Writing and writing instruction are facing a critical moment, an unprecedented techno-crisis, we hear. The word is spreading, the headlines announcing alarm. The death of English class. The end of student writing.
Students will outsource their writing assignments to ChatGPT or some other Large Language Model, a stochastic parrot only too willing to stand in for the student. Teachers will never see an authentic student essay ever again. Teachers are doomed! Doooooomed!
But AI couldn't threaten the writing classroom if we hadn’t spent the last few decades preparing the way. In too many classrooms, writing instruction is a fragile house of cards that was always going to collapse under the first stiff wind, and deserve to do so. As author and writing teacher John Warner put it, ChatGPT won’t kill any assignments that didn’t deserve to die. As Shannon Vallor put it, “AI can devalue our humanity only because we already devalued it ourselves.”
Writing instruction has long been wandering into a dark alley; AI was simply waiting at the end of that alley to mug its misdirected victim.
One of the oldest challenges of being human is both simple and profound. We are each individually locked into our own fleshy container, filled with ideas and emotions and images and impressions and memories, things we’ve experienced and things we’ve learned that we want to share with other humans, and yet we have no easy way to do that. We cannot project our thoughts and experience into another brain directly.
In his book On Writing, Stephen King answers the question “What is writing” with a simple line.
“Telepathy, of course.” And later, “a meeting of the minds.”
We dream of telepathy, mind-reading, of magical or technological ways to bridge the gap. We talk. We sing. We dance. We draw. We create entire languages and then expand and refine them in an attempt to make them better signifiers of everything bouncing around inside of us. And we write.
It is one of our most miraculous achievements. By making those marks on a page, we can cast our thoughts, our ideas, our feelings out across space and time.
But—and this is a big but—language, especially in written form, is both the most common and the most mysterious of human activities. It is an imperfect tool for making thought visible. We have no means to measure its effectiveness and accuracy, and we spend countless hours trying to interpret through the noise and parse the wording attempts of our fellow humans. “What did she mean by that?” can take up hours, even days.
That’s a central problem of education, where teachers are charged with reading the minds of their students. We can never be 100% certain that a student is truly and accurately communicating what’s in their head, because the only measure we have of what’s in their head is their communication. On any given day, we might be assessing what they hold in their head, or we might be assessing how well they are communicating it.
In many disciplines, we have worked our way around that central challenge. I can use straightforward objective measures to check for basic recall—do you recognize the desired piece of information when you see it? I can use performative tasks to create mini-artifacts of learning—what are the answers to these math problems?
But written language is the most complex and deep of human activities, and as soon as we try to reduce it to a simple performance or require that students produce an easily-measured artifact, we lose the plot. Asking students to perform a stripped-down version of writing, to write so that we can assess them (or worse yet, that some machine can assess them), is not the same as asking them to write for real.
Writing is thinking made concrete and visible. But by focusing only on the concrete and visible artifact of writing, and not the origin and process of its creation, the thinking behind it, we open the door to many other methods—non-thinking methods—to produce that artifact. And that’s where all our troubles, including those exacerbated by chatbots, began.
We have been living in a golden age of bad writing instruction. Text bots like ChatGPT are simply the latest step in a long slow march toward mechanical, performative writing.
This may be the oldest sort of in-school writing, the kind of performative writing in which the student jumps through hoops to prove to the teacher that she can jump through hoops.
Performative essays typically come with plenty of instruction about format and very specific expectations; the teacher may, even unwittingly, be suggesting that she has in her mind a picture of what the ideal essay for this prompt would look like, and she will be judging your work based on how closely it matches that ideal
This mechanistic model feeds on templates, as if there is some Platonic ideal of The Essay and all the writer is really doing is swapping out select words and phrases to match their assigned topic, trying to show they can pierce the ineffable to grasp that ideal.
The genre of performative writing tasks was boosted to even greater prominence by standardized tests, which come with strict and narrow requirements for what the writing-flavored artifact is supposed to look like.
