By ROBERTA MENDONÇA DE CARVALHO

 

Buried in the middle of a recent interview in the Chronicle of Higher Education with David J. Deming, dean of Harvard College, comes a line I know well. Deming says that learning is suffering. On the first day of my capstone class, I tell students the same thing in plainer words: learning hurts. I say it out loud.

 

But the sentence is not mine. I first, and perhaps only, heard it from my late uncle, a controversial college economics professor who was as bright as he was sarcastic. I heard it maybe a couple of times growing up, but never fully grasped it. Years later, while writing my master’s thesis, it came back to me. It fit even better when I wrote my PhD dissertation. I remembered it, and I felt the hurt. I still feel it when I research, write an abstract, construct a talk, or prepare a lecture. Learning hurts because thinking is hard. Producing knowledge means struggling with something you do not yet understand, questioning what you thought you understood, and staying with it long enough to produce something new.

 

I use AI myself, and it has changed how I write. As a non-native English speaker, I still cannot get over having my own English editor, available anytime I want. It means no longer begging people to read my words and correct boring mistakes like the difference between relation and relationship. My friend-and-sometimes editor, Ally Sabo, taught me that difference years ago. It still does not stick in my brain. AI reminds me every time. What it cannot do for me is the hard part: deciding what I think.

 

So when I read the constant barrage of stories about AI, I see arguments for or against as neither defeatism nor AI boosterism. It is a plain description of a problem that every one of us teaching right now faces. Statistics, likely out of date the moment we mention them, show that a great majority of college students are using AI. The choice in front of us was never whether to respond to that fact. It was only ever how.

 

AI is, among many other things, an extremely effective machine for removing that load. It offers a frictionless path from question to answer and, increasingly, from assignment to finished product. That is precisely why the struggle to learn matters in the age of AI. If some intellectual friction is necessary for learning, what happens when students suddenly possess a tool designed to remove it?

 

Banning AI doesn't necessarily restore the friction. It can simply move its use underground, where an instructor can no longer see it, coach through it, or understand what part of the intellectual work the student actually did. If learning hurts, then our job is not to preserve every old assignment exactly as it existed before AI. Our job is to preserve the productive struggle inherent inside it.

 

I happen to be someone who is fairly comfortable navigating without clear sightlines. My father was a merchant marine, and I spent much of my childhood aboard ships, crossing oceans and their dark nights, which left me with a certain tolerance for moving through conditions you can't fully see. Ignorance? Maybe. Even so, the waters here are genuinely dark, for me as much as for anyone teaching or learning. No settled playbook exists yet for teaching a room where, functionally, everyone has a research assistant, a ghostwriter, an English editor, and a graphic designer sitting in their pocket and/or backpack.

 

I am not banning AI. I don't think that's enforceable. But neither have I simply inserted AI into everything I teach. Instead, course by course, assignment by assignment, I have started asking a different question: Where is the part that is supposed to hurt? What intellectual work do I actually need the student to do?

 

The answer is different depending on the course—and sometimes depending on the assignment.

In Introduction to Urban Studies, I simply decided that there are moments when I need students to sit with a question and answer it themselves, without a device intervening between the question and their thinking. Note-taking went back to pen and paper. Vintage. Audacious.

 

In my other courses, I have leaned harder into what AI can't easily do for a student in real time: group work, in-class debates, discussion, and presentations. Students will absolutely use AI to help build their slides, and I started grading around it. An aesthetically pleasing presentation is now common ground.  The points increasingly live somewhere else: How do they present? Can they explain their choices? Can they defend an argument when I interrupt with an unexpected question? In some ways, we are back to the foundations that mattered before PowerPoint: not the slides, but the person beside them.

 

The same logic has reshaped smaller, more mechanical assignments. Building an infographic used to require substantial design effort; now a student can produce something visually impressive in seconds. So, I shifted more of the grading weight to what happens before and after the infographic: the outline developed in class, the evidence selected, the argument the infographic is meant to communicate, and the student's ability to explain those choices. The polished object still matters, but it no longer provides the strongest evidence that learning occurred.

 

The e-portfolio has changed for the same reason. Chapter summaries used to carry substantial heft but are now lightly weighted. Much more of the grade rests on students' responses to what they have read: what they make of an argument, how they connect it to other material, and what questions it raises for them. Can that thinking also be outsourced to AI? Of course it can. I am not pretending I have invented an AI-proof assignment. I have simply moved the weight toward the part of the exercise that I actually care about students doing. And secretly hoping it will work.

 

Now, I am relieved to spot imperfections; I almost long for them! A slightly awkward sentence, an inelegant slide, or a moment when a student has to stop and search for the right answer can tell me something a perfectly polished product no longer necessarily can — where the student is in the learning process.

 

In my capstone course, I've gone further and allowed AI outright, on one condition: Students must account for how they used it. Every semester, some still choose what I now call the analog way; they choose vinyl over Spotify. Each year, there are fewer and fewer. My Spotifyers keep a running diary of their interactions with AI over the semester and submit a short reflective essay on how, specifically, it helped their research. The action-plan report in my Urban Sustainability course follows the same logic.

 

None of this is AI boosterism. None of it is surrender, either. It's an attempt to keep the struggle where I can still see it. Maybe it means having night-vision glasses during those transoceanic nights.

 

And this has always been my instinct in the capstone, long before AI made it newly urgent.

On the first day of my capstone, I tell my students, “This will hurt.”

 

I tell them exactly how much an 8,000 -word research paper, built on data, not opinion, is going to demand of them. And I tell them, just as directly, that I will hold their hands through the entire process. I tell them that I go through the same kind of suffering in my own research and writing, and that this is simply part of the deal.

 

Suffering together makes it easier. It also builds those brain muscles.

 

AI hasn't changed that deal. If anything, it has made the case for it more urgent. The thing a chatbot cannot offer a student, no matter how capable it becomes, is a learning community that goes through the hard parts of thinking alongside them and helps them keep going anyway. And this is where my classroom comes back to a larger argument: universities still bring people together to learn, struggle, criticize, listen, think in community.

 

And suffer.

 

By the way, these are my metaphors, my ideas, my title. I used AI responsibly, mainly for editing and revising. And it still hurt.

 

Roberta Mendonça De Carvalho is teaching associate professor and director of undergraduate studies in the Urban Studies Program.