AI takes root in university curriculum

Learning to use AI effectively instead of banning it

Generative AI is ushering in an entirely new era in education. We spoke with four EPFL teachers about how the technology has impacted the way they’ve been giving their classes this year.

“It hit like a tornado and is upending both our relationship with students and their relationship with knowledge,” says Jamila Sam, who has been teaching coding for 30 years. In her experience, the arrival of generative AI three years ago was a real watershed moment. “It totally threw my colleagues and me for a loop, and we created a working group to address the issue. We’re starting to flesh out some ideas, such as how we can discuss AI with our students, keep them informed and suggest ways of using it. We’re figuring it out at the same time they are.”

Sam believes that the main problem with AI is that it gives you the impression that you’ve learned something. “The programs lull students into believing they can understand a concept without putting in the effort to follow the thought process all the way through.” She sees that reflected in exams, where students’ grades and skills acquisition are deteriorating, and in the online forums that were created during the pandemic. “Two years ago there were around 2,000 messages, but this year, there have been just 200,” says Sam. “Students seem to be looking for answers elsewhere and not digging any deeper.”

Theatrics and colored chalk

Sébastien Basterrechea, a lecturer in mathematics, warns students of the mistakes that AI can make. “For now, I see it as only an additional tool, like MOOCs were a few years ago,” he says. “Before students can properly assimilate a mathematical process, they must have a thorough understanding of it – and that’s not something AI can do for them. This kind of comprehension is what’s tested in exams.”

Basterrechea has two secret weapons for keeping students engaged: dynamic, theatrical lectures and colored chalk. “Mine isn’t just a Monday-morning analysis class – it’s the Sébastien Basterrechea show!” he says. “It’s a hard subject, so the least I can do is make it enjoyable.”

Nothing but pen and paper

“We’re in the process of fine-tuning the rules,” says Johan Rochel, an EPFL lecturer in the philosophy and ethics of new technology. He believes that generative AI has had a particularly disruptive impact on assessment methods. “We can forget about testing students with written assignments to be completed at home. All we would get are essays written by AI, and we wouldn’t know what the students actually grasped.” He believes the key is for teachers to be clear on what skills students are expected to acquire in their class. “Then they can use that as a starting point for deciding which teaching and assessment methods to use and what tools can be brought in.”

Rochel focuses his own classes on in-person lectures complete with case studies, teamwork and discussions. “From a teaching perspective, there are huge benefits to having people sit at the table and discuss things face to face,” he says. His exams are administered the old-fashioned way: two hours in a classroom with only a pen, paper and some written materials – but no internet.

In Rochel’s view, the main risk regarding AI is that skills will be handed over to machines. “If we automate critical tasks, what will happen if there are no more computers? Consistently turning to AI can encourage people to take the easy way out, leading to a loss of skills and AI dependency.” Learning to always check the facts and think critically is essential, but not always enough. “AI programs are designed to hit us where we’re vulnerable and short on time,” says Rochel. 

Enhancing human capabilities

Ken Holstein recently joined EPFL and is working specifically to reshape how emerging technologies are designed and used. He heads EPFL’s Co-Augmentation, Co-Learning, and Collaborative AI Lab (CoALA), which develops AI systems that aim to enhance human capabilities – such as thinking, learning, collaborating, reasoning and communicating – rather than simply automating tasks.

Holstein advocates for systems designed in collaboration with those who will use or be impacted by them – these users could be workers, local community members, teachers or students – based on their specific needs. “AI cannot be a one-size-fits-all solution,” he says. “When companies try to deploy the same AI systems globally for very different cultural, organizational and educational contexts, we often see clashes with local needs and values.”

He cites two major challenges in education. The first is that educational institutions sometimes fail to sufficiently involve experts in technology adoption decisions. Under pressure not to “miss the boat,” schools may end up adopting technology solutions that are not well-aligned with students’ learning needs or with teachers’ instructional needs. The second challenge is that faculty members must design activities and assessments that make it hard for students to simply “outsource” the cognitive effort required for learning to a generative AI system. “In addition, it’s also important to help students deeply understand why outsourcing certain tasks to AI will prevent them from developing the forms of human judgment and expertise that will be most critical in an AI-augmented world,” says Holstein.