When AI Wrote for MIT Students — and the Professor Noticed Right in the First Paragraph
Creative writing students and artificial intelligence in the same classroom — that combination might seem harmless at first glance, but what an MIT professor discovered at the start of a semester sparked a debate that goes way beyond the university walls.
Micah Nathan, novelist and professor of fictional and nonfictional writing at MIT since 2017, was gearing up for another workshop round like so many before. When two pieces landed on his desk for the first assessment with that suspicious polish, that too-perfect fit within narrative structures, he knew immediately: the students hadn’t written those.
And that’s where a story begins that matters not just to anyone who teaches or learns writing, but to anyone who’s ever wondered where the line falls between using AI as a tool and letting it think for you. 🤔
What happened in that MIT classroom became a mirror for something repeating silently in writing workshops, courses, and universities around the world.
This isn’t about banning technology or romanticizing creative suffering.
The point is more subtle — and more urgent — than it seems: when a student hands over the creative process to an AI, what exactly are they leaving behind?
What Happened in the MIT Classroom
Nathan didn’t walk into class ready to hunt for irregularities. He showed up, as he always does, prepared to hear what the students had produced over the week. But two of the submitted pieces had something different about them — they weren’t bad, quite the opposite. They were technically flawless, with tightly wound narrative arcs, characters with defined trajectories, and metaphors that sounded too natural to be natural. That was exactly the problem.
Good writing has friction. It has strange choices. It has that one sentence that shouldn’t work but does. These pieces had none of that. The prose was too polished for young writers, every character came pre-packaged, every metaphor was a collage without context — what Nathan described as a kind of involuntary pastiche.
He didn’t need any AI detection software. He simply knew. After years of reading work from students learning to master fiction, the professor knows the DNA of a human draft well — and those pieces didn’t carry a single trace of it.
The Confessions That Changed the Class
At the start of that workshop session, Nathan turned to the two students and openly said he knew AI had written their stories. He also made it clear that nobody was in trouble — MIT’s policies on AI use were still being defined, and the course syllabus itself left a gap. Beyond that, Nathan offered an honest reflection: if AI had been available when he was an undergrad, would he have resisted the temptation? Probably not.
What followed was a silence that lasted several moments, broken only by the sound of the room’s radiators. Then, a tearful confession: one of the students said she had used AI because she was afraid of looking stupid. She shared that she loved writing stories and hated that she had turned to a language model. But she couldn’t help herself — and described a sequence that resembles the progression of an addiction. First, she fed the AI her text just for a grammar check. The model suggested line edits and she accepted them. Then came the offer of structural edits. And finally, the AI offered to rewrite the entire piece.
The other student admitted he had never written a short story before. He had an idea but simply didn’t know where to start. When Nathan asked why he hadn’t sought help, the student shrugged.
Other students started raising their hands. One asked why it was bad for AI to write stories as long as the ideas were theirs. Another wanted to know the difference between using AI and using a human editor. And a third posed the question that perhaps echoed most throughout the room: why, at a university that launched one of the first AI research programs in the world back in 1959, were they even having this debate? Wasn’t artificial intelligence supposed to make everyone’s life easier and less stressful?
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The Response That Became a Teaching Moment
Nathan later acknowledged that the conversation following the confessions was one of the most productive teaching moments in his eight years at MIT. He explained to the class that writing shouldn’t be easy. That the process can be tedious, yes, but that doesn’t make it mechanical. Writing isn’t just the production of sentences — it’s endurance training through sustained attention. It’s a way of discovering what you think by trying to put it into words.
A large language model, an LLM, can reproduce the appearance of that activity. But it can’t replace it, because the value isn’t just in the object produced, but in the transformation that happens during its creation.
The professor didn’t penalize them in the traditional sense. Instead, the entire class returned to their texts with a new understanding of what was at stake. And the rewrites that followed were, according to Nathan, infinitely more interesting than anything AI could have generated. Imperfect, yes. But genuine — and full of the marks that make a text belong to someone.
AI Prose: Perfectly Mediocre
Nathan has a pretty precise description for the kind of text language models produce. He calls AI prose perfectly mediocre — a kind of inert shine that reads like a Frankensteinian amalgamation of creative writing workshop texts, an involuntary parody of the style it tries to imitate.
To illustrate this uncanny quality, he turned to a quote from poet Alfred Tennyson describing the character Maud in his eponymous poem: faultily faultless, icily regular, splendidly null — dead perfection, no more.
