Large Language Models and the Final Paragraph

I dread reading the final paragraph of the five-paragraph essay. When I teach it, I find myself resorting to platitudes I’ve told my students to eschew. Even with the conclusion of the best five-paragraph essay often reads the way a last swig of coffee filled with burnt grounds tastes. It’s like the back-end heel of a loaf of bread, which has also gone stale. You choke it down because you don’t want to waste it. And when a paper has really been brutally bad, given the effort of will necessary to drag my eyes over that last assemblage of half-hearted sentences soggy with the student’s tear-stained plea that the paper be over already, I’m as uninterested in that conclusion as the writer is. I feel like I’m in a hostage situation where I asked the students to trap me in a dinner party and demanded that they serve wilted lettuce for dessert.

I hope I’ve made it clear that I don’t get excited about reading the final paragraph. And yet, as we know, no good effort can be accomplished without a good end. Any real sense of adventure has to come with the dramatic tension of that final destination. While some may ask, “Why not move on from the constraints of the tired convention?,” I hope here to defend those constraints as inherent to the poetic wisdom of the traditional essay. So let me tell you why I think it’s absolutely essential for writing teachers to keep demanding that paragraph, composed without artificial intelligence, even if it means a diet of wilted lettuce for the foreseeable future.

In Writing at the End of the World, Richard Miller asks writing teachers to take seriously the question: What value can writing pedagogy have in the face of the apocalypse? From school shootings to the atom bomb, we at least had one assurance: learning to compose our words help us to get some purchase on composing ourselves. In 2005, he wrote the following sentence: “If you’re in the business of teaching others how to read and write with care, there’s no escaping the sense that your labor is increasingly irrelevant” (Miller, 5). The technology which existed at the time, by comparison to large language models today, make his 2005 fears for writing pedagogy seem almost quaint. Tertullian asked, “What hath Athens to do with Jerusalem?” Miller asks, and we must ask anew, “What hath teaching writing to do with this technological modernity?”

By disposition, I’m not inclined to share Miller’s fears about the significance of teaching students how to write college essays. Or at least I wasn’t until my own personal pedagogical apocalypse hit me on the day when my confidence in our craft should have been at its highest. On a cold spring day in Niagara Falls in 2018, I received a call that should have put me into a state of euphoria. I was being offered an assistant professor position at College of the Ozarks, and to my ears accustomed to adjunct pay, the offered salary sounded like a fortune. What’s more, I wanted to propose to my girlfriend, and the aforementioned-adjunct pay wasn’t a good foundation to build a life on unless we wanted to live in my grandmother’s attic for the rest of our lives. So the labor of three college degrees, and the cost of prayer, tears, and sleepless nights had finally paid off. I was getting the job, and I could propose to my beloved.

Yet barely two minutes before I received the call from C of O, I had been informed that over the weekend one of my students had committed suicide. I felt a numb guilt wash over the joy I wanted to feel for my long held hopes finally becoming a reality. Before I could call and tell my family my good news, I had to go and tell my Business Communication students that their classmate had taken his life. When a school administrator wrote me to ask if I could offer any thoughts about this student to speak some comfort to the parents, I was totally paralyzed. There was no way I could speak comfort to these parents. Yes, he was very good at analyzing workplace rhetoric. He understood the idea of a manager’s ethos being central for creating a safe environment for employers. So I was confronted with the doubts about which Miller wrote: “I have these doubts, you see, doubts silently shared by many who spend their days teaching others the literate arts” (6). What do my assignments have to say about the meaning of life? What I had been teaching him seemed so painfully banal in view of his mortality. Even though I surely couldn’t take responsibility for his death, I feared that my course did not respect his mortality, or that of any students I had taught. If I struggled to articulate what relevance writing pedagogy had for my students’ lives, I knew I had no answer to the question of what it might have to do with their deaths. What hath pedagogy to do with death?

Thinking about our students’ mortality is not pleasant work, especially as we look at our students with such love. I can’t help but have a sort of parental concern for my students, even as I try not to show it too much. They’re all going to leave us after a few years, so we better not get too attached, we might think. We have to work with so many students that we have to guard our hearts. But as I have been thinking about my visceral offense caused by so-called Artificial Intelligence, the thought that most presses upon me is that large language models are offensive because they do not respect the mortality of the user. A library with long stacks of books warns us that reading is a time-consuming venture. It creates a space that looms over our bodies—the long aisles, the rows of shelves, the dusty quiet as it muffles our steps and makes our coughs feel intrusive, all tell us that we are mortal beings traversing a great conversation that we only visit as brief interlopers.

