The AI College Split Is Already Here

The interesting question is not whether students will use AI. They will.

The real question is whether colleges are going to treat AI like contraband or like literacy.

A friend recently told me about his kids taking college classes at two different schools. One school is nearly anti-AI. The allowed use is narrow, cautious, and mostly prohibited unless the professor carves out a specific exception.

The other school has gone the opposite direction. AI is allowed almost anywhere, as long as the student is clear about how they used it, what they prompted, and where the human work begins and ends.

My guess is that the second model wins.

Not because it is cleaner. It is not.

It wins because it matches reality.

This feels a lot like the early internet panic

When I was in college, the internet was still new enough that professors did not quite know what to do with it.

There were rules about how many sources had to be actual books from the library. Internet sources were treated as suspicious by default. Sometimes any fact found online had to be cross-referenced against something published in print. Wikipedia was not really part of the picture yet, but the general mood was already there: the web was useful, but also a little too easy.

Looking back, some of that skepticism was reasonable. The internet did make lazy research easier. It also made good research dramatically more accessible.

The answer was not to ban the internet. The answer was to teach students how to evaluate sources, compare claims, follow citations, and understand the difference between finding information and knowing something.

AI feels similar, except faster and more destabilizing.

A search engine changed where students found information.

An LLM changes the shape of the work itself.

Disclosure is probably the new citation

The college that allows AI with disclosure seems closer to the right long-term model.

If a student uses AI to brainstorm, summarize readings, critique an essay, generate practice questions, debug code, or explain a concept three different ways, I want to know that. Not because it automatically makes the work invalid, but because it is part of the intellectual process.

Disclosure turns AI use from a hidden shortcut into something that can be evaluated.

What did you ask?

What did the model give you?

What did you accept, reject, verify, rewrite, or build on?

That is not a perfect system. Students can lie. Professors can overtrust the disclosure. Institutions can turn the whole thing into paperwork.

But it is still better than pretending the tool is not there.

A blanket ban mostly teaches students to hide the workflow. A disclosure norm at least creates a path toward judgment.

College may become less like job training

The bigger question is what college is for when the commercial value of many white-collar skills starts to fall.

A lot of modern higher education functions as job training plus accreditation. Learn to write reports. Build spreadsheets. Code. Research. Present. Produce acceptable professional work on deadline. Get the credential that says you can do those things.

But AI is already very good at the entry-level versions of those tasks.

By the time my oldest gets to college, a little more than a decade from now, I have a hard time imagining that "can write a competent report" or "can build a spreadsheet model" will carry the same market signal it once did. There may be much less need for humans to perform the mechanical middle of those workflows from scratch.

That could make college worse: more expensive, less connected to jobs, and more confused about its purpose.

It could also make college better.

Maybe it moves back toward something closer to its older ideal: learning how to think, argue, read, understand culture, study history, ask better questions, and live with complicated ideas. Less credential factory, more formation.

I am not sure that is what institutions will choose. But AI makes the old bargain harder to defend.

Personalized learning could be the real breakthrough

The obvious AI-in-school story is cheating.

The more interesting one is tutoring.

I think back to my own time in college and how often I was just stuck. The explanation in class did not land. Office hours might help, if my schedule worked. The library might have another reference that explained it differently. Classmates might be equally confused.

I remember one lab where the professor gave the class a task and walked out. People started talking, and it became clear that nobody really knew what he was asking us to do.

An LLM would have changed that moment.

Not by completing the assignment, though of course it could try. The more valuable thing would have been the ability to say: explain this another way, assume I missed the prerequisite, give me a smaller example, quiz me, now make it visual, now tell me where my reasoning went wrong.

That is a different kind of educational access.

A good AI tutor can look at every assignment, every quiz, every wrong answer, and every hesitation. It can build a model of where a student is behind and where they are ready to move faster. It can repeat itself without getting annoyed. It can offer a fifth explanation when the first four fail.

For students without access to great schools, great teachers, or patient one-on-one help, that could be huge.

The struggle might still matter

The uncomfortable part is that I am not sure all friction is bad.

Sometimes the struggle to get information is part of the learning. Finding the book, talking to the classmate, sitting with confusion, trying the wrong approach, and slowly building the internal machinery to solve the problem might be the point.

AI can remove useless friction. That is good.

It can also remove formative friction. That is dangerous.

Some students will use these tools to go further than traditional education would have allowed. They will ask better questions, get unstuck faster, learn more deeply, and build things earlier.

Others will learn a different lesson: the tool can do the job for me.

That works until it does not. Eventually everyone hits a problem where the model is wrong, the context is missing, the task is ambiguous, or the answer requires taste and responsibility. If you have never practiced figuring things out without the machine smoothing the path, that moment is going to be rough.

So maybe the real educational skill is not "use AI" or "do not use AI."

It is knowing when to lean on it, when to interrogate it, and when to put it aside.

The winning schools will teach judgment

I do not think colleges can ban their way out of this.

The tool is too useful, too available, and too deeply connected to the work students will be expected to do after graduation.

But I also do not think "use AI for everything" is enough. That approach skips over the hard part, which is teaching students how to evaluate the tool's output and know when they are relying on it too much.

The better approach is harder: make students show their process, disclose their tools, defend their choices, and demonstrate understanding in ways that cannot be reduced to a polished final artifact.

That may mean more oral exams. More in-class work. More project journals. More prompt logs. More reflection on what changed between the AI-assisted draft and the final version. More grading of judgment, not just output.

The schools that figure this out will probably produce students who are more capable, not less.

The schools that only prohibit AI may preserve the old rituals for a while.

But the world outside the classroom is moving on.