The Reading Skill That Makes AI Safe to Use in the Classroom

The essay was good. That was the problem.

A student in my literature class turned in an essay comparing a Filipino poem about environmental loss with a famous English one. On the surface, the paper did everything right: clean paragraphs, confident transitions, a thesis that arrived on time. And yet something had gone missing, in a way I almost failed to catch: the paper had set two poems side by side and quietly decided they were the same poem. 

Of course, suffering is not nothing, and the sentence, “Grief is universal, both works show that suffering unites us all,” is not false. But it proposes a kind of truth that also tends to erase. One poem carried a specific Philippine weather, a specific history of who gets left to drown and who gets warned in time; the other carried trenches, a European century, a different machinery of death. The student’s essay—fluent, helped along by a chatbot, and entirely sincere—had pressed two unlike things flat until the seam disappeared. And the seam was where the meaning lived.

This is the real risk, and not the one we keep naming. A widely shared 2025 study from the MIT Media Lab found that students who leaned on AI to write showed the weakest brain engagement and the least ownership over their own words. The worry hardened into a familiar verdict: the machine is making us lazy. But that verdict misses it. The machine is not making this student lazy; it is doing something more particular. And right now, as Philippine universities race to write their rules, we are answering the wrong question.

University after university is drafting the same memo: a policy on what counts as cheating, which tools are banned, which are tolerated, and how to tell. The committees are not wrong to worry, but notice its shape: a worry about detection and restriction, a list of what students must not do. Meanwhile, the second Congressional Commission on Education, EDCOM 2, in its final report released in January, traced a learning crisis that only deepens as students climb grade levels: nearly nine in 10 children entering Grade 7 cannot read at that level, and by 15 years old, the gap between a Filipino learner and the curriculum has widened to some five and a half years. Into that gap walks a machine that writes fluently on command—and the public conversation has largely settled on a single phrase: we must protect critical thinking.

I agree. But “critical thinking” is too large a thing to put on a syllabus. What you can teach is a specific habit of reading, and it is worth naming exactly.

Set two unlike things beside each other—two poems, two news reports, two accounts of one event, and resist the pull to decide they are really the same. Hold the difference open. Ask not which is correct but what each is doing, and why, and from where. Read each not as a finished verdict handed to you but as something made by someone, somewhere, with interests, inside a history that changes what it means. To compare, done well, is not to match but to judge and to choose, and to keep choosing, rather than letting the likeness between two things persuade you that the work is done.

That habit is called comparativism, and it is the defining way of reading in a discipline most Filipinos have never studied: Comparative Literature. Its thinkers do not all define it the same way; some treat it as a method, others as something closer to an ethics, but they agree on what it refuses. Comparison, done well, is never a tidy compare-and-contrast exercise; it is an act of judgment that declines to pretend the playing field is level. It sounds abstract. It is, in fact, the most practical reading skill of the moment.

Consider what the machine does when it compares. A large language model (LLM) is, at heart, an averaging engine: ask it to weigh two poems, two arguments, two accounts of a flood, and it produces something fluent and balanced, and that fluency is exactly the danger because it is built by smoothing. The model has read everything and sided with nothing. It returns the consensus. It is very good at finding what two things share and very bad at honoring how they differ, because difference is friction, and friction is what the smoothing removes.

Gayatri Spivak, who has spent a career rethinking what comparison even means, put the principle years before any of us typed a prompt: comparison is never really compare-and-contrast. It is, she wrote, a matter of “judging and choosing”—deciding what each thing is worth, and from where, and at what cost, rather than letting a discovered likeness end the thought. To find sameness everywhere, she warned, is to give in to the false promise of a level field. She was writing about how the powerful read the rest of the world: colonial empires, the unevenness no honest comparison can pretend away. Now the same sentence reads like a review of ChatGPT.

This is the operation that comparativism trains, and the machine cannot. Where the model resolves, the comparativist holds the contradiction open. Where the model presents its answer as finished, the comparativist remembers what the scholar Susan Bassnett argued: that comparison does not happen on the page at all. It happens in the reader, in the act of supplying the history a text leaves out: the context that, as Bassnett put it, can radically change what a thing means. The poem about the flood means one thing in a country where the floods are predictable, and the warnings are not, and another in a country spared both. The machine does not know which country you are reading from. You do.

