For the past couple of weeks, as students have submitted their final papers and faculty have immersed themselves in grading, academics on social media have been focused almost exclusively on a single topic: artificial intelligence (AI) cheating.

The problem is serious enough that Princeton University has just turned its back on 133 years of tradition. Since 1893, faculty members have left the room during exams, trusting students to behave honorably. No more. The university has now announced that proctors will be present during exams to prevent students from using their phones to cheat with AI.

Some might bristle at the word “cheating,” but let’s be clear: when students submit work that is not their own, that is cheating, plain and simple. There may be some gray area if the professor is allowing or even encouraging AI use, although, to my mind, that simply makes the faculty member complicit. But most professors, at least in the humanities, do not allow students to write with AI, much less encourage it. Professors are doing their best to discourage it and, in many cases, actively prohibiting it—except, perhaps, for certain very narrow applications such as research.

That’s why humanities faculty’s social media posts have been mostly bemoaning the fact that nearly all their students use AI on their written assignments. They know it, but can’t exactly prove it in a way that would stand up in a disciplinary hearing, much less in court, and it’s more than their career is worth to level allegations that might get them sued or even fired. And yet they’re incredibly frustrated because their students aren’t learning the things they need to learn in humanities classes—specifically, how to think and how to write.

As Massachusetts Institute of Technology writing professor Micah Nathan put it in a recent essay for the Guardian:

Writing isn’t just the production of sentences – it’s the training of endurance by way of sustained attention. It’s a way of learning what one thinks by attempting to say it. An LLM can reproduce the appearance of that activity, but it can’t replace it, because the value lies not only in the object produced but in the transformation that occurs during its making.

Or, in the words of Ralph Waldo Emerson, nearly 200 years before LLMs were even a thing, “Hence arises a grave mischief. The sacredness which attaches to the act of creation, the act of thought, is instantly transferred to the record.” For students taking humanities courses, the point is not the piece of writing; it’s the writing itself.

Humanities professors know this. They know their students are missing out on whatever small chance they once had in the modern corporate university to engage in deep thought, to struggle with putting those thoughts into words—activities indispensable to intellectual growth. And yet they feel powerless to do anything about it. Some, in a last-ditch attempt to stem the tide of AI non-writing, are going old-school, requiring students to write essays by hand in class. But this creates an extra burden for instructors, and it doesn’t work in every situation. Some classes are too big or configured such that in-class writing is nearly impossible. Moreover, it doesn’t take into account the fact that learning to write well involves working through a series of drafts, which cannot be done in a 75-minute class period.

Of particular concern are asynchronous online classes, where there is literally no practical way to proctor students’ writing without creating hours of additional work for instructors while remaining, at best, only marginally effective. For instance, instructors could require students to record themselves while writing. We could also require students to write their papers in person at campus testing centers, but that would defeat the purpose of offering asynchronous online courses, especially when many students do not even live within the institution’s service area.

I propose a technological solution to this moral problem. Sure, a moral solution would be better. It would be nice if we could educate students about the ethics of unauthorized AI use and shame them into doing the right thing. But we all know that isn’t going to happen. It did not stop internet plagiarism a generation ago. What did work, though, was a technological tool—Turnitin and similar sites. Once professors started using Turnitin, instances of students simply taking material from the internet and trying to pass it off as their own declined dramatically.

Just as AI is far more sophisticated than the internet, so must a technological tool to discourage AI cheating be more sophisticated than Turnitin. It can’t just be another type of detector, because “AI detectors” are notoriously unreliable. Texts as diverse as passages from the Bible and the Declaration of Independence have been flagged by various “detectors” as being substantially AI-generated. (Note that the sentence I quoted from Micah Nathan, above, would almost certainly be flagged as well, due to his use of the em-dash and the “it’s not A, it’s B” construction.)

What we need is something that goes far beyond a detector. We need a preventer. I’m no technologist, but I will try to explain in rough terms what I think might work.

Suppose every university required students to use some sort of software whenever they interacted with the institution’s learning platform—software that would prevent them from accessing AI at the same time. Or perhaps students could log into something similar to a VPN before connecting. Either way, they would not be able to use the learning platform without first accessing this tool, whatever form it might take, and the tool itself would prevent them from visiting AI sites while using it. It would also include a built-in word processor that would not allow cutting and pasting from outside sources, and students would be required to complete all their writing, including rough drafts, through this tool. (This might work for short-answer exams as well.)

One objection I heard when I ran this idea by a colleague was that students would just use AI to write their papers on another computer, then manually copy it into the learning-platform tool I’m proposing. To that I say: maybe. But if they could not simply copy and paste, and instead had to transpose the whole thing sentence by sentence, at least they would be doing some writing and probably learning something. Whereas now, by simply prompting AI to write their paper for them and then uploading it, they are learning nothing.

But perhaps there’s a way to eliminate that shortcut, too, as well as any other workarounds bright—if morally bankrupt—students might invent. I have no doubt brighter minds than mine could come up with solutions to all these problems within this very rough framework.

The bigger question is this: Can the human mind create a problem that the human mind cannot ultimately solve? I don’t believe so. Even in the humanities, we have our technologists. It’s time for them to put their big brains to work on the problem of AI cheating, so the rest of us can get back to the business of teaching thinking and writing.

Follow Rob Jenkins on X.