I knew my writing students were using AI. Their confessions led to a powerful teaching moment | Micah Nathan

I person been school fabrication penning astatine MIT since 2017. Many of my students past wrote fabrication successful mediate school, and very fewer person knowledgeable a due workshop, truthful astatine nan commencement of each semester I connection these directions for writer and scholar alike:

double quotation markRead nan communicative astatine slightest twice. Mark what useful and what doesn’t – underline awesome sentences, emblem clunky syntax, gaps successful logic and unrealistic dialogue. Ask yourself: does nan communicative work? Why aliases why not? What could amended it? Answer successful a signed missive to nan author, attached to their story. Give your honorable opinions. Remember that an effective adjacent reappraisal demands adjacent reference of nan matter accompanied by a boldness of spirit.

As nan directions foreshadow, astir of nan clip we’re discussing why we didn’t for illustration nan communicative being workshopped, because penning a bully communicative is immensely difficult moreover nether nan champion conditions, particularly for Stem-centric undergrads who thrive wrong a building of quantitative problems and solutions – systems wherever there’s a correct reply and a cleanable method for arriving astatine it.

Fiction penning isn’t quantitative. Good penning feels bully to read; bad penning feels bad. An effective shop is simply a paradox: students must supply textual grounds to support nan qualitative arsenic if it were nan quantitative. This is simply a terrifying imaginable for nan habitually superb student, to beryllium successful stony soundlessness while their classmates and professor slash astatine their work. The enactment of confronting that panic is, itself, an acquisition for nan writer, because penning is some conveyance and alloy for reasoning – absurd made concrete, feelings translated into words. This is what galore writers talk astir erstwhile they mention to bully prose arsenic not conscionable poetic expression, but communication. Thus, erstwhile we knock a writer’s work, not only are we criticizing their artistic choices, we’re besides criticizing – and here’s wherever it tin get individual – nan writer’s feelings and their expertise to pass them.

It’s a batch for nan ego to absorb. Prior to a fewer years ago, nan only measurement a fabrication writer could protect their ego was to either salary personification other to constitute for them, aliases edifice to plagiarism. AI changed each that.

‘Dead perfection’

AI’s prose is perfectly mediocre, producing nan benignant of inert gloss that sounds for illustration a Frankensteinian amalgam of MFA-workshopped writing, an unintentional parody of nan style it mimics. The resultant stories and essays are simulacra of thought, generated via shape nickname learned from millions of human-penned words, rooted successful nary peculiar acquisition by nary peculiar person. AI penning reminds maine of Tennyson’s explanation of nan beautiful Maud successful nan titular poem:

double quotation markFaultily faultless, icily regular, splendidly null
Dead perfection; nary more

Insightful readers consciousness that emptiness moreover if they can’t articulate it. They consciousness that nan assemblage moves without a brain. By contrast, student-written fabrication is gloriously flawed, a struggle connected nan page betwixt what nan writer is trying to opportunity and what’s really being said. The prose stumbles successful a measurement reminiscent of a foal learning really to walk: moreover successful their trembling legs I spot hints of early grace. Such clumsiness is necessary; its absence would beryllium impervious of nan foal ne'er having learned to walk.

Death and taxes; technophobia is nan 3rd certainty. In 1565, astir a period aft Gutenberg invented nan printing press, nan Swiss intelligence Conrad Gessner was already worrying astir nan “confusing and harmful abundance of books”. An 1889 article successful Nature claimed nan telephone is nan astir vulnerable of each inventions “because it enters into each dwelling. Its interminable web of wires is simply a perpetual menace to life and property.” Now we’ve added AI to nan database of worries: a 2025 MIT Media Lab preliminary study recovered that participants who utilized ChatGPT to constitute essays showed little neural connectivity than those who wrote without assistance.

Other studies pass of akin dangers, from not-yet-peer-reviewed reports pinch self-explanatory titles specified arsenic “AI Assistance Reduces Persistence and Hurts Independent Performance” and “Generative Artificial Intelligence Reliance and Executive Function Attenuation: Behavioral Evidence of Cognitive Offload successful High-Use Adults”. Dire stuff, if proven true. But immoderate nan peer-reviewed findings whitethorn be, nan cardinal informing is difficult to disregard and doesn’t require a study for validation: by letting students routinely and thoughtlessly usage AI, we’re weakening their minds. That informing shaped really I addressed AI successful my syllabus. Specifically, really I planned to discourage its use:

double quotation markPlaying nan AI-detection crippled drags maine into a surveillance mindset that undermines nan shop environment. If you usage AI, it reveals your predisposition toward writing. Do you want to make art, aliases conscionable move successful text? Do you want to really study really to write, aliases conscionable dress to do so?

I was definite my questions would shame them into compliance moreover without an definitive prohibition. So astatine nan commencement of past semester, erstwhile I publication 2 of my student’s stories for nan first shop and knew wrong their opening paragraphs that some stories were written by AI, I was hurt. I was besides worried, because I realized that for nan first clip arsenic a penning professor, I had to woody pinch students producing words without work, which wasn’t rather plagiarism and wasn’t rather paying for personification other to do nan job, but it felt for illustration a benignant of naive chicanery; a perversion of nan statement betwixt writer and reader.

