Let me tell you a story that will make you laugh, then make you furious.
A professor at Brown University—one of the most selective institutions in the world—did something radical. He replaced the standard take-home final with an in-person exam. The result? Scores plummeted by 50%. Not 10%. Not 20%. Half.
This isn’t a cheating scandal. It’s a confession.
The professor, who taught a course on machine learning, suspected that students were using AI to complete take-home assignments. So he ordered a proctored, in-person final. Suddenly, the same students who had been acing every online quiz couldn’t answer basic conceptual questions. The grade distribution collapsed.
You’ve probably suspected it too. That high-achieving friend who aced every online test might not be as brilliant as they seemed. The coworker with the perfect LinkedIn profile might be running tasks through ChatGPT. We’ve all felt the quiet unease—the sense that the game has changed, but the rules haven’t.
But here’s the part that stings: these were Ivy League students. Admitted for their intellectual horsepower. The same people who will run our hospitals, our courts, our tech companies. And when given the choice between thinking and cheating, they chose the path of least resistance—empowered by tools that let them outsource their cognition.
We cannot choose to become idiots. That’s what the professor wrote in his email to the class after the exam. And he’s right. But the real idiot here isn’t the student—it’s the system that pretends take-home exams measure anything real.
Let’s be honest: the traditional take-home final was already a joke before AI. Open-book, open-note, unlimited time. It rewards endurance, not understanding. Now add ChatGPT. You get a perfect score on a test that never tested you in the first place.
The 50% drop isn’t proof that students cheated. It’s proof that our entire model of assessing human competence via unsupervised homework is obsolete. Punishing students for adapting to a broken incentive structure is like blaming fish for swimming in polluted water.
So what’s the solution? Not more AI detectors (they’re unreliable). Not draconian surveillance (that’s dystopian). Not going back to hand-written blue books (that’s nostalgic).
The real answer is harder: redesign how we measure understanding. Oral exams. Project-based assessments. Real-time problem-solving. Anything that forces the student to actually think, in real time, with their own brain.
But here’s the uglier truth: that costs time and money. It requires smaller classes, more graders, better pedagogy. Universities won’t do it because it’s expensive. They’ll keep the current model because it’s cheap and scalable—and then wag their fingers when students exploit its flaws.
The elite education system is a Ponzi scheme of trust. We trust the brand, not the product. And now the product is being exposed.
This story from Brown is not a one-off. It’s a warning. If the smartest students in the world can’t resist the temptation to outsource their thinking, then every school, every employer, every credential must ask: What are we actually measuring? And what happens when we stop measuring competence and start measuring compliance?
The 50% drop is a mirror. Look into it. You might not like what you see.
FAQ
Q: Isn't this just one anecdote? Maybe the in-person exam was harder or poorly designed.
A: The professor was the same, the content was similar, and the students were the same cohort. The only variable was the setting. A 50% drop is statistically catastrophic—far beyond normal variance. This is a pattern, not a fluke.
Q: Should all colleges go back to in-person exams?
A: Not exactly. In-person exams are one band-aid. The deeper issue is that take-home assessments were already weak. We need to redesign evaluation methods entirely—oral exams, project defenses, real-time problem-solving. Surveillance-only solutions fail and breed resentment.
Q: Maybe AI is making education obsolete, and that's fine. Why not let students use AI?
A: Because then we're no longer evaluating students—we're evaluating their ability to use AI. That's a different skill. If the goal is to certify human understanding, then AI-assisted answers undermine that. If the goal is to certify tool use, then overhaul the curriculum and be honest about it.