Attentive readers feel that emptiness, even when they can’t quite articulate what’s wrong. They sense the body moving without a brain behind it. In contrast, fiction written by students is gloriously flawed — a visible struggle on the page between what the author is trying to say and what’s actually being said. Nathan compares that prose to a foal learning to walk: even on wobbly legs, he sees hints of future grace. That awkwardness is necessary. Its absence would be proof that the foal never learned to walk at all.
Writing Workshops and AI: A Real Tension
Writing workshops have always been spaces of vulnerability. It’s where you show an unfinished draft to other people and hear what didn’t work. That process has an intentional discomfort built into it — it exists so the writer learns to separate from the text, to see what they wrote with distance and then come back with more precision.
When artificial intelligence enters that equation as the primary writer, that entire cycle gets cut short. The student jumps straight to a text that’s already been through a kind of automatic polish, and the workshop loses its main working material: the honest draft.
As Nathan observed, an effective workshop is essentially a paradox: students need to provide textual evidence to support qualitative assessments as if they were quantitative. This is especially challenging for STEM students, who are used to problems with clear answers and clean methods to reach them. Fiction doesn’t work that way. Good writing feels good to read. Bad writing feels bad. And learning to navigate that subjectivity is a fundamental part of the exercise.
That doesn’t mean AI has no place inside a workshop. Several educators and writers who work in this format have been experimenting with uses that make sense — using language models to generate variations of a scene the student already wrote, to explore alternative points of view for the same character, or even to identify repetitive patterns the author can no longer see after so many re-reads. In those cases, the tool works as a mirror or a possibility generator, but the voice, the choices, and the starting point still belong to the writer.
The difference between that kind of use and what happened at MIT is exactly this: who’s in control of the creative process. 😬
Technophobia Is Nothing New — But the Warning Is Legitimate
Nathan places the reaction to AI within a historical context that helps put the debate in perspective. Death and taxes are universal certainties, and technophobia might be the third. In 1565, nearly a century after Gutenberg invented the movable type press, Swiss scientist Conrad Gessner was already worried about the confusing and harmful abundance of books. An article published in the journal Nature in 1889 labeled the telephone the most dangerous of all inventions because it enters every dwelling and its endless network of wires is a perpetual threat to life and property.
Now it’s AI’s turn on the worry list. And this time, the warnings come backed by data. A preliminary 2025 study from the MIT Media Lab found that participants who used ChatGPT to write essays showed lower neural connectivity than those who wrote without assistance.
Other studies point in the same direction. A report titled AI Assistance Reduces Persistence and Hurts Independent Performance suggests that AI assistance lowers persistence and damages independent performance. Another, titled Generative Artificial Intelligence Reliance and Executive Function Attenuation, presents behavioral evidence of cognitive offloading in adults who use AI at high frequency.
Nathan acknowledges that these studies still need broader peer review to be considered definitive. But he argues the central warning is hard to ignore and doesn’t need a study to be validated: by allowing students to use AI routinely and thoughtlessly, we are weakening their minds.
The Creative Process Isn’t a Problem to Be Solved
There’s a very common misunderstanding about what the creative process is, especially among students who are just starting out. The idea that writing begins when you know what you want to say — and that everything before that is just an obstacle — is one of the most limiting beliefs a writer can carry.
In practice, writing is a form of thinking. The text isn’t the result of thought; it is the thought happening. When you hand that process over to an AI, you’re not saving time — you’re skipping the most important part of the exercise. It’s like asking someone else to do your workout while you wait for the results to show up on your body.
Nathan connected this idea to a powerful literary reference. In his 1946 essay, Confessions of a Book Reviewer, George Orwell describes himself surrounded by unread books, constantly inventing reactions to works about which he has no spontaneous feelings. Orwell argues that producing reviews in high volume under tight deadlines not only distorts the work of reading — it distorts the person doing it. The mechanical fabrication of responses corrodes judgment and standards collapse.
What Orwell is describing is what happens when language is produced under conditions that disconnect it from thought. The reviewer performs the form of a response without having actually responded. What Orwell couldn’t have anticipated is that this condition would eventually be outsourced even further up the chain. When a workshop fills with AI-generated fiction, every writer and reader becomes the reviewer Orwell describes.
Orwell closes his essay arguing that criticism would be healthier if it were slower, more selective, and less industrial. The same argument now applies to fiction writing. AI accelerates the writing process, but it’s not selective at all — and, in an ironic cycle, it turns the act of creation into the kind of mechanical task it was supposedly meant to automate.