As nan first shop started that night, I turned to nan ostensible authors and told them I knew that AI wrote their stories. I didn’t request AI-detection package to know; I conscionable knew. The prose was excessively polished for a young writer, nan arcs excessively tidy, each characteristic prepackaged, each metaphor a pastiche without context. I told nan people nan shop couldn’t proceed because I won’t springiness feedback to an writer who doesn’t exist, but I assured nan would-be authors that they weren’t successful trouble. MIT’s policies regarding AI usage were successful flux, and my syllabus offered an opening. Besides, had AI been disposable during my undergrad years, would I person resisted its help? Of people not. The pedagogical borderlands person ever been filled pinch students questing for shortcuts. The exertion changes but nan quest remains.

For a fewer moments, each was quiet isolated from nan classroom’s ticking radiators. Then, a teary-eyed confession: 1 of nan ostensible authors said she only utilized AI because she was frightened of looking stupid, of being criticized for bad writing. She said she loved penning stories and hated having utilized AI. But she couldn’t extremity herself, recounting a series akin to an addict’s descent: astatine first she fed her communicative into AI for a grammar check, it suggested statement edits and she accepted, past it asked if she wanted structural edits, past it offered to rewrite nan full piece.

The different would-be writer admitted he had ne'er written a short communicative earlier and he had an thought but didn’t cognize wherever to start. I asked him why he didn’t scope retired to maine for help. He shrugged.

One of nan different students raised her hand, saying she didn’t understand why it was bad for AI to constitute stories arsenic agelong arsenic nan stories are based connected their ideas. More students spoke: 1 wanted to cognize really utilizing AI was immoderate different from utilizing a quality editor. Another wanted maine to reply why, astatine a assemblage that launched 1 of nan world’s first AI investigation programs successful 1959, were we moreover having this debate? Isn’t AI meant to make everyone’s life easier? Less stressful? Isn’t nan constituent of AI to free humans from nan tedium of rote tasks?

The speech that followed their confessions was 1 of nan astir productive school moments of my 8 years astatine MIT. Writing, I told them, isn’t expected to beryllium easy, and of people it tin beryllium tedious but that doesn’t make it rote. Writing isn’t conscionable nan accumulation of sentences – it’s nan training of endurance by measurement of sustained attention. It’s a measurement of learning what 1 thinks by attempting to opportunity it. An LLM tin reproduce nan quality of that activity, but it can’t switch it, because nan worth lies not only successful nan entity produced but successful nan translator that occurs during its making.

Bringing backmost friction

In George Orwell’s 1946 essay Confessions of a Book Reviewer, Orwell describes himself surrounded by unread books, “constantly inventing reactions towards books astir which 1 has nary spontaneous feelings whatever”. High-volume, on-deadline reviewing, he argues, does not simply deform nan activity of reference – it deforms nan self. The mindless manufacture of responses erodes judgment, and standards collapse.

Orwell is describing what happens erstwhile connection is produced nether conditions that disconnect it from thought: nan reviewer performs nan style of a consequence without having really responded. What Orwell couldn’t person anticipated is that this information would yet beryllium outsourced upstream. When a shop fills pinch AI-generated fiction, each writer and scholar becomes nan reviewer Orwell describes.

Orwell ends his effort by arguing that disapproval would beryllium healthier if it were slower, much selective, and little industrial. The aforesaid statement now applies to penning fiction. AI speeds nan penning process, but isn’t astatine each selective, and – successful an ironic rhythm – turns nan enactment of creation into nan benignant of rote task it’s meant to automate.

Going forward, my argumentation is now plainly stated: I don’t want students utilizing AI to constitute their work. I want their words. I want entree to their thinking, their voice, their struggles to find what they want to opportunity and nan champion measurement to opportunity it. I want to spot what happens erstwhile personification tries to move done connection without an intermediary finishing nan thought.

This is simply a pedagogical position, not a civilized aliases method one. The shop only useful if there’s a writer successful nan room, personification whose reasoning is visible connected nan page, and who tin speak straight to that thinking. Using AI to constitute not only nullifies nan full adjacent reappraisal conception – we’re present to shop each other, not to shop AI slop – it besides guarantees a weakening of nan muscles needed to wrestle pinch writing.

The threat isn’t that AI will switch writers aliases render nan shop obsolete. It’s that students are becoming accustomed to bypassing nan clash that erstwhile revealed their process.

Since that night, our workshops person changed successful ways I didn’t anticipate. We talk much openly astir frustration, astir nan moments erstwhile a draught resists its ain author. I still thatch trade – form, structure, revision – but find myself returning to nan hostility betwixt thought and language, nan stories wherever abstraction refuses to return shape. We talk why their reasoning matters, that their struggle to construe thoughts into connection isn’t grounds of failure, but a motion of growth. Even when, and particularly when, words fail. What my students and I now defender isn’t a bound against machines truthful overmuch arsenic a sanctuary for authorship, a spot wherever everything connected nan page and not yet connected nan page belongs to an existent person.

• Micah Nathan is simply a novelist, essayist and MIT teacher successful fabrication and nonfiction penning whose books see Gods of Aberdeen and Losing Graceland. His fabrication and essays person appeared successful Vanity Fair, nan Paris Review, Little White Lies, Kinfolk and elsewhere

• Spot illustrations by Cristina Spanò