We are ending the academic year with the feeling of stepping on ground that has disappeared. What is the value of learning when you can generate an essay that sounds good in thirty seconds? Next year, we cannot limit ourselves to updating content. Brave changes are needed, at the height of the challenges posed by the omnipresence of generative AI in students' lives.
The evaluation system is obsolete. It is a textbook incentive problem. The students themselves say it frankly: if they make an effort and do not use AI in their submissions, they get a worse grade than their peers who do use it. The culture of effort is dynamited when the result is the only objective.
Faced with this, I began to value imperfection. A poorly connected sentence, a half-baked but genuinely thought-out idea, is worth more than an impeccable paragraph that, I'm sorry, I now doubt is theirs. I have a vocation to teach, but suddenly I have become a detective of traces of humanity in the submissions I correct. And this causes me immense emptiness. Is this, really, the best we can offer?
A recent study by researchers from Stockholm University and Hong Kong University has tracked over 25,000 Chinese secondary school students for two and a half years: probably one of the longest investigations to date on the real impact of generative AI on learning. In the short term, it seems efficient: when students start using chatbots for homework, grades rise by about 18% and they finish it 30% faster. The problem arises when exams come: without AI and handwritten. In two years, grades drop by more than 25% for 80% of students. Furthermore, the impact is not distributed equally among all profiles: it is stronger in social sciences than in STEM subjects. Discernment, reasoning, connecting ideas, and arguing; this is where the failure is most evident.
Let me give you an example: undergraduate students, final exam in sociopolitical thinking. They have a vignette with a current news item and I ask them for an essay from a sociological perspective. They always have a blank sheet stapled, where they can make notes before the final written work. What I find is a huge gap between the ideas they have in their draft — names, authors, well-listed concepts — and the written text. The feeling is that activating knowledge to apply it to a case is very difficult for them: if they answer the case, they forget the authors. If they bring up authors and concepts, they mention the case in passing and almost anecdotally. The conclusion is that they have good memory, but their ability to argue fails. The new problem that AI brings is that externalizing cognition is the easiest and most efficient path for a system based on results and speed.
Before falling into nostalgia, we need to create learning environments where we accompany students to appreciate the process and embrace error. Evaluate the journey, the doubts, the comings and goings. This means creating spaces — oral, written, dialogued — where shortcuts are not rewarded or given space. Metacognition, understanding how you think and not what you think, is the reward of the journey. It will be from here that we can accompany them to the next stage: how to use AI as a cognitive scaffold and not as an external brain.
