AI Did Not Introduce Cognitive Offloading. Education Had Already Set the Stage.
Between the Horns
Tim Moon
Between the Horns
I have been trying to grapple honestly with the role of AI in education, without catering to either side of the current dilemma. Each provides a poor alternative.
This debate appears like a bull. One horn warns that artificial intelligence will permanently weaken human thought. The other promises that it will personalize learning, democratize expertise, and finally deliver the educational revolution we have been promised for decades.
Each horn sees something real. A student who repeatedly hands memory, writing, interpretation, and judgment to a machine will lose something. A faculty that is never exercised will weaken. AI does make it possible to escape difficult thinking faster and more completely than any previous classroom technology.
The promise has substance too. A patient tutor that can question a student at any hour, explain an idea in several ways, expose a weak inference, or generate a serious counterargument is not a small thing. Used with discipline, AI can deepen a student’s encounter with an idea.
But we have seen the larger promise before. The computer, the internet, and the tablet entered school carrying hopes of access, personalization, and liberation. They brought real benefits. They also brought distraction, passive consumption, shallow research, and entire classrooms staring into portals. No machine has ever relieved us of the human work of education.
We are often told to take the bull by the horns. I think that is the wrong move here. The horns appear to point in opposite directions, but they are attached to the same head, and the premise in that head matters more than either prediction.
Go for the Head
Doom and utopia share the same skull. Both assume that AI determines the educational outcome.
In both stories, the teacher becomes a bystander, the student becomes a recipient, and the school becomes a place where technology happens to people. The two camps disagree about whether AI is good or evil, but they grant it the same sovereignty.
AI changes the conditions of learning. But it does not relieve us of deciding what learning is for, which struggles are formative, and who remains responsible for the thought.
That does not make the tool neutral. AI answers quickly, smooths language, resolves uncertainty, and makes completion easier. Tools train the people who use them, and a tool designed to reduce effort will usually do so unless the teacher intervenes with clear instructions, sequencing, and direction.
Once the tool receives most of the causal power, human responsibility thins out. Schools can blame AI for corrupting students. Vendors can credit it with saving them. Both avoid the harder question: What forms of thinking must students still do for themselves, and how will a teacher know that the work was truly theirs? That question leads backward, before AI, to the habits schools had already built.
The Stage Was Already Built
Cognitive offloading is the transfer of mental work to agents outside the person. We offload memory to a note, calculation to a calculator, direction to a map, and judgment to a rule. Offloading is not always foolish. Human civilization depends on tools, records, institutions, and other people carrying work no single mind could carry alone.
Education still has to decide which work develops the student. That work cannot be surrendered without cost.
Long before ChatGPT, schools had begun transferring more of that work away from the learner. Memory moved to the handout and then to search. Judgment moved to the rubric. Argument moved to the five-paragraph formula. Attention moved to the screen. Knowledge became something temporarily retrieved for an assessment and then released.
This history is uneven, and no serious argument should romanticize every older classroom. Traditional schooling often confuses mechanical rigor with cognitive rigor, obedience with learning, and memory with understanding. Still, the larger direction is hard to miss. We built systems that rewarded the appearance of learning while asking too little about the reality beneath it.
A polished paper stood in for understanding. A passing score stood in for durable knowledge. Compliance stood in for intellectual ownership. The student learned to produce the acceptable artifact because it was the ticket to the show.
Then AI arrived.
It did not write the script. It stepped onto a stage education had already built. The scenery was in place. The cues had been given. The role was waiting: produce the paper, complete the worksheet, answer the question, produce the artifact.
We are blaming the actor for performing the role education had prepared for it.
Granted, a late arrival can still make a bad play worse. AI can industrialize cognitive offloading. It can create the appearance of thought without the inconvenience of thought, at speed and in language polished enough to fool both student and teacher. The prior weakness makes the present alarm more serious because the tool can now scale what the school had already normalized.
Friction Is the Point
Intellectual strength grows through resistance. We remember by recalling. We learn to write by struggling to give form to an idea. We develop judgment by making distinctions, living with uncertainty, testing our reasoning, and standing on conclusions.
AI can remove that resistance before the student has fully felt the problem. Ask for an essay, and it writes one. Ask for an interpretation, and it supplies one. Ask it to decide, summarize, compare, or conclude, and it will often perform the very act the student was supposed to practice.
Some friction is merely mechanical. Formatting a citation, correcting a transcription error, or repeating a routine calculation may consume time without deepening understanding.
Formative friction is different. It requires the student to remember, explain, distinguish, defend, and revise.
