What is the best path forward for AI in academia?

A professor at a state university told me last month that half his students are submitting essays written by ChatGPT. He can tell. They all sound the same — confident, verbose, wrong in the exact same way. He said he spends more time now trying to prove cheating than teaching.

A parent called me worried her daughter's college applications would get filtered out by an AI before a human ever saw them. She's probably right.

A prospective student asked me if his CS degree would be obsolete before he graduated. Also probably right.

The academic system is built on a stack of assumptions that AI is making irrelevant. Essays prove you can think. Grades prove you mastered something. Degrees prove you can do the work. All three of those signals are breaking at the same time, and the institutions that grant them are moving too slowly to matter.

The question isn't whether AI should be in academia. It's already there. Students are using it. Professors are drowning in it. The question is what happens next.

From the professor's side: you cannot win an arms race against detection. The tools students use to generate text are evolving faster than the tools that catch them. Every detector has a false positive rate, and every false positive is a student you accused of cheating who didn't. That destroys trust faster than actual cheating does.

The path forward is to stop assigning work that AI can do. If the task is "write a five-paragraph essay on the causes of World War I," the task was already bad pedagogy. It tested recall and structure, not thought. AI just made that obvious.

Assign work that requires judgment, synthesis, a point of view. Assign work where the student has to defend a claim in real time, out loud, to you. Assign work that builds on something they did last week that only they have context for. The goal is not to make the work AI-proof. The goal is to make the work worth doing even if AI helps.

From the parent's side: your kid's application is going into a system that is under siege. Tens of thousands of essays that sound like they were written by the same person. Admissions officers are overwhelmed. The signal-to-noise ratio has collapsed.

The path forward is to make your kid illegible to the average AI output. That means specificity. Real stories, real failures, real obsessions. If the essay could have been written by anyone, it will get lost. If it could only have been written by your kid, it will land.

It also means thinking harder about what college is for. If the degree is just a credential, the credential is losing value faster than tuition is rising. If the degree is access to a network, a lab, a professor who changes how your kid thinks — that still matters. The parents who win the next decade are the ones who stopped paying for the signal and started paying for the thing itself.

From the prospective student's side: your CS degree might be obsolete before you graduate. That is not an argument against the degree. It is an argument for using the four years differently.

The students who will win are the ones who treat college as a build environment, not a certification program. You have access to compute, to researchers, to people solving real problems, to four years where failure is cheap. Use it to build something. Use AI to build faster. The gap between students who graduate with a transcript and students who graduate with a portfolio of shipped projects is about to be the widest it has ever been.

The degree still matters as a credential. It will matter less every year. The thing you built in the margins will matter more.

The academic system is not going to solve this from the top down. The incentives are wrong. The institutions move too slowly. The path forward is for the people inside the system — professors, parents, students — to stop waiting for the system to catch up and start making different choices now.

Professors: assign work that is worth doing even with AI. Parents: make your kid's application illegible to the noise. Students: use the four years to build, not just to pass.

The system is breaking. The people inside it do not have to wait for it to be fixed.

— Mir