Generative artificial intelligence is forcing higher education institutions to rethink the way they teach and assess. That is the argument of Dan Schwartz, a specialist in the learning sciences, in an interview published on 23 September 2026 by EPFL (interview by Cécilia Carron). In his view, the issue goes beyond whether or not to allow these tools in the classroom: it is about redefining what students actually need to learn.
A professor of educational technology at the Stanford Graduate School of Education and director of the Stanford Accelerator for Learning, the researcher was in Switzerland in early September for the launch of CLUE, the new joint centre of EPFL and ETH Zurich dedicated to the learning sciences. A useful clarification: this is an interview, not the presentation of a study. The views reported reflect the analysis and suggestions of a specialist, not published experimental results.
A gap between students and teachers
Dan Schwartz points out that in the 5th century BCE, Socrates already feared that writing would destroy people's memory and their ability to tell truth from falsehood. Today he observes several attitudes in higher education institutions: teachers who bring AI into their courses after adopting it in their research, some who develop tools specific to their discipline (for example, a tool that compares a video of a novice surgeon with that of an expert to identify the differences between their movements), others who are mainly trying to prevent cheating, and finally those who reject AI outright.
According to him, the difficulty lies in the fact that "AI tools are particularly good at performing the kind of tasks we traditionally give students". Hence a gap: for some learning tasks, students are ahead; for research and complex problems, it is the teaching staff.
Measuring the ability to learn
On assessment, his position is clear: since students will use these tools in their working lives, "why not also let them use them in the final exam and rethink the way we assess?" He draws a parallel with the debates over the calculator, which already raised the question of what a pupil absolutely must be able to do alone.
He distinguishes between two goals: measuring mastery of taught content, or measuring the ability to learn what comes next. For the second, he sketches out an idea: an assessment that would observe how a person learns while solving a problem. An AI could spot that the person keeps repeating a strategy that does not work, and observe whether they manage to try a new one. This is a prospect, expressed in the conditional, and not an existing system.
What we still need to know
"You need to know enough to be able to interpret what it produces," stresses Dan Schwartz. Some knowledge must become automatic, like understanding a word when reading. Other things, such as proportionality or percentages in mathematics, are fundamental concepts that need to be understood in depth. On the other hand, a procedure with no fundamental concept behind it could, in his view, be handed over to AI.
Courses to be redesigned
"We are still very much in an experimental phase," the researcher acknowledges. What interests him are situations where AI helps humans rather than replacing them. In a lecture, AI can take notes, which he considers of little interest. In small-group work, on the other hand, a system could notice that one person is dominating the conversation and suggest that another student ask a question, or support novice teachers. These are forms of collaboration he describes as "promising", without presenting them as already deployed.
As for the risk of students putting all their questions to AI, he plays it down. But "if the whole course is designed in such a way that the AI's answers are always enough, then there is probably a problem with the course design". He admits that rethinking a course represents an enormous amount of work, while believing that at some point, carrying on exactly as before will no longer work.
Adapting and including
To prepare students for a rapidly changing world, Dan Schwartz points to an underestimated difficulty: recognising that a situation requires learning something new. He argues for an attitude of openness towards novelty, strategies for making sense of new information and a sufficiently solid knowledge base. He also cites an English teacher, one of whose family members was deaf, who used AI to create a tool capable of reading a text and producing a character using sign language, a product too specific to interest a large company.
CLUE, the Swiss framework
CLUE (Joint Center for Research on Learning, Understanding and Education) was officially launched on 10 September in Bern. Hosted at EPFL by the Center for Learning Sciences (LEARN), founded in 2018, it aims to develop multidisciplinary research on human learning, train a new generation of specialists and bring research closer to practice on the ground. Artificial intelligence is among its themes, notably through the RAISE initiative (Responsible AI for Swiss Education). The centre also plans to work on continuing education for teachers. At this stage, these are announced goals.