In-class exams that require students to write out answers to questions are formative for memory. Short in-class essays that capture a student’s feelings, convictions, and patterns in thinking are formative.
Reverse engineering an essay is another formative tool. Providing a well-written essay and having a student write a prompt by hand to reproduce a similar essay can be very demanding.
Teaching students to ask the right questions is as critical as teaching them to provide the right answers. Writing a prompt that can closely replicate such an essay is formative. These are just a few ways to retain and restore cognitive friction in the AI Age.
Asking students to answer questions and defend responses orally also creates formative friction.
This is how seasoned traditional, or classical, teachers from the last century would have handled the challenges of AI in this century. They would not have panicked or been intimidated, nor should we.
The governing principle is friction relocation. Use AI to carry some of the mechanical work, but the thinking effort that develops the person must remain with the student. Grade the process along the way to a polished product, and weigh the process higher than the finished product itself.
Sequencing matters. The student should encounter the question, attempt an answer, recall what is known, and commit to a claim before asking the machine for help. AI can then expose a weak inference, ask for evidence, introduce a counterargument, or test whether the student can explain the idea without borrowed vocabulary.
After the tool has helped, the teacher should be able to ask a simple question: What work remains inside the student? If the answer is only approval, selection, or cosmetic editing, the machine carried too much. When the student must still explain, defend, and take ownership, the tool may have made the work harder in the way that matters. That standard makes the proper role for AI much clearer.
Interlocutor, Not Scribe
The strongest educational role for AI is to be an interlocutor, not a scribe. A scribe produces language on the student’s behalf. An interlocutor meets the student in dialogue and returns responsibility through questions.
A scribe says, “Here is your answer.” An interlocutor asks, “What do you mean? What evidence would change your mind? Where does your claim contradict itself? Can you explain it without the vocabulary you just borrowed? What is the strongest case against you?”
That is closer to the work of Socrates than the work of a ghostwriter. It provides the teacher another instrument for pulling thought from the student, provided it is visible and transparent.
The process must be visible. A finished paper is no longer reliable evidence of the thinking that produced it. Teachers need drafts, oral defense, in-class work, remembered knowledge, conversation, and a record of how a claim changed and developed. The student becomes the object of education, and the finished product becomes one piece of evidence in a larger record.
That leads to an irony schools should face. Using AI well may require less general computer use at school. Access to the machine should be limited and targeted. Initial thinking will often need to happen by hand, in speech, or in an observed classroom setting before AI enters the process.
A consistent written and oral record of each student provides stronger evidence of authenticity than a detector or a single polished paper. AI may help the teacher organize that record, but the teacher interprets the growth, and the student retains ownership of the work. Over time, the accumulated record gives the teacher evidence that no isolated product can provide.
The Harder Place
The space between the horns is not safe. It is the place representing the full weight of a bull.
A blanket ban protects familiar assignments. Broad adoption allows a school to buy access, announce innovation, and accept the vendor’s promises as a theory of learning. Both choices are easier than redesigning the intellectual life of the classroom.
The alternative requires teachers to separate mechanical work from work that forms and reveals thought. As a general rule, formative processes like thesis development, rough drafts, etc, should occur predominantly at school, where the teacher can gauge the process. Mechanical work can be done at home.
This has a price. It will consume class time and slow coverage. Teachers may need to assign fewer papers and examine them more closely. Schools will need time for conversation, handwritten work, and oral defense. Students will lose some of the convenience of completing important work privately. Administrators and parents will have to stop treating a large stack of polished products as proof of learning.
Unless schools are willing to pay that price, responsible use will remain a slogan, and the adults who design school use will continue placing the burden on the tools.
Who Still Does the Thinking?
AI did not introduce cognitive offloading into education. Education had already set the stage. The technology entered a system practiced in mistaking performance for formation, and its default use will deepen that mistake.
The next decision is practical. It concerns when AI enters the work, what it is allowed to do, what evidence the teacher keeps, and which capacities the student must continue to exercise.
Pedagogy determines who still thinks. By pedagogy I mean the full human design of the project or curriculum, the purpose of the lesson, the sequence of effort and help, the limits placed on AI, the evidence a teacher requires, and the responsibility the student retains.
AI should enter after the student has recalled what is known, struggled with the question, and made a first attempt. At that point, it can question, pressure, and extend the student’s thought without becoming its author.
The future of education will be decided by whether schools can recover and document a clear account of the thinking that forms a person. AI arrived on a stage that education already built. What happens next is still our responsibility.