Faculty across the nation have declared war on AI-assisted academic dishonesty, and the arsenal is impressive: surveillance software, detection algorithms, honor code pledges signed with the solemnity of a mortgage closing. According to Inside Higher Ed, the war is not going well.
The Detector That Flagged Its Own Author
Kim Manturuk, executive director of the Center for Excellence in Teaching, Learning and Online Education at Georgia State University, ran an experiment on herself. She fed a popular AI detector a chapter of her own 2006 doctoral dissertation — written, one presumes, entirely by a human, because generative AI did not exist yet and neither did the iPhone. The detector flagged it as 39 percent AI-generated.
This is the tool faculty are relying on to adjudicate careers. Manturuk's point, dear reader, is not subtle: if the software can't tell a 2006 dissertation from a ChatGPT session, it cannot be trusted to tell your sophomore's lab report from one either. Detection tools, per the article, are biased, easily circumvented, and prone to false positives — three strikes that would get any other product pulled from shelves.
Honor Codes: Please Do Not Cheat, Sincerely, Management
Then there's the honor code pledge — the academic equivalent of a screen door on a submarine. Students sign a promise not to use AI, and then, Manturuk argues, go straight back into a system engineered to punish the smallest stumble and reward compliant box-checking over actual curiosity. Asking a student not to cheat in a course with no margin for error is a bit like asking someone not to run for the exit once you've announced the building is on fire. The incentive structure does the talking; the honor code just watches.
What Apparently Does Work
Even now, — as if this were not enough — the article does eventually get around to solutions, and none of them involve software. Manturuk points to scaffolded projects, process-based feedback, and mastery-based grading: breaking assignments into stages, grading the thinking rather than just the final product, letting students revise instead of dying on the first draft.
She cites Duke biology professor Mohamed Noor's flipped-lecture model, where class time goes to discussion instead of transcription, and Georgia State's vertically integrated project teams — students collaborating across year groups on real, ongoing problems instead of generating discussion-board filler for an unread audience of one TA. The through-line is human connection: small-group instruction, even inside large lecture halls, so a student's work is legible to an actual person and not just a rubric.
One colleague, Manturuk notes, quit online teaching altogether after students started submitting AI-generated answers to prompts explicitly asking for personal examples — a category of assignment previously thought cheat-proof by virtue of requiring, well, a life. Sources confirm even that is now automatable.
The Diagnosis Nobody Wanted
The article's real thesis, buried under the pedagogy jargon, is uncomfortable: AI cheating isn't a discipline problem, it's a design problem. Students aren't reaching for the machine because they're morally deficient. They're reaching for it because the course offers no room to fail safely and no reason to believe anyone's paying attention. The fix on offer — better course design, more feedback, actual human contact — has been sitting right there, available without any AI involved whatsoever, for roughly as long as universities have existed.
The students are cheating their way through a system that was already failing them. The machine just made it faster to notice.
Sources: Inside Higher Ed — The Best Defense Against AI Cheating
