The Oath Against the Machine

In the early eighties I studied at the Budapest Technical University, and this is how you handed in a programming assignment. You wrote the code out by hand. You carried it to the machine room. Someone punched it onto cards. You waited a day for your slot on the computer. The next morning you picked up the printout and found out you’d mistyped a variable name, and you started the wait all over again.

The card puncher did not make me a better thinker. It vanished the moment something faster came along, and I don’t remember anyone shedding a tear.

I keep coming back to that, because I’ve now watched the same fight happen four or five times, and it always runs the same way. So when I read that a university in Budapest is asking students to sign a declaration — under criminal liability — that they used no AI in their coursework, my first thought wasn’t shock. It was: I’ve seen this before.

Go back far enough and it’s almost funny. In 1299 the city of Florence banned the new Hindu-Arabic numerals from official records. The reason given was fraud: with one stroke of the pen you could turn a 0 into a 6, and the neat new digits made forgery too easy. The old guard kept their Roman numerals and their counting boards. The trouble was that the new numbers let a trader do in a minute what used to take an afternoon. The ban held, on and off, into the fifteenth century. Then business won, the way business always does.

Do you see the shape of it yet?

It keeps repeating. The slide rule ran engineering until Hewlett-Packard sold the HP-35 in 1972; within about five years, colleges were dropping their slide-rule classes. Calculators were banned, then argued over, then welcomed — the maths teachers’ own body signed off on them in 1980. And the people who warned that mental arithmetic would rot were right. It did. Maths teaching didn’t bring the old drills back. It just changed what it tested.

So I understand the urge to reach for the oath. I also think it beats itself. A signed promise not to cheat is a confession that you can no longer tell whether someone cheated. If you could tell, you’d just give the work a bad mark. The oath is a teacher reaching for the police because he’s run out of eyes.

And the eyes really are gone. Vanderbilt turned off Turnitin’s AI checker in 2023 because it flagged too much honest work as fake, and it was worst of all toward students writing in a second language. Meanwhile, people were paying essay mills to do their homework long before ChatGPT showed up. Ban the tool and the wish to dodge the work doesn’t go anywhere.

So is this just the calculator argument in a new coat?

No, and here’s why I think so. Every tool before this one made a skill you already had run faster. A calculator makes a good mathematician quicker; it can’t write a proof for a student who couldn’t have written it. AI can. It can hand back strong work to someone who doesn’t have the skill at all. And that breaks the quiet bet under nearly every essay grade — that if you can produce the argument, you must be able to build one. That bet survived the abacus, the slide rule and Google. It doesn’t survive this. This isn’t a small step up. It’s a different thing.

Take the number everyone shared. A Brown University professor found that take-home essay scores above 90% dropped below 50% when the same students did the same work in a locked, watched room. Only 2 out of 86 kept their marks. Everybody read that as a story about AI.

Read it the other way round. Fewer than three students in a hundred could redo their own take-home work when someone was watching — and that hole didn’t open the week ChatGPT arrived. Maybe take-home essays never measured the single student in the first place. AI just made it too obvious to ignore.

Now the part that really stings a professor. The phone in the student’s pocket knows more than he does. GPT-4 was sold as scoring in the top 10% on the bar exam; a later, harder look put it closer to the middle of the pack against real lawyers — but it stings either way. So what is the person at the front of the room still good for?

The honest answer from history is that authority doesn’t die, it moves. The printing press ended the university’s grip on the written word, and the university lived by leaning on the thing a pile of printed pages couldn’t do: making you argue and defend your case out loud. The rare skill now isn’t holding the fact. It’s telling good work from bad, and knowing which questions are even worth asking. A school can still own that job — if it’s brave enough to say so.

And it’s been done. Medical schools swallowed the stethoscope, the X-ray and the digital patient record. Every one of those could have been called the end of the doctor’s hard-won knowledge. Instead they moved the test away from memorized anatomy and toward reasoning at the bedside. They stopped testing what the machine now holds, and started testing what it doesn’t.

See what that does to the whole “should we allow it or not” debate? It makes it the wrong debate. The real question is what the exam is even for. If it’s there to check that you can produce text, then AI is a killer and the oath is just holding back the water. If it’s there to check that you can judge, steer and fix what the machine gives you, then banning the machine is like teaching bookkeeping and forbidding the spreadsheet.

The university itself will probably come through this. What’s really at risk is the diploma. An economist named Michael Spence won a Nobel for a simple point: a degree is worth something only as long as it reliably tracks real ability. Cut that link and you get what a used-car market gets when buyers can’t spot the lemons — they stop trusting all of it. That, not the lecture hall, is what the oath is really guarding. And an oath can’t guard it. Only assessment people can trust can.

What would that look like? Three moves are already on the table. Split the exam the way maths split around the calculator: watched, AI-free papers for the core, and AI-assisted projects that look like the real job. Grade the work, not just the result — the drafts, the prompts, the edits, the way a coding team is judged by its history and not only its final program. And bring back the oldest test there is: the spoken defense. Let the student prepare with every tool on earth, then sit them down and make them think it through out loud, where no model can take their seat.

I could be wrong about all of this. Maybe it really is Google all over again. Back in 2011 psychologists showed we’d already stopped storing facts and started storing where to find them, and the world kept turning. Writing by hand is better for your brain than typing, and we still cut cursive from schools the second the job market stopped asking for it. The “it’s good for you” argument, on its own, has never once won.

Every earlier round of this fight ended when the school changed what it measured. This round is asking whether it can change quickly enough — before the diploma quietly stops meaning anything.

So here’s what I’d leave you with. What would a university have to make you do before you trusted its degree again?