AI in the Classroom Is Creating a Generation That Can’t Think

When a Fields Medal-winning mathematician at UC Berkeley says he has given up grading homework for correctness because his students are simply feeding problems to artificial intelligence, the rest of us should pay attention. Professor Richard Borcherds now awards points for turning work in on time and leans almost entirely on in-person, device-free exams. He calls the arrangement “not a very good system,” but he sees no better alternative. Grading for accuracy, he notes, only incentivizes AI use and penalizes honesty.

There is a real efficiency argument for AI outside the classroom. It can eliminate arithmetic errors, speed routine calculations, and free professionals for higher-order work. Inside education, however, the same tools produce a net loss. Students who never wrestle with the steps of a problem never learn the steps. They leave school able to summon answers but unable to generate them, which is a far more fragile form of competence. A society that cannot perform basic reasoning without a machine is not liberated; it is dependent.

I saw the practical consequences when I taught a mixed undergraduate and graduate course. Homework correctness counted for a larger share of the grade than the exams. Students had days rather than minutes, and they could consult their notes. In a demanding subject that seemed only fair: the work was hard, the time pressure was low, and the assessment measured whether they had actually mastered the material rather than whether they could sprint through a closed-book test. Remove the ability to grade that homework for correctness and the balance collapses. The only remaining instrument is the timed exam. A class that was merely difficult becomes nearly impossible to pass for anyone who is not already fluent under artificial constraints. Assessment reduces to one of two forms—methodical, open-ended practice or high-stakes speed trials. Students have always preferred the former. Widespread AI use is forcing instructors toward the latter.

Humans learn by trying, failing, correcting, and trying again. That process is slow and sometimes frustrating; it is also how understanding is built. When the feedback loop is short-circuited by an algorithm that supplies the finished product, the learning never occurs. The grade may still appear, but the skill does not. Over time the gap widens between those who can operate without the tool and those who cannot. The former remain useful when systems fail, networks drop, or problems arrive in forms the training data never saw. The latter do not.

We have already seen this pattern once. When universities first widened admissions and lowered standards with dollar signs in their eyes, average student preparedness fell. Rather than accept higher failure rates, many institutions simply pressured faculty to lower the bar—curving grades, diluting content, or redesigning assessments so fewer students failed. The same dynamic is now set to repeat. Heavier reliance on timed exams in difficult subjects will drive scores down again. If administrators respond as they did before, they will again demand that faculty ease standards rather than let large numbers of students fail. The result will be no meaningful standards at all: the bar resting on the floor, and a new generation of graduates entering the workforce both clueless about the underlying skills and incapable of performing them without a machine to do the work for them.

None of this requires romanticizing the pre-AI past. Calculators, search engines, and software packages have long assisted serious work. The difference is that those tools still left the user responsible for the reasoning. Generative systems that produce complete solutions on demand do not. Until universities find ways to restore genuine practice, whether through oral exams, in-class work, or redesigned assignments that AI cannot simply complete, the default will be more weight on timed tests and less genuine competence. That is not progress. It is a quiet surrender of the very capacity education is supposed to cultivate.