At several recent college commencements, graduates booed speakers who praised artificial intelligence. One student put the contradiction plainly to the Associated Press: students are discouraged and even penalized for using AI, then asked to applaud a speaker celebrating it.
The student was right to notice the contradiction. Students are being told that AI will define the world they are entering, while many classrooms still treat the same technology as contraband.
That should embarrass universities.
Universities are still arguing about AI as if the main question is cheating. That is too small a question.
I am not an AI skeptic. As a professor of engineering, I use these systems seriously, and I think students should too. A student who graduates unable to use AI well is already behind. Sending students into the world without serious AI practice is irresponsible.
But using AI seriously is not the same as outsourcing your mind to it. The danger is fluent work with no ownership behind it.
A student can produce a polished answer without really understanding the question. A draft can look decorated. An email can seem responsible. The words arrive. The tone is right. Nobody is home.
The question, “Did you use AI?”, is obsolete. The better question is: “Did you still think?” That question should force professors to rethink what the classroom is for.
For centuries, much of university teaching has relied on a familiar rhythm: professor lectures, student listens, student goes home, student completes an assignment, professor grades it. This model is simple, scalable, and deeply embedded. It survived the textbook, the calculator, the computer, and the internet. It may survive AI too, mostly because universities are very good at preserving old machinery.
AI has exposed how fragile many traditional assignments were. Take-home essays, polished summaries, coding assignments, and problem sets no longer measure what they used to measure. We can write disclosure rules. We can ask students to make honor promises about what they did or did not use. We can build detection tools. But this is a trench war, and universities are not going to win it.
The line between “using AI” and “not using AI” has been blurring for years. We already accept spellcheck, autocomplete, suggested sentences, and search engines. They are primitive versions of the same story. So where exactly do we draw the line?
The policing instinct is too narrow. A student can avoid AI and still not think. A student can use AI intensely and think very hard. The difference is whether the student can stand behind the answer.
We should stop treating AI as contraband and start treating it as equipment. The classroom should be where students learn to use that equipment under pressure.
Students should use AI before class. They should ask naive questions, request examples, test their understanding, and arrive with better confusion. Unlike a book or a search engine, AI can meet a student at the edge of their own understanding. Students should use AI during class too, not secretly under the desk, but openly as part of the work.
Then the professor has to do the part no sycophantic chatbot can reliably do: apply pressure. Why is this answer right? What assumption is hidden? What would make it fail? Can you derive it yourself? Can you explain it with the laptop closed?
The classroom should become an AI laboratory: a place where machine answers are tested, attacked, repaired, and defended. The professor’s job has shifted. It is not merely to deliver information, because information is now everywhere. The professor’s job is to create intellectual pressure: conceptual, argumentative, technical.
A student should be able to say: “Here is what AI told me. Here is what I think is wrong. Here is what I still do not understand. Here is my best attempt to defend the idea without reading from the machine.” That should be ordinary classroom work, not a confession.
This does not require a grand reform. It can start course by course. Bring one AI answer to class and attack it. Find the hidden assumption. Find the fake confidence. Find the missing step. Replace some take-home work with live defenses. Ask where the machine helped, where it failed, and what the student can still explain with the laptop closed.
If we do not make this shift, we will drown in borrowed thinking while trying to police our students. We will get fluent answers, polished emails, plausible essays, and technically decorated AI slop.
None of this weakens the case for AI-free evaluation. It strengthens it. Some moments should remain brutally simple: a blank page, a problem, a pencil, and the student’s own mind. Written exams, oral exams, live derivations, and in-person defenses must stay with us. They are how we find out whether AI-assisted work has become understanding.
At the same time, we should let go of something. For centuries we have asked students to memorize integration tricks, reproduce derivations they may never use again, and recite formulas they could find in any handbook. That demand was a tax students paid because no better tool existed. The tool exists now.
What matters is whether they can tell when an equation is being misused, whether the assumptions are honest, whether the answer is plausible. Judgment, not retrieval, is the new test.
The printed textbook changed what students could read. The internet changed what they could find. AI changes what they can ask. It can sit next to a single student, infinitely patient, and answer their actual confusion. That deserves better than another fight over plagiarism detection.
The classroom we need is one where AI is everywhere, and borrowed thinking has nowhere to hide.