Schools Are Catching the Wrong Kind of Cheating

In June, I attended a three day conference outside Orlando where over a thousand educators talked about artificial intelligence. The keynote speaker, C. Edward Watson, projected some interesting statistics generated by Turnitin, the plagiarism service many schools use regularly. In its first year of AI detection, Turnitin reviewed more than two hundred million student papers, flagging twenty two million of them as significantly AI-generated. The company had also quietly revised its false-positive rate from one percent to roughly four. Watson did the multiplication out loud: 4 percent of 22,000,000 is 880,000. In a single year, the tool we bought to protect academic integrity told schools that 880,000 pieces of honest student work were deemed fraudulent.

That number is everything we get wrong about AI in schools compressed into a single statistic. We have decided the problem is cheating, and that the answer is detection. We are wrong on both counts.

Let’s start with detection. A presenter from Jacksonville University told the room that she used to run every paper through an AI checker, until the morning she typed an original piece of her own and watched it come back 60 percent machine written. An audience member shared that he had fed in a dissertation written in 2012 that came back as 80 percent AI generated, 10 years before ChatGPT was released! The explanation belongs on the wall of every faculty lounge: “It sounds like human writing because AI was trained on human writing.” The detector flags people who write the way the machine was taught to write, which is to say, it flags good writers.

The cost of this lands on students, and it is corrosive. One of them told Watson that after she finishes an essay, she goes back through and makes it worse on purpose by using simpler words and clumsier sentences, because if she sounds too polished, she’ll be accused of using AI. We have built a system that punishes clarity and teaches the smartest teenagers to hide it. I am not sure what skill we think we’re developing.

Then there is the harder truth, the one that should end the detection conversation entirely. Even a perfect detector catches only the laziest, “straightforward” cheating. Watson described the student who works backwards, who asks AI to build a detailed outline, what to argue and in what order and what to cut, and then writes every word of the paper himself. No detector on earth will flag that paper, because the student wrote it. And yet every act of actual thinking, the part where learning is supposed to happen, was done by the machine. The cheating we can catch is the kind that barely matters. The kind that matters is invisible.

So the question was never whether a machine touched the page. The question is whether any thinking happened. And that turns out to be a much older question than artificial intelligence.

At the same conference, a session walked participants through a study from MIT that measured students’ brain activity while they wrote, some with a search engine, some with nothing, and some leaning on a chatbot. The press did what the press does. “Is AI Making Us Stupider?” ran one headline. The researchers were irritated enough by the coverage that they added a line to their own FAQ: asked whether their work has proven that AI makes us stupid, they responded, “The answer is no. Please do not use words like ‘stupid.'” What the study found is more useful than the headline… the damage tracked with passivity. The students who used AI to skip the work accumulated what the researchers called cognitive debt, and the ones who used it to push their own thinking did not.

I have been arguing a version of this for years, long before there were electrodes on scalps. Students don’t struggle with the assigned chapter anymore; they paste it into a chatbot and ask for the themes. But the struggle is the whole point. The friction is the learning. Remove it and you have removed the reason to read at all. One recent poll found that 58 percent of adults between 18 and 34 say they use AI specifically to avoid learning something. They are not cheating the institution, but rather they are cheating the one person the institution exists to serve.

This is where I should admit the limits of my own authority. I run schools and a tutoring company, and I also never managed to pass on my love of reading to my own kids, which stings more than I usually let on. So I don’t offer any of this as a man who has solved the burgeoning problem. I offer it because I watch, every week, the difference between a student who uses these tools to think faster and one who uses them to avoid thinking at all. The tool is identical. The student is not.

If thinking is what we’re trying to protect, then something almost funny follows. The skills we spent twenty years calling impractical, close reading and the patient construction of an argument, are suddenly the entire game. In a future dominated by AI, the humanities are the most imperiled by AI, and the most important form of AI literacy. You cannot supervise a machine whose work you can’t evaluate. Watson’s term for the position all of us now occupy was “AI bosses.” If you can’t find your own fingerprints on every line of the output, you aren’t running the tool; the tool is running you. A boss who can’t tell good work from bad is no boss at all.

This is why I grow uneasy when a school proposes to solve AI by bolting on a standalone “AI literacy” class, as though it were a software tutorial. AI is a superb instrument for an expert who can judge what it hands back, and a dangerous one for a novice who cannot. The machine never says “I’m not sure.” It answers everything with identical confidence whether it’s right or an invention. The defense against that is the oldest subject we have, taught with more urgency than we’ve managed in a generation: how to read closely and reason your way to whether something is true, i.e., critical thinking.

Most recently, a number of universities have started bringing back the blue book, the cheap stapled booklet I last filled out in college. It is tempting to read that as retreat, a nostalgia for the world before machines. I read it the other way. We are not trying to go backward. We are trying to find out, in a room, with a pen, whether a young person can still think. We spent two decades telling them the humanities were a luxury they couldn’t afford. The machines may have just made them the curriculum.