A Former Professor Estimates 90% of His Online Students Used AI for Coursework

A former Arizona State University professor says he suspected at least 90% of his online students used AI to complete all or part of their coursework before he left teaching earlier this year. His estimate comes without verification, he concedes, but it frames a problem now surfacing across higher education: homework, the traditional unit of measurement for learning, has lost its evidentiary value.

The account, published as a first-person essay on CNET, comes from a former professor of practice at the Walter Cronkite School of Journalism and Mass Communication at Arizona State University. His reflections were prompted by a story last week about a UC Berkeley math professor who no longer bothers grading his students' homework because they are all using AI.

What did grading look like before and after ChatGPT?

The author describes pre-AI grading as demanding but useful. "Reviewing the results of carefully crafted assignments was at best a barometer of how well I was relaying the concepts and crucial elements of the course," he writes of his tenure as a professor. Grading homework and providing feedback, he argues, is arguably the toughest part of being an educator.

That barometer broke after ChatGPT launched in late 2022 and tech giants worldwide followed suit, releasing increasingly sophisticated chatbots and now AI agents (software tools that can carry out multi-step tasks with limited supervision). Grading, he writes, "became deeply depressing," and then something worse: "an infuriating exercise in futility."

The telltale signs piled up in student submissions. Discussion posts littered with complex references to materials outside the course's accepted reading. Students who appeared to be instant masters of the subject matter from day one. He compares the phenomenon to a classic forgery tactic: "forge your mom's signature the first time so you can keep forging it forever." A student who starts the semester producing polished work establishes a baseline no one questions later.

How did he try to detect AI-generated work?

In advanced communications courses, the author says reviewing homework turned into a hypervigilant hunt for the hallmarks of AI-generated text and ideas. His tools for catching it were sorely lacking. "The beast was cunning, quick and never slept," he writes of AI-generated coursework.

His detection method was manual and, by his own admission, mostly ineffective. He ran a smattering of assignments through a few widely available AI chat tools and compared the outputs against student-submitted work. Most of the time, the method was inconclusive. That admission matters: his 90% figure rests on professional suspicion, not confirmed cases, and he states plainly there was no way to verify it.

The pattern he observed had a consistent structure. Despite the lack of verification, he was fairly certain some online students had never done any of the work themselves across an entire course. When those students faced assignments they could not outsource to AI — simulations and recorded oral presentations — they reliably struggled to maintain the standard they had set. But the arithmetic of grade inflation protected them. "When you get A's all semester, one C or D doesn't really matter," he writes. "The AI cheating math still shakes out."

What happened when he confronted a student?

His small in-person classes produced direct confrontations, and one case in particular made him rethink the purpose of his work.

A student gave a presentation early in the semester. Asked a few basic questions in front of the class about how they developed their ideas, the student was flummoxed, unable to explain the circumstances that led to what was supposed to be their own creative output.

The two spoke privately later that day. The student admitted to using ChatGPT to create part of the assignment and agreed never to use it again in the course. On the next assignment, the same discussion took place, and the same promise was given. It did not stick.

That repeated violation led him to his supervising dean, where the institutional response he describes reveals the deeper problem. Asked whether the case should go to the academic integrity board, the dean told him it was at his discretion — and, candidly, that the university and college were still grappling with what stance to take in the face of wide AI adoption. Arizona State's current institutional position, as quoted in the essay, reads in part: "Our focus is on using AI not only to address educational challenges, but to do so in a way that respects and uplifts all learners."

That statement is aspirational rather than operational. It does not tell a professor what to do with a repeat offender, which is precisely the situation the author faced.

Why does this matter beyond one classroom?

The author anticipates the counterargument: cheating is a natural impulse, on taxes or the Monopoly board, and sometimes people put a thumb on the scale because the game looks rigged or too hard to beat.

His answer is that education is not a game. School environments can be competitive, but learning is not something at which a student wins. He describes it as a personal journey meant to unlock fluency a person can call on at will — like knowing how to speak a language or effortlessly opening a door.

Students who outsource the work of learning, he argues, take for granted the value of being able to perform expertise with ease. They also risk ruining their credibility at their own peril. AI functions for today's students as a skeleton key, in his framing: it grants instant access to great grades and passable answers that a busy, bored teacher cannot flag as 100% fake.

His closing position is blunt about the stakes for educators. Until our brains are plugged into our computers 24/7, he writes, AI-assisted work is a crutch that educators cannot afford to quibble about any longer.

What remains unmeasured?

The essay's limits deserve attention alongside its claims. The 90% figure covers only his online students and comes from comparison sampling that he himself calls mostly inconclusive. He offers no detection-tool results, no confirmed-case counts, and no breakdown by assignment type. The UC Berkeley case is cited secondhand, from a story he read, without the professor's name or the course involved.

What the account does establish is the institutional gap. One professor suspected near-universal AI use, documented at least one repeated violation, brought it to his dean, and received discretion instead of policy — from a university whose published stance emphasizes AI adoption rather than enforcement. If those conditions hold elsewhere, the homework era's end will not arrive by decree. It will arrive quietly, one ungraded assignment at a time, until institutions decide what evidence of learning they will accept instead.