AI Didn’t Create the Cheating Problem—It Exposed an Assessment Weakness
It will come as absolutely no surprise to most educators that our education system has not changed in any truly fundamental way for roughly the last 150 years. One could probably make the case that, structurally speaking, it has not changed in a meaningful way for even longer than that. But just stop for a moment and think about how much everything else has changed in that same span of time—technology, transportation, medicine, communication, manufacturing—nearly every sector of modern life except education.
Is there anything else we rely on in our daily lives that has remained functionally unchanged since the antebellum period? When the modern education system as we know it today was taking shape, transportation still relied on horses and carriages. Flight was decades away. Even the telephone would not be invented for several more years. And yet the core structures we still use to educate children—how we group students, how we assess them, what we treat as evidence of learning—were largely designed in that world.
That alone should give us pause. Education is one of the most important institutions in any society, shaping not just individual outcomes but the future of that society itself. One might reasonably expect that something so important would have undergone major structural evolution over time. Instead, most of what has changed has been surface-level. We swapped chalkboards for whiteboards, whiteboards for smartboards. We moved from handwritten diagrams to overhead projectors, then to slideshows, then to projecting everything digitally. These are not useless changes, but they are largely cosmetic. Under the hood, the system has stayed mostly the same.
Now we are trying to educate 21st-century students—students who have instant access to the totality of recorded human knowledge and now AI systems capable of organizing and explaining that information in ways that feel coherent and convincing—using assessment models designed when the bowler hat was the height of fashion. And then we act surprised when those assessments begin to fail under pressure. Cheating, of course, is not new. It has existed for as long as schoolwork has existed and likely always will in some form. But AI did not create cheating; it democratized it. In the past, cheating still required a certain level of competence—enough knowledge to fake understanding, enough skill to imitate a voice, enough awareness to convince someone that the work was their own. Today, every student effectively carries a highly capable tool that can generate convincing academic work on demand.
What this means is that entire categories of assessments that had been elevated to serve as our default evidence of understanding, is now the exact kinds of tasks AI is particularly good at emulating. For years, we assumed that if an assignment was completed, that completion itself signaled comprehension. Not because that assumption was carefully examined, but because historically producing the work did require understanding. In that context, completion and comprehension overlapped closely enough that we treated them as the same thing.
AI breaks that overlap. When a student can generate polished responses, explanations, creative work, or even step-by-step reasoning without fully understanding the underlying concepts, completion can no longer stand in for mastery. This is not a sudden failure of student ethics, motivation, or effort. It is a design issue. The problem is not that students stopped thinking; the problem is that many of our assessments were never designed to survive in a world where thinking itself can be convincingly simulated.
As I often tell my students when they ask whether they are allowed to use AI, “I already know that ChatGPT can pass fifth grade. The question is whether you can.” Work that exists primarily to be graded, rather than to make thinking visible, can now be produced without understanding at all. If assessment is meant to tell us what a student actually knows and can do, then this should concern us—not because AI is uniquely dangerous, but because it reveals how much we were relying on proxies for understanding rather than evidence of it.
At this point, it becomes tempting to frame the issue as one of discipline, motivation, or enforcement, but that is a mistake. If an assessment can be completed convincingly without understanding, then stricter rules, longer assignments, or harsher penalties only address symptoms. What needs to change is not student behavior, but assessment design itself.
If we are serious about preserving the integrity of education, then the task before us is not to chase new technologies with reactive policies, but to rethink what our assessments are actually designed to do. Assessments should exist to demonstrate mastery, not merely to generate a score. As I often remind my students, I am not the one who gives them a grade. They give it to themselves. I simply record the evidence of what they show me.
It was this realization—about what AI reveals rather than what it threatens—that led me to begin studying how assessments might be designed so they cannot easily be outsourced. After hours of reading, thinking, and an entire summer spent building, tearing down, and rebuilding my own ideas, I arrived at a framework centered around five core pillars. I will explore these pillars more deeply in future articles and in my upcoming book, Beyond AI: Elevating Human Intelligence—A Practical Guide to AI-Resistant Assessment Design. For now, I want to introduce them briefly in hopes they might provide a framework for AI resistant academics. They are split into five major, and often overlapping, domains:
Performance focuses on assessments that require students to demonstrate understanding in real time—through speaking, explaining, solving, or interacting. What matters is not just the answer, but the thinking as it happens.
Production emphasizes assessments where students must construct something coherent and complete. Authorship becomes visible when students must make and sustain decisions across an entire piece of work that builds on itself, not just produce isolated responses.
Personalization grounds assessment in a student’s specific context, experience, or perspective. AI generalizes well; humans are specific. The more an assessment depends on who the student is, their background and personality, the harder it is to outsource authentically.
Permutation requires students to adapt, modify, or recombine knowledge rather than simply repeat it. Showing the transfer of understanding between mediums. Changing variables, constraints, or contexts reveals whether understanding is flexible or superficial.
Process makes learning visible over time through drafts, revisions, checkpoints, and reflection. When the path matters as much as the product, authorship becomes much harder to fake and much easier to recognize.
Together, these pillars shift assessment away from disposable outputs and toward evidence of thinking. They offer a way forward that does not rely on banning tools or abandoning rigor, but on designing assessments that naturally foreground human understanding.
Functionally speaking, the goal is not to “change” education for the sake of change. It is to preserve and protect what actually matters. Assessments that protect student voice, authenticity, rigor, and understanding are not just defenses against AI; they are safeguards for the intellectual integrity of the entire system. I have already heard stories from higher education where students submit AI-generated work that is then graded by AI-assisted systems, with humans acting only as intermediaries. That is not education. That is automation masquerading as learning. High-quality assessment must put students back in the role of creators and teachers back in the role of guides with judgment, expertise, and responsibility. The task ahead is not to reject modern technology, but to teach students how to use it wisely—knowing when it can elevate understanding and when restraint itself becomes an essential
intellectual skill.
Comments