Education has a friction problem, and software developers just solved it. That is the disaster.
By instantly correcting awkward sentences, debugging broken code, and summarizing dense philosophy texts, generative artificial intelligence has become the ultimate academic shock absorber. It rescues students from the exact moment of cognitive frustration where actual learning occurs. In doing so, these systems act as an omnipresent, digital helicopter parent, hovering over every keystroke and intercepting every potential mistake before the student can feel the sting of failure. The result is a generation of students who are intellectually pampered, computationally dependent, and increasingly incapable of independent critical thought.
We are trading long-term competence for short-term convenience. To understand how we got here, we have to look at the biology of learning itself.
The Mechanics of Cognitive Softening
Human brains do not retain information simply because it was presented to them. They retain it because they had to work to understand it.
Decades of cognitive science research point to a concept known as desirable difficulties. Pioneered by psychologists like Robert Bjork, this principle states that obstacles that make learning more challenging and slower during the initial phase actually lead to better long-term retention and transfer of knowledge. When a student struggles to recall a fact, reconstruct an argument, or find a bug in their code, their brain is forced to build and reinforce neural pathways.
Generative AI short-circuits this entire biological mechanism.
To illustrate this, consider a hypothetical scenario involving two computer science students tasked with writing a basic search algorithm. Student A spends four grueling hours staring at a terminal. They misplace a semicolon, misinterpret a variable scope, search through documentation, and fail repeatedly. Eventually, through sheer persistence and trial and error, they locate the error and fix it. Student B, impatient after five minutes of confusion, copies their broken script and pastes it into an AI chatbot. The chatbot identifies the error instantly, explains the fix, and provides the corrected code.
On paper, both students turn in a working assignment. Both may receive the same grade. In reality, Student A learned how to debug. They developed the mental stamina required to sit with frustration. Student B learned only that a machine can solve their problems on command. The next time Student B faces a novel problem that does not fit neatly into an AI's training data, they will lack the cognitive toolkit to solve it.
The Tyranny of the Immediate Answer
Our current culture of efficiency has blinded us to the utility of being stuck. In an older educational environment, being stuck was an active state. You went to the library, you debated with classmates, or you simply walked away and let your subconscious mind work on the problem.
AI creates a highly convincing illusion of competence. When a student asks a conversational agent to explain the causes of the Peloponnesian War, they receive a beautifully structured, five-bullet-point summary in three seconds. The student reads it, nods, and believes they understand the topic.
This is a dangerous self-deception. Psychologists have long documented the "Google effect"—the tendency to forget information that can be easily found online. Generative AI takes this to an extreme. Because the chatbot synthesizes the information for the user, the student does not even have to perform the minor cognitive labor of filtering search engine results. The brain, which is an energy-saving organ, recognizes that it does not need to store this information because the external tool is always available.
But a brain that contains only pointers to external data is not an educated brain. It is merely a terminal. When the tool is removed—during a high-stakes exam, a job interview, or a real-world crisis—the intellectual scaffolding collapses, leaving behind an empty room.
The Eradication of the Bad First Draft
In the humanities, the damage is even more insidious. Writing is not merely a tool for communicating thoughts that already exist; it is the physical process through which we formulate those thoughts in the first place.
Almost all good writing begins with a bad first draft. It is in the messy, frustrating struggle of trying to drag chaotic ideas out of the mind and arrange them into structured, grammatical sentences that we discover what we actually believe. If you do not write poorly first, you never learn how to think clearly.
When a student uses AI to generate an essay outline, suggest thesis options, or polish their prose, they are skipping the most important part of the intellectual journey. The machine does the heavy lifting of conceptual synthesis, leaving the student to act as a glorified editor.
Editing is a useful skill, but you cannot edit a blank mind. The essays produced through this collaborative process are often grammatically perfect but intellectually sterile. They lack the strange, idiosyncratic leaps of logic, the raw vulnerability, and the productive errors that characterize genuine human development. We are teaching students to produce flawless, mediocre work rather than flawed, brilliant work.
The Quiet Panic in the Lecture Hall
Ask university instructors what they are seeing on the ground, and the diagnosis is remarkably consistent across disciplines. It is not a panic about plagiarism; it is a panic about stamina.
Students are increasingly unable to tolerate ambiguity. If a prompt does not have a single, clear-cut answer that can be fed into a prompt box, they freeze. They struggle to read long-form texts because they are accustomed to having machines distill five-hundred-page books into one-page summaries. Their capacity for deep, sustained attention is being systematically eroded.
This represents a fundamental shift in the power dynamic between student and tool. Historically, educational technologies like calculators or spellcheckers outsourced the mechanical, low-level tasks, leaving the high-level conceptual work to the human. A calculator does the arithmetic, but the student must still set up the algebraic equation.
Generative AI does the exact opposite. It outsources the high-level conceptual work—the synthesizing of ideas, the structuring of arguments, the creative problem-solving—while leaving the human with the low-level task of copy-pasting, formatting, and proofreading. We have inverted the educational pyramid.
Shifting the Metric from Polish to Process
We cannot ban these tools, nor should we try. The technology is here to stay, and pretending otherwise is a form of institutional denial. But we must radically change how we evaluate intellectual growth.
If the final product of education—the essay, the computer program, the laboratory report—can be generated to a passing standard by an algorithm in seconds, then the final product can no longer be the primary metric of student evaluation.
We must stop grading the destination and start grading the journey.
This means a return to analogue, high-friction assessment methods. It means viva voce oral examinations where students must defend their ideas in real time, face-to-face with an instructor. It means in-class writing sessions using pen and paper, where the messy scribbles, the crossed-out paragraphs, and the struggles are visible. It means grading the revision history of a digital document, evaluating how an idea evolved over days of labor rather than how polished it looks in its final seconds.
We must intentionally design difficulty back into the curriculum. We must celebrate the student who fails publicly, struggles openly, and takes the long, inefficient path to understanding. If we continue to allow machines to smooth out every bump in the road, we will find ourselves with a generation of minds that can only walk on flat ground.
