For centuries, university education in Europe was in Latin. Not because someone flaunted it, but because Latin was the language of science, French language which made it possible for a student from Prague to read the work of a professor from Paris, for an Italian mathematician to discuss with an English astronomer, without translators and without borders. Latin was not an ornament but the infrastructure of intellectual life.
And when, in the 18th and 19th centuries, national languages began to take over its role, part of the academic community resisted it with arguments that sounded convincing at the time. Latin maintains intellectual discipline. Prevents superficiality. It preserves the continuity with the ancient tradition. All of this was true, and none of it mattered - while this battle was being fought in the universities, the world outside the walls had already made up its mind. Scientific works began to be written in English, German, French. Engineers designed in their own language. Trade contracts were drawn up in a language that the contracting parties understood. Universities that insisted on Latin became places where knowledge was stored, but not where it was born. Those who switched to living languages, although it seemed like a capitulation at the time, became dominant.
Two centuries later, we are at a similar crossroads. The subject of dispute, however, is no longer language but something more subtle: the cognitive infrastructure of education. In less than three years, artificial intelligence has become part of the daily work of engineers, programmers, doctors, lawyers, and journalists. And yet in classrooms and amphitheatres, an almost identical debate is taking place. Allow artificial intelligence? Ban her? Detect? Restrict to specific subjects? Are students using AI cheaters or victims of bad professors? Are the professors who use it innovators or is it a betrayal of the educational mission?
This debate is already lost, as the one about the Latin language was once lost. Not because AI is good or bad, but because the world outside the university has decided.
The question, therefore, is no longer whether to allow AI in education. The question is: if AI is now part of the intellectual infrastructure, what exactly does it change in the way we learn and teach?
THE MYTH OF THE ESSAY
At the outset, it is worth resolving an assumption that distorts the debate. The image that the university functioned without serious challenges before ChatGPT is not correct. The written essay, which has become the center of much debate in the last two years, relies on the assumption that a student sits alone, for hours at a desk, and writes. That assumption, hands down, has not been true for a long time. Services like Chegg, Course Hero, and even simply paying older students to complete an essay for a student have been around for decades. What has really changed, therefore, is not the appearance of cheating, but its democratization. For decades, students from wealthier families had access to tools that made their work easier. AI has, in one fell swoop, made that access available to everyone. The assumption of the essay as an original independent work has become unsustainable for everyone, not just some.
Is it then possible to go back to pen and paper, na proctored exams, banning AI in classrooms? Those who propose it start from the assumption that if the university does not let AI in, it can continue to operate according to the old model while something completely different is happening outside. This assumption does not hold. The university is not a closed institution. It exists to prepare people for the world beyond its walls. Young engineers, doctors, lawyers, economists spend four or five years at university, and then enter a profession where they are expected to use the tools used by that profession on their first day of work. A degree earned in an environment where AI is not mentioned is not a degree that protects tradition. It is a degree that does not prepare for a job.
LEARNING WE HAVE KNOWN UNTIL NOW
Anyone who has studied can remember hours spent over one page of a book, over one formula, banging their head trying to understand what the writer actually wanted to say. The book is written for the average reader, with the author's assumptions about students' prior knowledge and examples from personal milieu. When a student hit an obstacle, there were three options. Wait for the consultation, look for another book or, most often, give up and go over the problematic part, with the hope that everything will somehow come together later.
AI solves that problem in real time. A student who does not understand one passage can look for five other explanations, with five different analogies, with examples adapted to his own prior knowledge. Hours and days wasted on the same page are no longer inevitable, which essentially changes the quality and speed of learning itself.
The second problem that AI solves is more subtle. It's about what cognitive psychologists call unknown, holes in knowledge that the student cannot detect on his own because he does not know what he does not know. Bloom's 1984 study documented that one-on-one tutoring outperformed group instruction by two standard deviations (the average tutored student outperformed about 98% of group-taught students). With the help of artificial intelligence, the student can ask his digital tutor: "Ask me a question to which I would give the wrong answer", or: "List the mistakes that students most often make in relation to this term". A hole that would drag on for semesters disappears in twenty minutes.
There is also an aspect that is rarely acknowledged: shame. Students are often embarrassed to ask the same question a third or fourth time, especially in a group. With AI, that obstacle no longer exists. Many things remained unexplained during the studies precisely because the student did not have the courage to ask again. Learning has not been the same since the advent of AI.
A TOOL THAT BALANCES
For centuries, the quality of higher education was extremely unevenly distributed. The best professors taught at a handful of institutions, and those institutions were entered through narrow doors, geographical, financial and class. The rest learned from those that the environment offered them.
The Massive Open Online Course (MOOC) revolution through platforms like Coursera and edX promised to solve this, but failed. Stanford's video lecture was still one-way: you could watch, but when you got stuck studying, the professor wasn't available.
AI is the first tool in history that democratizes not only information, but cognitive service. The same GPT is available to a student in Novi Sad and a student in Cambridge, with the same patience, the same explanations and the same ability to adapt to the student's level of knowledge. But not everything is equal. Access to laboratories, a circle of selected peers, a network of professional connections, and the prestige of the degree on the labor market remain stratified. What is rising globally is the ceiling of competence.
There is another dimension of this equalization that is perhaps even more important than the spatial one. It's about time. A forty-five-year-old engineer who missed a certain field can now master it in a few months with an AI tutor. Until recently, this would have required a return to school or a long absence from work. The ability to update competencies becomes more important than initial education.
photo: apHarvard
THE ILLUSION OF COMPETENCE
But from that same democratization comes something that needs to be seriously considered. It would be naive to say: if AI knows everything, why should we remember? Why should a student spend years acquiring knowledge when an assistant who knows more is just a click away? This logic, as reasonable as it sounds, leads in the wrong direction. AI is not a substitute for knowledge. AI is a multiplier of the knowledge we already have. And multiplication has one property worth understanding: if you multiply zero by one thousand, you get zero. The difference between a doer and a non-doer, when both are using the same AI, is not marginal but essential.
