The Medium Is the Mind: What AI Is Doing to Education

All student names in this article are pseudonyms.

Marshall McLuhan’s phrase “the medium is the message” is often treated as a slogan about television or advertising. Its actual claim is more radical. A medium is not a neutral container. Its form changes the scale and structure of human perception and social action[1]. Radio, television, social networks, and short-form video do not merely distribute different content. They train different kinds of attention, create different rhythms of public life, and alter what feels important or real.

Human beings have always externalised parts of the mind. Writing externalised memory. Maps externalised spatial representation. Calculators externalised computation. Search engines externalised retrieval. But generative AI can intervene between confusion and understanding. It can formulate the question, select the evidence, summarise the source, construct the argument, simulate doubt, and produce the conclusion. Earlier media transmitted information. AI increasingly mediates the cognitive process itself.

The crucial question is no longer only what information passes through the medium. It is what happens when the medium begins to perform the thinking through which information becomes knowledge.

Looking at the Future Through the Young

One of the best places to observe the social reality being constructed around a new medium is among the young. As danah boyd observed, teenagers’ path through a networked world “provides valuable insight into how technology is being integrated into and shaping everyday life”[2].

And because youth is institutionalised through schools and universities, habits of attention, inquiry, cooperation, authority, and judgment are gradually turned into social practices that help shape the reality inherited by subsequent generations.

My own interest began with an ordinary university assignment. I was asked to review another student’s essay. The document looked academic: it contained theory, method, analysis, and conclusions. The sentences were grammatical and the paragraphs connected. Yet after reading it, I felt that I had consumed words without receiving information. One generic idea had been repeated in slightly different forms, some references could not be located, and the source books appeared not to have been seriously engaged with. I was disappointed because it had become clear that I had spent my time reviewing AI slop. I wrote at the time: “The student pretends to write. The university pretends to assess. The AI produces the evidence required to maintain the illusion.”

Then came a much larger case.

In spring 2026, Brown University economics professor Roberto Serrano gave 86 students in his welfare economics course a take-home midterm. The historical average had usually ranged from 65 to 80 percent. This time, the average was 96 percent [3].

Serrano and his graders entered the questions into ChatGPT. Its responses resembled many student answers: “kind of correct, but very off and with a very convoluted style,” he said. He changed the final to an in-person examination. Eighteen students dropped the course. Nine remained enrolled but did not take the final. The final average was 48.6 percent, and nineteen students failed. The numbers do not prove that every individual used AI, but the sequence was striking.

What Teachers Are Seeing

Jan Sedenka, a lecturer at the University of Skövde, described a quieter version of the same transformation. In one assignment, students increasingly selected the same arguments and produced similar conclusions.

“In general, students are writing longer texts, but the texts are also becoming more similar to one another,” he told me.

He has redesigned his course, “Sales Techniques and Communication”. Students now submit work, receive feedback, revise it, and discuss their reasoning in seminars. “The process has, in some ways, become more important than the final result because producing a text is now so easy,” Sedenka said.

He had also tested whether AI could generate the supposedly personal part of an academic assignment. “I was shocked by how convincingly AI could produce a reflection even when I gave it no personal input.”

A reflective paragraph can now exist without reflection. A polished argument can exist without intellectual ownership. The external signs of learning are separating from learning itself.

Why Students Cannot Simply Refuse AI

I went into the streets of Stockholm and spoke with several students and their answers made a simple anti-AI position impossible.

Sara, a gymnasium student, has dyslexia. “It can read questions and information aloud, and it can summarize information. That makes it easier for me to understand what we are learning.”

For her, AI does not necessarily remove the educational task. It makes the task accessible.

Elin, a music student, described a different gap. “Not everyone has a parent at home who can help them. Sometimes my parents cannot help me, so I ask AI.”

AI offers what educational systems and families cannot always supply: immediate attention, unlimited patience, explanations at the required level, and help without embarrassment. “You can ask as many questions as you want,” another music student said.

