What university do we want in the age of AI?

Universities five or ten years from now will be very different from the ones we know today, and yet we keep planning them as if the world were changing slowly. The recent discussion about our five-year institutional development plan inevitably reminded me of the Soviet Union’s five-year plans: meticulous documents that tried to anticipate everything, from steel production to the potato harvest. They were exercises in control over a future that even then resisted being tamed; today that illusion is even more fragile. If for much of the twentieth century decades could pass between a scientific discovery and its social impact, in the last ten years we have seen technologies born, mature and become obsolete within the span of a single university degree.

In that context, it is striking that many university strategic plans mention artificial intelligence almost as an obligatory nod, with no concrete indicators or actions reflecting that AI is already here, sitting like an elephant in the room. It is not a matter for the future: students already use it to do homework, study, translate, summarize, program and even decide which courses to take. Faculty use it too — sometimes quietly — to prepare materials, generate code, design rubrics or draft reports. The question is no longer whether AI “will arrive” at the university, but what we will do with the fact that it is already part of the learning ecosystem, whether we like it or not.

This demands rethinking how we conceive the transmission of knowledge. I always tell my students that they did not come to university to accumulate formulas, but to learn how to learn. Today it makes very little sense to assess the ability to memorize, or to solve by hand a particularly convoluted integral that a piece of software, or an AI assistant, solves in seconds. Far more relevant is that they know in which situations they need to set up an integral, how to model a physical phenomenon, how to translate a real-world problem into mathematical language and, above all, how to question whether the model they are using makes sense. Critical skill lies not in the last line of the calculation, but in the first lines of how the problem is framed.

The difference is not rhetorical; it shows up in everyday teaching practice. For almost two years we were developing an online General Physics course (FIS100) taught in the summer. One of the most exhausting tasks was writing hundreds of questions for practice quizzes, each with detailed feedback, all coded in LaTeX and in the right format to upload to the institutional platform. Many hours of academic work went into finally building a bank of more than a thousand carefully reviewed questions. At the time, that was what had to be done if we wanted to offer a high-quality, interactive course with plenty of practice.

Today, while I teach the Thermodynamics and Heat course (FIS131), I can ask an AI tool to generate twenty questions at the level of a classic textbook like Sears and Zemansky, with feedback included and in a format ready to import into our question bank. Within minutes I have them on the platform, available for students to practice ahead of a test or exam. It is quite likely that the online course that took us almost two years to build could today be designed in half a year, maintaining or even improving the quality of many of its components. The difference is staggering: where we once had a bottleneck of academic time, we now have an assistant that multiplies our production capacity.

The experience is not exclusive to teaching. In the world of programming, AI already generates blocks of code, suggests functions, detects errors and speeds up tests that previously required entire teams of junior developers. In administrative and management tasks, it automates report writing, consolidates databases, schedules meetings, sorts emails and even drafts internal policies. In service companies, it takes over part of customer support, the initial handling of queries and the production of standard reports. It is not that AI “brushes against” these areas: it is already replacing concrete tasks, reducing the need for certain entry-level positions and forcing workers to move toward more complex, more creative or simply different roles.

If we extrapolate this trend a few years ahead, the university landscape changes profoundly. Within five years most universities will probably offer hybrid courses where the basic content leans heavily on AI-generated or AI-curated resources, while in-person sessions concentrate on discussion, problem-solving and projects. Within ten years it is not far-fetched to imagine personalized systems that accompany each student throughout their entire trajectory, proposing learning paths, materials adapted to their level, exercises targeted at their gaps, and continuous assessments far finer than our current tests and exams.

Faced with this, the question we should ask ourselves as an academic community is not so much “how do we defend ourselves from AI”, but “what kind of university do we want to build in this new environment”. If we let external forces decide for us, we run the risk of becoming mere certifiers of standardized competencies generated by outside platforms. If, on the other hand, we meet the change head-on, we could concentrate human time — the scarcest resource — on what AI still does not do well: shaping judgment, guiding vocations, accompanying personal learning processes, cultivating curiosity and sustaining spaces for critical thinking and interdisciplinary dialogue.

That means honestly reviewing our institutional development plans. It is not enough to mention artificial intelligence in one line of the diagnosis or in a generic objective. We need concrete decisions: how does assessment change so that it makes sense in a world of omnipresent assistants? Which transversal skills do we want to set our graduates apart in ten years, beyond today’s specific techniques? What continuing education will we offer our own faculty so they can integrate these tools without fear and with judgment? What research do we want to drive around the social, ethical and economic impact of these technologies?

The future of the university is not written, but its room for maneuver shrinks if we keep planning as if nothing had changed. Perhaps the real lesson of the old Soviet five-year plans is not that planning five years ahead is useless, but that it is dangerous to do so as if the future were a simple extrapolation of the past. The university we will be in five or ten years depends less on guessing the exact curve of AI adoption, and more on daring today to answer an uncomfortable question: are we really thinking about the university we want, or just updating, once again, the university that already was?