None of this means the tool is the enemy. Comparison has always made two motions: the work of finding connection—what travels between texts, what one tradition borrows or rewrites from another, which Bassnett called literary transfer—and the work of judgment, deciding what each thing is worth, from where, and at what cost. The machine is almost miraculously good at the first; it can read ten thousand novels and map a pattern no lifetime could catch. What it cannot do is the second, because judging requires a someone, situated somewhere, with something at stake—and the machine is no one. So the meeting of discipline and technology is not a contest the human is bound to lose. It is a division of labor: let the model do the matching it is built for, and let the reader do the judging it cannot, taking its fluent output not as a verdict but as one more text to be weighed against everything she knows that it does not.

And here the Filipino reader has an advantage most of the discipline’s European founders never imagined. Consider a student of mine from Negros. She thinks in Hiligaynon, studies theory in English, and reads the country’s literature in Filipino, often in a single afternoon. She does not experience them as one smooth language. She feels the seams, the places where one tongue holds what another cannot. When she reaches for the Hiligaynon word for a particular kind of longing and finds that neither English nor Filipino quite carries it, she is not failing to translate; she is reading the gap itself, registering that the three words are not labels for one feeling but three slightly different feelings, shaped by three different worlds. Spivak called translation the most intimate kind of reading, something more demanding than swapping one word for another: a refusal to assume the two words are equivalent at all. My student will not have read Spivak, but she knows it in her body before lunch. That instinct is the seed of comparativism. It is not foreign, and it is not elite. For a multilingual nation, it is closer to a birthright—one we have simply never thought to teach as a skill.

The strongest objection to all of this is that I could come across as too indulgent. Somewhere there is a reasonable skeptic—an education reformer, say—who has read the same reports I have, EDCOM 2’s dismal findings among them. Against numbers like these, here comes a literature instructor proposing we teach a graduate-seminar concept to the whole country. Comparativism? These students need to read a paragraph and keep a job. And if a free chatbot helps a working student in Tacloban draft a scholarship letter she could not have written alone, who is the humanist to call that a loss? The last thing a system this strained needs is a more expensive, more elite thing to teach.

I want to take that seriously, because most of it is true. Access is not a luxury, and a tool that lowers the barrier for the student the system has already failed is, on balance, a good. A policy that meets this moment with bans and detection software (punishing the students with the least support) would be both cruel and useless. On all of this, the skeptic and I agree.

I will concede the harder half first. The founders of this discipline thought comparativism is expensive: Spivak argued it demanded a whole apparatus: deep training across many languages, faculty seminars, consortia pooling what no single campus could afford. The full discipline is costly, and a system like ours in the Philippines that is stretched this thin cannot build it.

But the apparatus is not the skill. Bassnett spent her career arguing almost the opposite of her own field’s grandeur: that comparison is not a bounded discipline but a method; it is something you do to any two texts, in any classroom, in any language, with no laboratory and no license required. The apparatus is costly; the habit is cheap. A nursing instructor can teach it with two conflicting patient charts. A senior high school teacher can teach it with two news accounts of the same barangay election. It costs a question—what is each of these doing, and from where?—not a budget line. And the multilingual student I described is not the exception here; she is the rule. A country that already reads across three languages before lunch is not poorly positioned to teach reading across difference. It is the best-positioned country I can think of.

I will not pretend this is easy to do at scale. No system stretched as thin as ours can simply draft a new literacy into every syllabus. But difficulty is not the same as misplaced priority. The skill is not in competition with foundational literacy; it is what foundational literacy is for. We do not teach reading so that students can be fluently agreed with by a machine. We teach it so they can disagree with one.

So return to that committee and its memo. The question on the table is: how do we stop students from using this? It is the wrong question, and it was from the start. The students will use it. The honest question, the one a memo cannot answer but a classroom can, is what kind of reader we are sending into a world that now speaks to them in one smooth, confident, agreeable voice.

Because that voice is not going away, and it is very good at what it does. It hands back the consensus with the corners filed off and the seams pressed flat. A reader who cannot feel the seams will take it and call it knowledge. A reader trained to read across difference will take the same output and do the one thing the machine cannot do for her: weigh it. That reader does not need to be a literature major. She needs to have been taught, somewhere, by someone, that comparison is an act of judgment and not a search for sameness.

I think of that student and her flattened poems, and I do not blame her. No one had taught her there was a seam to look for. That is not her failure; it is ours, and it is still fixable, but only if we stop writing rules about the tool and start teaching the skill that makes the tool safe to hold. Put comparativism in the nursing syllabus and the journalism seminar and the senior high reading class, not as the property of a dying humanities department but as the basic literacy of this moment. I am inclined to believe a country already half-fluent in reading across difference is not the worst place to begin. It may be the best.

ALSO READ:

The Pope, the Machine, and What We Owe Each Other

Is There Such a Thing as Ethical AI?