This is not even merely a matter of convenience. A 2026 meta-analysis of 35 experimental studies involving more than 4,000 students found a moderately positive overall effect of ChatGPT on learning outcomes, including problem-solving, critical thinking, engagement, and motivation. Crucially, however, these benefits depended on how the technology was integrated into learning [4].

The French engineering students I interviewed used it to enter unfamiliar subjects and reduce long documents to their central points. Students also described using it to create practice quizzes, structure projects, understand mathematical steps, and see examples of expected formats.

McLuhan’s example of the electric light helps explain why these uses matter. He describes electric light as a medium without content of its own. Whether it illuminates a hospital operating room, a baseball field, a factory, or a home is secondary. Its deeper message lies in the way its existence changes the conditions under which human activity takes place. Electric light extends the day, reorganises work and leisure, and makes forms of activity possible that previously depended on time and place. The medium changes the environment before we even begin to ask what people choose to do with it.

Generative AI may be producing a similar transformation in education. Its significance cannot be reduced to whether a student uses it to transcribe and summarise lectures, explain an equation, generate a quiz, or write an essay. These are its applications, not its full message. The more consequential change is that students increasingly learn in an environment where explanation, feedback, summarisation, translation, organisation, and even argument are available instantly and on demand. Once those conditions exist, the expectations surrounding study begin to change with them.

AI is therefore becoming difficult to refuse not simply because it is convenient, but because it gives students something they genuinely need within an educational environment that is already adapting to its presence. It is available when the teacher is busy, the parent cannot help, the text is inaccessible, or the student is ashamed to ask the same question again. Refusing the tool does not restore the conditions that existed before it. The conditions themselves are changing.

The Cognitive Cost

The same students also described the other side.

“I do not use my brain enough when I use it,” Elin said. “Then you do not have to do the work yourself. I feel lazy when I use it.”

One student described AI as removing “the middle stage” of searching through five or ten sources. Another admitted: “When I use AI, I usually do not check its sources because I trust it.”

This middle stage is often treated as inefficiency. Sometimes it is. But it is also where the student actually learns to distinguish a claim from evidence, compare interpretations and construct a position.

Liana, a university student in Belgrade who also teaches Swedish, makes an important observation about this AI-shift: “I am in an environment where people generally do not rely very heavily on AI, so it is strange for me when I hear that someone has used AI for something extremely simple, something they could easily have done themselves,” she said. “But I have seen other classmates or colleagues use AI even for tiny problems.”

Thus, for many students, delegating cognitive work to AI is no longer a deliberate decision but a default response that precedes the question of whether assistance was needed at all. AI use is becoming automatic.

A 2026 study of 299 STEM students found that routine GenAI use was associated with lower levels of reflection, critical thinking, and the intrinsic need to understand. The concern, the authors argue, is that repeated reliance may gradually turn cognitive delegation from an occasional shortcut into a habitual response [5].

The French students I talked to were equally direct. AI helped them work faster, but one said it “slightly reduces how much we learn” because the system completes part of the work in their place.

Students feel this ambiguity. One admitted using AI to write an initial essay, memorising it, and changing it to sound more human. “I suppose I am cheating,” she said, “but it is mainly because I have a hard time studying at home by myself.”

That sentence contains the whole problem: misconduct, loneliness, practical need, self-awareness, and shame.

McLuhan’s theory starts from the idea that a medium is an extension of the human being. Generative AI introduces something different. It not only extends human capacities such as mathematical calculations or information search. It can partially take over the cognitive processes through which information becomes knowledge. It was designed to be a “helpful assistant,” but this helpfulness can also become unsolicited. AI often generates meanings and offers interpretations even when the user has not explicitly asked for them. In such cases, it can take over part of the cognitive process before the user has completed it independently, while at the same time introducing its own compiled formulations into a space where the user’s own meaning may not yet have formed. If the message of a medium is expressed through the way it changes human perception and action, what happens when parts of perception, analysis, and formulation are themselves delegated to the medium?