These performative artifacts are for the ease and convenience of teachers and for the people who try to evaluate what teachers do. It’s all well and good for scholars to debate at inconclusive length whether or not Hemmingway or Chopin or Morrison are great writers and thereby open discussion to what the characteristics of great writing might be, but the front office needs a grade for Chad by next Tuesday, and the state needs to know if Chad’s teacher is highly effective or not, and besides, Chad is no Hemmingway, so let’s stop talking about aspects of writing quality and decide if Chad has produced a high scoring artifact.
Or maybe Chad’s teacher is thinking, “How am I supposed to judge whether this is any good or not? I’m not sure I can identify high quality writing, but I can damned well tell whether or not Chad followed the directions.”
Performative writing artifacts are attractive because they can be scored by an algorithm. Scoring by algorithm pre-dates scoring by computer. In the early days of the Pennsylvania state writing assessment, essays were scored by tables full of actual English teachers. We had the benefit of our professional expertise, but we were also quickly trained to follow a rubric which was an algorithm designed to guide us to the correct score. Soon, the state determined that anyone could be trained to follow that rubric, and regular teachers were displaced by temps answering classified ads.
Nothing follows an algorithm like computer software. But it’s important to remember, as vendor after vendor touts a new AI software product that can score essays, that computers do not and can not read and understand language in any sense in which we ordinarily use those words. They can only examine the external qualities of the writing and compare those qualities to however many millions of samples they have “learned” from. They too can only judge how well Chad has followed some rules; they just happen to be able to store a gazillion more rules than Chad’s overstressed human teacher. As demonstrated by the work of Les Perlman and his team at MIT over the past decades, computers are unequipped to deal with any inventive or unusual language use, while simultaneously being ill equipped to notice unusual content (like Abraham Lincoln leading the US through World War II).
Teachers at my school were able to crack the code of scoring well on state writing assessments with just a few simple rules. Recycle the prompt. Fill up lots of space, even if you have to be redundant to do so. Do not worry about factual accuracy. Use some big words, even if you’re not sure you’ve used them correctly (I was always partial to “plethora”). And it works. Our students scored well every time.
The performative model has been nurtured and cultivated in schools. It’s a performance students put on for their teacher, and, during the last couple of decades, for whatever lightly trained humans or barely competent software will actually set eyes (or sensors) on it.
This was the model favored by David Coleman, architect of the ELA standards for Common Core, but he had not invented it. He just seems to have internalized it from his own education. Writing for school is about writing to satisfy the requirements of the teacher. The teacher has spent weeks putting ideas into the student’s head, and Coleman’s description of how to read can be boiled down basically as “read this literature as preparation for writing a paper about it for your professor.” In Coleman’s world reading and writing are a closed loop that circles around a classroom. Having absorbed the teacher’s instruction, the student’s job is now to present the material for assessment.
Students are admonished to stay within the four corners of the text. At no point does this closed loop intersect any other part of reality—not with how the student understands the world, not with a grasp of culture, not with how the author communicates ideas that might spur action or vision of new generations.
So we crank out a parade of students who are equipped with performative writing skills that are of no direct use in the world, no help in the quest to bridge the gulf between one human and the world with which they want to communicate. Some students may rebel or encounter a teacher who swims against the tide (there are plenty out there), or they may, on their own, beat their experience into something useful. But mostly students have been taught not to write, but to comply with the algorithm.
ChatGPT and other Large Language Models are a reckoning, because they can also follow the algorithm, and unlike students, they don’t have to learn to block off portions of their brain to do it. Teachers are freaking out over the possibility that students may now be able to pull up an undetectable piece of cheating, a fully competent essay generated in seconds by a computer algorithm. While some folks are asking the question, “How shall we ever detect computer-assisted fraud,” the better question to ask is, “If this essay can be polished off by a computer, why are we asking human beings to write it?”
ChatGPT understands nothing, has no feelings or insights to convey, is no smarter than the paper on which we print words. It is stupid, stupid as a rock. It does what it does simply by following an assortment of algorithms, and so we have to ask—
Have we been teaching students to write, or to follow algorithms, to perform the steps laid out for them?
The question should make us uncomfortable.