In The Presence of the Word, Walter Ong reminds us that our communication technologies modify how our minds interact with our senses, that artificially captured sound puts us out of touch with the temporality of spoken words “to a degree unthinkable for typographic man,” so that the “present sensorium is dismayingly mixed and we are hard put to understand it” (9). But while Ong sees such technologies as opening up new kinds of understanding, artificial intelligence smothers the sensorium, cloaking the fact that, as Ong reminds us, the human “communicates with [the] whole body through all senses” (1). The large language model lies to mortality by having no structured admission of the finite nature of the user who encounters the vast, shapeless pit of hidden verbal calculation with a speed that has no respect for the organic process of cognitive development which comes from the slow, plodding quest for the one true word.

Tolkien’s last line of his classic Beowulf essay reads, “It is to idols that men turned (and turn) for quick and literal answers.” A book, however thick, tells the reader that it has a first page and a last and admits the limited claim it has to the reader’s life. Artificial language models have no similar structure for modeling a good death. Their regurgitations remind me of the lifeless idols described in Psalm 135:15-18: “The idols of the nations are silver and gold, made by human hands. They have mouths, but cannot speak, eyes, but cannot see. They have ears, but cannot hear, nor is there breath in their mouths. Those who make them will be like them, and so will all who trust in them.” The new idols are beginning to imitate voice and sight, but such innovation merely amplifies the illusion of thought. They remix and review dead language like a verbal zombie, unable to let the words die and find authentic afterlife. The original human speaker is forgotten, and the large language model looms like the cowled darkness of the Nazgul, hissing promises of endless, definitive speech that has no life because it has no death.

Artificial Intelligence can produce an imitation of the five-paragraph essay, but a well-crafted five-paragraph essay esteems life’s time-sensitive realities. Unlike the sausage-grinder perfection of a large language model’s imitative essay-product (like square “cheese products” in plastic-wrapped single slices that don’t melt naturally but do remind the palate of the idea of cheese), the five-paragraph essay thinks in terms of limitations and marries the need life has to meet ordered structure with the sparking energy of real thought. As Ong puts it, “Speech itself as sound is irrevocably committed to time” (40). The five-paragraph essay, when well executed, never forgets the time-bound nature of communication. The introduction tells your reader why they should care. Why? Because they’re going to die some day. Try to write in a way that matters for people who are going to die some day. Try to craft a thesis statement that respects the limited time of the audience. Craft a topic sentence that respects the mind of the audience. Craft a paragraph that respects the audience’s nervous system. Good writing respects the body and the soul of the reader.

Like the sonnet, the five-paragraph essay traps investment in truth felt in the heart and forged in the mind by means of its life-respecting limitations. When done well, such essays are structured to capture the urgency of the present word as the only bridge between the past and the future that carries the lifeblood of thought from one mind to another, or even of one mind to itself. Artificial intelligence blunts this interaction by creating the impression of a lifeless lump of now. Only corpses are entities of the now and only the now, and the past and present working upon them are the carrion crawlers that benefit from death. Likewise the large language model freezes the user into a passive receiver of the artificial vulture’s pseudo activity, and the mind mistakes the computer’s calculations for one’s own mental life. He who would keep his life must lose it, and the lack of lost life in the hollow verbal spurts of a large language model drowns the agency of the user. User. The same word we employ for the drug addict, by the way.

So I dread the final paragraph not really because it’s often bad, but because not only do I confront my students’ mortality there, I also confront my mortal frailty as a teacher. I only have so much time with them, and that sense of helplessness overwhelms me when I realize I will inevitably fall short of my aspirations to help them write better, live better, and die better. So here’s an alternative to wilted lettuce. When we ask students to follow through from the birth of the idea in the introduction to the celebration of it in the conclusion, we might consider that when they learn to say a final word well, it is a practice for what we used to call the good death.

I know all of this might sound morbid, but to my medievalist mind, memento mori is a life-giving strategy. It’s when we forget mortality that we really become morbid. So the final paragraph might be a kind of palliative care for the student’s grade, or a place for the student to remember living the life of the mind, like a fond memory from your youth. You’ve lived a long four paragraphs to get to this point, dear Essayist Oakenshield—it’s time to reconcile with Bilbo and affirm that if we all valued food and cheer above hoarded gold, it would be a merrier world. But you’ll only write that conclusion with a noble end if you lived nobly from word to word and sentence to sentence and paragraph to paragraph yourself. That’s how you pass into the West and avoid plunging into the cracks of Mount Doom. So don’t chat with the Ringwraith: instead, write a good conclusion yourself, because it’s good practice for your final paragraph.

Image via RawPixel