This creates a new kind of loop. The medium no longer only changes the environment in which thinking takes place. It begins to intervene in the process of thinking itself. AI does not replace thinking altogether, but it can make parts of the processes through which thinking develops optional. The cognitive cost may therefore lie not simply in doing less work, but in practising less of the intellectual process through which judgment, understanding, and independence are formed. The shift is therefore also a shift in subjectivity: the medium can now produce cognitive content before the human subject has fully formed or articulated it.

The Polarisation Hypothesis

Sedenka made the most disturbing observation. Reflective students can learn more with AI. They arrive with ideas, knowledge, and judgment, then use the system to organise, challenge, and extend them. Students who lack an academic approach may use the same system as a substitute for the process they have not yet developed.

“I would therefore say that the strongest students learn more with AI,” he said. “Students who use it merely as a shortcut may learn less.”

This suggests a hypothesis about intellectual stratification.

AI may not equalise cognition. It may amplify the difference between those who can evaluate it and those who depend on it. One group will use AI to think more deeply, move faster, and control increasingly powerful cognitive systems. Another may produce fluent texts and plausible analyses without developing the independent structures required to understand or challenge them.

The darkest possibility is a society divided between people who retain authority over meaning and people who can only generate convincing forms. The ability to think could become a scarce resource while the appearance of thought becomes a mass-produced commodity.

This is a hypothesis, not a prediction established by four interviews. But the mechanism is already visible.

When Does Assistance Become Substitution?

Universities, with their slow-moving institutional machinery, still seem remarkably unable to address what AI use in education actually looks like. In a digital marketing course I took, the instructor warned us not to use AI for assignments, adding that he would know if we did and that there would be serious consequences. The room responded with collective poker faces. I remember internally rolling my eyes. Having worked closely with these technologies, I knew how unreliable such confidence was. I later used AI while working on the assignment. Nothing happened. The instructor did not detect it.

That experience found an almost exact reflection in my interview with Liana. Asked whether her university gave students any guidance on appropriate AI use, she answered: “No, there are no recommendations. They just say not to use it at all. That is it.” A few moments later, however, she described the practical reality just as plainly: “Everyone uses AI in some way.” The contradiction matters.

The answer cannot be a futile attempt to remove AI from education. Nor can universities continue evaluating final documents as though the conditions of authorship had not changed.

Assessment must move toward verified understanding: oral defence, supervised work, source discussion, revision histories, and questions that change the original conditions. As Sedenka put it, teachers must conduct something resembling an interview: “You gave this answer. What would happen if I changed the conditions?”

The purpose is not to catch a student using forbidden software. It is to establish whether the student can explain, challenge, and transfer the knowledge.

Several students I spoke to articulated a surprisingly similar boundary. “You can get help from it,” Sara said, “but you should not let it do all of the work.” These are good words, but they raise an important question: How should we define the boundary between unnecessary friction that can be outsourced and productive friction that is itself part of the learning process and therefore must be preserved?

McLuhan’s insight returns here. AI’s deepest message is not contained in any individual answer. It is contained in the habits the medium normalises.

The future of education will not be decided by whether AI is present. It is already present. It will be decided by whether a human mind remains meaningfully present inside the work.

References

[1] McLuhan, M. & Lapham, L. H. Understanding Media: The Extensions of Man. (The MIT press, Cambridge (Mass.) London, 1995).
[2] Boyd, D. It’s Complicated: The Social Lives of Networked Teens. (Yale University Press, New Haven, 2014).
[3] Whitford E. Brown Professor Suspects Majority of His Class Used AI to Cheat [Internet]. 2026 Jul 8. Available from: https://www.insidehighered.com/news/faculty/learning-assessment/2026/07/08/brown-professor-suspects-most-his-class-used-ai-cheat

[4] Wu, X., Zhu, P., Zhang, J., Yin, M. & Wang, Y. ChatGPT’s impact on student learning outcomes: a meta-analysis of 35 experimental studies. Humanit Soc Sci Commun 13, 684 (2026).
[5] Choudhuri, R., Sanchez, C., Burnett, M. & Sarma, A. Thinking Less, Trusting More: GenAI’s Impacts on Students’ Cognitive Habits. Preprint at https://doi.org/10.48550/arXiv.2601.22430 (2026).