A physician familiar with internal medicine or a lawyer familiar with case law can ask AI for analysis and immediately recognize when the tool is hallucinating, when it is offering an argument that does not hold up in a given jurisdiction, when it is missing something specific to the case before them. The one without that foundation reads the same answer and nods, and what's worse, he doesn't know how much he missed, because the answer still seems authoritative to him.
It is a phenomenon that cognitive psychologists call the illusion of competence. AI systematically produces a sense of familiarity without real effort. The student reads the explanation, everything seems clear, the logic follows and the subjective feeling is: I understood this. But what the student actually experienced was the recognition of meaningfulness rather than the production of understanding. When a student tries to explain something on his own, he encounters holes, questions he doesn't know how to answer. That conflict is learning. AI eliminates it.
Empirical evidence is accumulating. A University of Pennsylvania study found that students who used ChatGPT for practice solved 48 percent more problems, but scored 17 percent lower on a comprehension test. A 2025 study by Michael Gerlich found a negative correlation between the use of AI tools and critical thinking, especially among younger users.
And the consequence is more serious than it seems at first glance. The gap between an average and a great expert, with AI, can grow by an order of magnitude. AI does not equalize competence, but highlights the value of fundamental knowledge. It was naively expected that the tool would raise everyone's level of knowledge and skills and thus bring the top and the middle closer together. The reverse happened. For the best, AI has become a multiplier, one human now does what until recently required ten, and the market pays them dramatically more for it. For those in the middle, that same tool erases what set them apart, because what they do can now be done by almost anyone, so the market pays them dramatically less.
WHAT CHILDREN SHOULD LEARN
What does all this mean and what should we teach young people? The naive opinion is - and children often use this argument - if we have the answer to every question in our pocket, why remember dates, formulas or grammar rules?
This is an epistemologically serious mistake and takes us back to what it was in 1987, in the book Cultural Literacy, formulated by the American educational theorist ED Hirsch. Comprehension is not a general reading skill, it is a function of background knowledge. A reader who knows what interest rates and bonds are can read a financial news story and understand its implications. A reader who does not know this will read all the words, but will not understand why the news is important. Without memorized facts, complex thinking becomes not only ineffective, it becomes impossible, because the time it takes to Google something or check with an AI breaks thought into fragments, and thought requires continuity.
But it's not just about speed of thought. There is another reason why knowledge must remain in the human head and not just in an external tool. People don't read books by turning them into a neutral summary. Ten readers of the same book do not remember the same book. It is not only Heraclitus' river, but also a natural fact about reading: what a person takes away from the text depends on what he entered the text with. Everyone understands it through their own prior knowledge, profession, temperament, generational experience, language and social position. That imperfection of human memory is the source of different interpretations. Culture, science and public discussion do not live on everyone extracting the same essence from the text, but on the fact that different people see different problems in the same text.
AI, in contrast, tends to reduce text to a smooth, average version of meaning, delivered with false certainty. This is useful when we need a quick factual review or understanding of technical documentation. But if society begins to learn from summaries, and then from summaries of those summaries, knowledge does not deepen, but rather stagnates. The problem isn't just that AI can make mistakes. The problem is that it can erase precious differences in human interpretations. What appears to man as limitation, imperfect memory, selectivity, even the noise of personal experience, in education is often an advantage. This is precisely why knowledge must retain the human core.
photo: apHarvard
WHAT THE UNIVERSITY IS ACTUALLY SELLING
From all this follows the question of what happens to the institution of the university as a whole. The mere transmission of knowledge becomes commodified. Mentoring, expert judgment, professional guidance are becoming rare and precious. A university that sticks only to the mere transmission of information loses its purpose. The one who opts for depth becomes more expensive, but also more valuable.
Today, our faculties work according to the model of four plus one or three plus two, with an accreditation period of seven years for the entire program. This means that an electrical engineering student who enrolls in 2026 will graduate in 2030 from a program formed in 2020 or 2021. In normal times, this is tolerable. In the era of AI, that's enough to make a degree out of date on the market.
The proposal is an engineering one. Align the accreditation cycle with the speed with which the scientific field is changing. For the first two years, the basic knowledge and foundations of the profession remain on the seven-year cycle, because that knowledge does not change. The third year, advanced basic and initial production knowledge, is changed and accredited every three years. Fourth year, advanced knowledge actively sought by industry, for one year. Master's studies, the most advanced knowledge used by top companies and research organizations, also for one year, with the continuous possibility of complete revision. Different layers of the system have different update frequencies. The base remains stable, the top is constantly updated.
Let's go back to Latin. For centuries the universities refused to leave it, and the world outside the walls decided long before them. The institutions that were slowest to admit it paid the highest price. AI does not eliminate the need for education, it multiplies it. But the multiplier only works where it already has something to multiply. Never in the history of education has it been more important to leave college with the knowledge that the program really entailed, not with a diploma as a piece of paper. With AI, the difference between the two becomes visible on the first day of work.
Serbian faculties cannot win the global race for AI infrastructure. They can, however, quickly recognize that the depth of mentoring has become more important than the size of the amphitheater, that the number of students is less important than the network being built, and that the defense of "traditional teaching" is actually the defense of a model that the world is abandoning. Adaptation does not ask for money that we do not have. He asks us to align teaching with the world we are letting students into, instead of the world they came from
professors.
The author is a professor at the Faculty of Technical Sciences in Novi Sad