Writing as a performative algorithmic act is, like a ChatGPT essay, empty of any meaning. It is not meant to communicate (and when students try to use it that way, they are too often slapped down). And, unfortunately, it has spread beyond the walls of schools. Demagogues and marketing executives treat language as a tool to draw desired behavior out of others. A former student of mine started his career writing short puff advertorial blurbs, like three sentences about a light-up toilet seat. It wasn’t, he said, really writing. The internet has increased the demand for writing (now known as “content”) that is not meant to communicate between human beings, but to fill up space, to attract eyeballs and/or the search engine bots, to then hand over the eyeballs to advertising.
It is its own empty form, and I suppose that mastering the empty performative forms of student writing might prepare students for that work—except that of course that kind of artifact can now be extruded swiftly and easily by AI. Knowing how to write like a robot is no help when the robots finally show up to do the job for far less pay.
I’ve talked to teachers and writing gurus who believe that if we can teach students how to manage these empty forms, they will somehow later in life be able to fill the forms with meaning and purpose and personal communication, and perhaps there are a favored few that manage that trick. But the vast majority of students get stuck on the question of “Why are we doing this empty exercise?” And the answer is nothing deeper than, “It’s a trick you must perform for your teacher and the state to get out of here and move on to a stage in life when you won’t ever have to do it again.”
Why teach students a skill that is not the real thing in hopes that the resemblance will somehow lead them to discover the real thing later?
Communication is the point of writing for real. It’s the whole object, the reason to bother with any of it. To start with the empty performance is like buying a suit for your future spouse when you are ten years old. And now, on top of thinking “Well, this seems pointless,” students can add, “Besides, there’s an app for this.”
Learning to perform the algorithm is about learning to fake. Fake interest in the topic. Fake the voice of a person writing about the topic. Fake some kind of developed insight or understanding. When begging my students to write authentically, I would say, “Life is too short to sign your name to a lie.”
The performative essay is also about compliance. It is about how well the student (and increasingly, also the teacher) will comply with the instructions, the requirements, the algorithmic demands. We find ourselves in a culture that values students who are willing to subsume their own ideas, their own expression, their own human desire to be understood by other human beings. That’s the chilling layer beneath this debate; if you are willing to give up freedom of expression, what other freedoms can you be convinced to give up?
Writing ought to be for real. It ought to be exciting and deeply human, carrying the full flush of putting yourself out into the world, to hurl your thoughts out into the intra-human void in hopes of connecting with others, of being seen and heard.
With all that in mind, this book is not about a concrete specific program—a set of teacher and student algorithms to follow. Trying to tweak the algorithm into a more human form is not the answer. Not lipstick on a pig, but a synthetic flesh face on a robot frame.
I started the thinking that went into this book long before generative AI started to grab the world by the eyeballs. ChatGPT did not create any of the issues discussed here, but it brought them into focus and demanded that we start thinking more carefully about what writing instruction—and writing itself—is for.
The focus here is on establishing a culture of writing in a classroom, about establishing for the teacher and the students (and parents) a mindset that allows writing for real not just to happen, but to thrive, and to become a tool that young humans can carry out into their lives, a real, living, vital human thing. All writing humans can use this set of values for re-focusing on the work. If these values are the foundation of a writing program, the program, whatever specifics are built into it (even technological ones), will be solid. In teaching, as in writing, knowing what you want to do makes it much easier to see how to do it. The spotlight here is on the classroom because that is the one place where we address writing deliberately, but the need for humanity and intention in our writing extends well beyond the classroom walls.
A thousand years ago, my eighteen year old self wrote an essay about why he wanted to go into teaching—a belief that if people were better at reading and writing and communicating, the world would be a marginally better place. There isn’t a corner of our world that would not benefit from fully authentically human communication. Perhaps by reflecting on what happens in a writing classroom, we can find some help in dealing with writing in the larger world.
Still here? This summer I took some time to finally hammer out a version of the book about the teaching of writing that has been rattling about in my head for too many years. This is the first chapter from that work, which has been sitting and waiting for me to figure out what to do next. In the meantime, I want to share this small bit.
