AI Isn't Killing Education. It's Exposing What Was Already Broken. - Forbes

Every few weeks, another headline warns that artificial intelligence is gutting academic integrity. Colleges are scrambling to lock down exams, install detection software, and rewrite honor codes. That panic is understandable. But the real story is simpler: AI is forcing education to confront a long-tolerated problem it has avoided for decades.

That problem is regurgitation — and AI has exposed how much schooling still rewards it.

The Real Cheat Was Never AI

For most of modern schooling, an "assessment" has meant one thing: produce the answer, description, or conclusion the instructor already has in mind. In high school, that usually means reciting facts. In college, it means echoing the analysis or framework the professor presented in lecture. Either way, the system has long rewarded figuring out what the grader wants, then supplying it.

That model started cracking the moment information became a Google search away. Memorized facts stopped being valuable currency once anyone could look them up in seconds. Even so, many classrooms kept running on the old operating system, asking students to demonstrate they could reproduce what a screen could already tell them.

AI hasn't created that flaw. It has just made it impossible to ignore. A tool that can generate a competent essay, compare two theories, or summarize a reading in seconds proves the point: regurgitation was never the skill worth testing.

The Uncomfortable Fix: Teachers Have to Do More, Not Less

Here's the part nobody wants to hear: the antidote to AI-assisted cheating isn't better surveillance. It requires more rigorous teaching and a clearer purpose.

At Brown, economics professor Roberto Serrano moved a take-home midterm to accommodate students who felt anxious in classrooms after a campus shooting — and watched the class average jump to 96 out of 100. When he switched the final exam back to an in-person format, the average for students who showed up fell to 48. The institutional reflex to episodes like this is to bring back proctors and lock the room down. But that response treats the symptom, not the cause: it leaves intact an assessment built around a single, gradable, AI-replicable output.

Faculty who are actually solving this problem are doing something harder: redesigning the work itself.

Evan Goldstein notes, in the Chronicle of Higher Education, two examples of professors effectively using AI. At Montana State University Billings, history professor Jennifer Lynn built a dedicated "working day" into her syllabus, where students bring notes and outlines to class to draft in front of her — removing the opportunity for an AI-written paper to simply appear on a due date. At Adler University, clinical psychology professor Richard Niolon requires students to submit verifiable quotations from source material, along with reflections that tie their arguments back to in-class discussions, making the origins of the ideas traceable. Elsewhere, instructors are flipping the relationship entirely, using AI as a sparring partner for peer feedback or as a low-stakes tutor instead of treating it as contraband.

The common thread isn't technology — it's slowing down. Professors who've moved students toward in-class collaborative work, whether debugging code together or wrestling with a difficult text as a group, report something unexpected: a stronger sense of community and less appetite for outsourcing the thinking. A student who can't explain a design decision or describe what they tried and abandoned hasn't done the work — regardless of which tool touched the final file.

What This Actually Looks Like

Across disciplines, educators are converging on a similar toolkit for keeping assessment honest in an AI-saturated environment:

Process-based grading. Break a paper or project into outline, draft, and revision stages, each graded individually. A polished final PDF is no longer the only evidence of learning.

Oral and live examination. Students receive a question on camera, get a short window to think, then explain their reasoning out loud. There's no prompt that fakes real-time comprehension.

Local, personal application. Assignments that ask students to connect a theory to a campus issue, a regional event, or an interview with a working professional are designed to resist generic AI output almost by definition.

Hands-on, offline artifacts. Physical lab work, minimum viable prototypes, and annotated hard copies of readings keep the evidence of thinking tethered to the student rather than to the screen.

Kim Manturuk mentions in Inside Higher Education that at Duke, biology professor Mohamed Noor flipped his large lecture, breaking students into small groups that he circulates through in real time. At Georgia State, a team of instructors built vertically integrated project teams so students could apply coursework to problems they actually care about while building relationships with peers and instructors along the way. None of this is about banning a chatbot. It's about making the work itself worth doing, honestly.

Motivation, Not Detection, Is the Real Variable

It's worth being honest about the limits here. When students are motivated, technology — AI included — is a remarkably effective tool for teaching and practice. When motivation is absent, no amount of proctoring software can fix it. Detection tools chase a symptom; they don't create the meaning that makes a student want to do the work in the first place.

That's the deeper argument for treating AI as an asset rather than a threat: it removes the option of coasting on regurgitation, and it puts the burden back where it belongs — on designing courses that make genuine understanding more rewarding, and frankly easier, than trying to engineer the perfect prompt to fake it.

Teachers didn't ask for this disruption. But the institutions that lean into it — rebuilding assessment around process, dialogue, and application rather than a single graded output — will end up teaching something AI still can't fake: reasoned thinking. The real opportunity AI exposes is not to replace education, but to force it to become more rigorous, more honest, and more human.

15 Ways to Design Around AI Misuse in the Classroom

Detection software will always be one step behind the newest model. A more durable strategy is to design assignments and classroom routines that make outsourcing the thinking to AI difficult, pointless, or beside the point. Here are fifteen approaches educators are using:

- In-class drafting days. Dedicate class time to outlining and drafting under the instructor's eye, so the first version of a paper is never produced unsupervised.

- Staged submissions. Require an outline, a rough draft, and a final draft as separate, individually graded checkpoints, so a single polished document can't appear out of nowhere.

- Oral defenses of written work. After submitting a paper, students explain and defend their choices in a short one-on-one conversation. A student who can't discuss their own argument hasn't done the work.

- Live, on-camera questioning. Pose a question, give a brief thinking window, and have the student answer aloud in real time — a format that resists pre-generated responses.

- Handwritten or in-class exams. Move high-stakes assessment back into supervised time with pen and paper or a locked-down device.

- Require traceable sourcing. Ask students to submit verified quotations, page numbers, or annotated excerpts from assigned texts, tying their arguments to material that an AI wouldn't have access to.

- Tie assignments to in-class discussion. Require reflections that reference specific moments, disagreements, or comments from that week's class session, which generic AI output can't replicate.

- Localize the prompt. Ask students to apply a concept to a specific local event, campus issue, or a person they've interviewed, rather than a generic topic that an AI can answer from training data.

- Use process portfolios. Have students submit drafts, notes, sketches, and revision logs alongside the final product as evidence of the thinking that led to it.

- Assign hands-on or physical artifacts. Lab reports tied to an experiment actually run, prototypes actually built, or fieldwork actually conducted keep the evidence tethered to the student.

- Built-in peer collaboration and debugging sessions. Group problem-solving done live in class produces evidence of understanding that's hard to fake individually.

- Ask "why," not just "what." Design questions that require justifying a choice, defending a tradeoff, or critiquing an alternative approach rather than simply stating a conclusion.

- Rotate and personalize prompts. Generate multiple versions of an assignment or problem set so answers can't be easily shared or copied wholesale.

- Incorporate current or unpublished material. Base assignments on very recent events, unpublished data, or original source material that an AI model wouldn't have seen.

- Make the reasoning process itself the grade. Shift rubrics to weight the quality of a student's reasoning trail — false starts, revisions, and justifications — as heavily as the correctness of the final answer.

None of these approaches requires banning AI outright. Together, they shift the target of assessment away from a single reproducible output and toward the kind of process, judgment, and explanation that a chatbot still can't supply on a student's behalf. The takeaway is simple: design for understanding, not replication.