of the value of good writing—then we’ll have gone along way toward improving our educational system for reading and writing. Motivation can’t be automated.

Meaning Lab

Discussion about this post