With support from a Senior Fellow Comenius grant and the Amsterdam UMC Innovation Award, Houben and Koens, both affiliated with Amsterdam UMC, are developing KARLA: an AI-driven tutor that allows students to practise with realistic cases and gradually sharpen their clinical reasoning.
“Students lack the space to make mistakes”
The problem is familiar to many lecturers. Students learn a lot, but have relatively little opportunity to practise applying that knowledge in realistic contexts. Textbooks offer structure, but not interaction. Koens: “In current education, students sometimes lack the space to make mistakes and learn from them. While that’s exactly what is important for developing clinical reasoning.”
“And that need also comes directly from students themselves,” Houben adds. “They want to practise more often, preferably in a setting that’s as close to real-life practice as possible. Not another chapter to read, but experiencing what it means to have a patient in front of you.”
KARLA: a tutor that deliberately doesn’t give the answer
KARLA, short for Klinisch AI-gestuurd Redeneer Leer Applicatie (English: Clinical AI-driven Reasoning Learning Application), was developed in response to that need. Through Canvas, students are presented with a case: a patient, a complaint, a context. Then the real work begins.
“KARLA asks: what are you thinking?” says Koens. “And it doesn’t stop there. Students have to reason, make decisions and substantiate them.”
What stands out is what KARLA does not do. While generative AI often provides quick answers, this tutor is deliberately restrained. “We recently had a student testing the system,” Koens explains. “They said: with ChatGPT I get the answer immediately, but here I really have to spell out my reasoning. And that’s where the real learning happens.” KARLA guides with small hints, asks follow-up questions and keeps students actively engaged. “It’s not about reproducing knowledge, but about reasoning,” says Houben.
Why an AI tutor?
The idea for KARLA originated in Houben’s Comenius proposal, focused on new forms of practice for clinical reasoning. Koens saw an opportunity to apply AI: “Not as a replacement for the lecturer, but as an extension. AI is not a goal in itself,” she says. “But it can add something we currently lack: immediate, personalized interaction at scale.”
Where traditional e-learning is often linear and predictable, an AI tutor can adapt to the student. It can respond to what someone understands (or doesn’t), probe further, slow down, or speed up. “It’s essentially the person you turn to when you get stuck,” Koens says. “A kind of personal coach, always available.”
That search for realism goes further. Houben is also experimenting with AI-generated telephone consultations, in which a ‘patient’ describes symptoms in everyday language. “The difference from a real consultation is hardly noticeable anymore,” he says. “And that’s exactly what we want: a safe yet realistic environment where students can practise.”
Finding the balance in AI-supported education
KARLA is not a standalone tool, but part of a broader movement within VU Amsterdam towards more active, personalized and practice-oriented education. That also raises questions for lecturers. When does AI add real value? How do you ensure students continue to learn, rather than copy answers? And how do you design education where technology strengthens the learning process instead of taking it over?
Projects like KARLA show that this balance is possible. Koens: “We see KARLA primarily as a complement to education. During practicals, there is space for interaction with lecturers and for students to practise together. The AI tutor allows students to do additional practice independently, but it certainly does not replace the lecturer.”
Looking ahead
The development of KARLA will continue over the next two years, with an official start in September 2026. The first phase focused on research and design; the coming period will centre on testing, further development and implementation. The ambition is clear: “Not just to help students acquire knowledge, but to let them experience what it means to be a doctor,” says Houben.
Or, as Koens puts it: “Ultimately, you want students not only to understand what is medically going on, but also to feel: okay, what do I actually do now, with this patient in front of me?”