Experienced writers know this in different ways, but they always arrive at the same conclusion: first drafts are bad because they need to be. The mess of a first draft is evidence of a cognitive process at work. When AI replaces that draft with something polished, the student never goes through the stage of productive discomfort that generates real understanding of their own voice, their own recurring themes, their own technical limitations that need to be worked on.
And the most ironic detail of all is that the more a student uses AI to replace their writing, the less they develop the ability to use AI in a truly creative and collaborative way in the future — because they lack the personal repertoire needed to ask good questions, identify what’s wrong in generated outputs, or adapt the results into something that sounds like them. 🎯
How Educators Are Responding to This
The response from most educational institutions is still in the realm of prohibition or detection — tools that identify AI-generated text, usage policies that vary from classroom to classroom, and penalties for anyone who gets caught. This approach has clear limitations: detection tools aren’t foolproof, the policies create a relationship of distrust between professor and student, and none of it addresses the central question, which is why the student turned to AI in the first place.
Nathan reflected on exactly this when designing his course syllabus. Before the incident, he had included a note that explicitly avoided a surveillance posture:
Playing the AI detection game drags me into a surveillance mindset that undermines the workshop environment. If you use AI, it reveals your orientation toward writing. Do you want to make art or just turn in a text? Do you actually want to learn to write, or just pretend you’re learning?
He was confident those questions would be enough to discourage use. They weren’t. And that forced him to rethink his approach.
The New Policy: Clear and Direct
After the incident, Nathan shifted to something more explicit. Now he states openly: he doesn’t want students using AI to write their work. He wants their words. He wants access to their thinking, their voice, their struggles to find what they want to say and the best way to say it. He wants to see what happens when someone tries to move through language without an intermediary completing the thought.
He makes a point of emphasizing that this is a pedagogical position, not a moral or technical one. The workshop only works if there’s a writer in the room — someone whose thinking is visible on the page and who can speak directly about that thinking. Using AI to write not only nullifies the entire concept of peer review — we’re there to workshop each other, not to workshop generic AI text — but it also guarantees the weakening of the muscles needed to wrestle with writing head-on.
The educators getting the most interesting results are the ones who changed the question. Instead of asking how to prevent AI use, they’re asking how to redesign assignments so the process is visible — and so the value lies in the journey, not just the final product. This includes requiring process journals alongside submitted texts, conducting real-time revisions during class, using AI itself as comparative analysis material, or structuring assignments so the starting point is so specific and personal that it could hardly be outsourced without it being obvious.
Some writing workshops around the world have started including dedicated sessions on artificial intelligence in their curriculum — not to teach how to use it, but so students understand what’s happening when a model generates text. When you understand that AI is essentially making a statistical prediction about which words should come next based on existing text patterns, it becomes easier to see why the output is always competent and almost never surprising. And it’s in the surprise — in the unexpected sentence, the strange metaphor that works, the narrative choice that goes against convention — that literature truly happens.
What Changed After That Night
Nathan reports that since that workshop session, something shifted in the classroom dynamic in a way he hadn’t anticipated. Students began speaking more openly about frustration — about the moments when a draft resists its own author, when the words refuse to cooperate with the idea trying to be born.
He still teaches technique — form, structure, revision — but he notices he keeps returning to the tension between thought and language, to the stories where abstraction refuses to take shape. Class discussions now revolve around why each person’s thinking matters, why the struggle to translate thoughts into words isn’t evidence of failure but a sign of growth. Even when — and especially when — the words fail.
The real danger, according to Nathan, isn’t that AI will replace writers or make the workshop obsolete. It’s that students are getting used to bypassing the friction that once revealed their process.
What he and his students now protect isn’t exactly a boundary against machines. It’s something more like a sanctuary for authorship — a place where everything on the page and everything not yet on it belongs to a real person.
The MIT episode isn’t a scandal. It’s a symptom of a transitional moment where the tools have evolved faster than the contexts in which they’re used. Students aren’t villains for turning to AI — they’re responding to real pressures with the tools available to them. The question that remains, and will stay relevant for a long time, is how to ensure that the development of writing — with everything it carries in terms of self-knowledge, voice building, and tolerance for an imperfect process — isn’t silently outsourced in the name of convenience.
Because what’s at stake isn’t just a text. It’s the ability to think for yourself. ✍️
