Categories

• Arts and Social Sciences
• AI
• Research
• Magazin

Teacher feedback is often more critical, but also more concrete - and perhaps more effective precisely for that reason.

Shouldn't schools be doing more to teach learning strategies anyway?

Schneider: With Lehrplan 21 and the new school curricula, there is already an attempt to move away from merely conveying subject matter and toward cross-disciplinary skills. Society and technology are changing so rapidly that you can no longer study with a single predefined goal in mind. Learners have to be equipped to plan, monitor, direct and reflect on their own learning process. This also includes distinguishing which information is reliable and understanding how to verify AI output. These skills have to be taught in school. This creates tension, because subject-specific pedagogies often place more emphasis on their own subject than others. However, many of today's problems can no longer be solved within a single field alone.

What might cross-disciplinary teaching look like?

Schneider: One goal would be project-oriented work that is based on solving problems. Learners are presented with a problem and have to tackle it from various disciplinary perspectives. That corresponds to a lot of situations that you'll encounter later in the working world: you face a problem, work together with others and have to link together different bodies of knowledge. This would have to start upstream, in how teachers themselves are trained. At the upper-secondary level in particular, there needs to be a stronger move toward skills that aren't tied to individual subjects, but that are cross-disciplinary in nature.

So in the future, will teachers also have to become AI experts?

Weidlich: No, their primary job will still be to teach well. They don't have to be AI experts to achieve that, but they do have to use digital technologies in ways that support learning. AI is part of a larger debate that began back with the internet, Google and digitalization. Information is available around the clock, and that changes the role of the teacher. The teacher is no longer just someone who conveys subject matter, but is increasingly a learning guide, someone who provides structure and a source of pedagogical support. AI-specific challenges arise with exams in particular. When learners can easily produce solid finished products with AI, we have to question what we actually want to assess: is it the finished product or the learning and thinking process behind it?

What solutions are there for exams?

Weidlich: You have to change the incentive structures. When only the most polished possible end product counts, it's natural to choose the path of least resistance. Exams should better capture how students engage with content and build understanding. This constitutes a shift from product to process. There are various ways to do this. Some exams have to be AI-free: on site, oral or with pen and paper. That's a pedagogical decision. In addition, there can also be AI-integrated exams where learners work with AI output, assess it critically, make improvements and reflect on the results.

Schneider: I think we have to make the learning process more visible. So far, our university and school curricula are often geared only toward that one particular exam. But it's also important not to only assess learning processes at the end, but instead to guide and measure students' progress as well. For example, when AI is used, students could bring in AI output and discuss with their teachers what's correct, what's problematic and what could be done better.

How can we support instructors and teachers through this transition?

Schneider: We need pedagogical support across the board - not as dictates handed down from above, but more in the form of supportive guidance. You can look at what teachers are already doing and identify opportunities for small adjustments: building in self-tests, using study-partner systems, structuring lectures differently, enabling feedback loops. Changes of this kind can create added value. At UZH, we are building an AI tutoring system where instructors can jointly develop courses and exchange ideas about how AI can be meaningfully integrated into teaching. And together with the Education and Student Affairs Office, I'm working on a project to advance innovative ways of designing courses. UZH is also working on AI-based support systems, for example an AI buddy to help students structure their studies and to show instructors things like which courses cover similar content and where there are points of thematic overlap. This could help knowledge become more interconnected instead of remaining siloed off in isolated courses.

When an AI tool generates student output and another AI tool evaluates it, you get a closed loop within the same knowledge space.

If even people's buddies are powered by AI, what role do social factors still play in learning?

Schneider: Perhaps the biggest role of all. We are social creatures, and we seek interactions with others in order to exchange knowledge and create meaning. AI-based learning is based on probabilities. A human teacher, by contrast, can respond better to a learner's history, overall situation and emotional state. This capacity for empathy is something that AI can only simulate at best. That's why the human in the loop remains central. When an AI tool generates student output and another AI tool evaluates it, you get a closed loop within the same knowledge space. Universities have to be able to think beyond this, in new directions.

Joshua Weidlich, you recently conducted a study about how feedback on learning outcomes is perceived differently when it comes from AI, peers or teachers and what the effects of this are. What did you find?

Weidlich: We compared reactions to the different kinds of feedback and found that feedback from AI and peers - meaning fellow students - was perceived more positively than feedback from teachers. Our study found that on average, teacher feedback was seen as more critical and less fair than AI or peer feedback. At the same time, though, teacher feedback was linked to the strongest improvements when students had to revise the assignment. That fits the pattern mentioned earlier: what feels good while learning isn't necessarily the most helpful. One explanation for these results could be that AI writes friendlier responses. A follow-up study did in fact show that AI feedback contains more positive statements. By contrast, teacher feedback is often more critical, but also more concrete - and perhaps more effective precisely for that reason.

What role do emotions play in feedback?

Weidlich: They're an important component. Good feedback also contains criticism and can therefore be unpleasant. Feedback that's too insulting will be rejected. Many of us are familiar with how people initially react to critical feedback: saying it's not true, not wanting to hear it. Only later does real engagement with the feedback occur. Being skilled at giving and receiving feedback also means being able to navigate this tension.

Schneider: In general, it's not only positive emotions that are important for learning. Confusion can also promote the learning process. It feels negative at first, because you don't understand something. If learners accept and try to resolve this feeling of irritation, a deeper understanding can emerge. Emotions are closely tied to motivation, interest and knowledge.

Given the rapid advance of AI, will teachers even be needed in the future?

Schneider: Yes, and if anything, the developments around AI underscore the importance of teachers even more. We need people with subject-matter and pedagogical expertise to direct, structure and provide critical guidance to learning processes. At the same time, the role of the teacher is changing. A teacher still has to convey foundational knowledge and provide a sense of orientation. After that, they can step back a little and act as a learning guide. So, a teacher is not only someone who imparts knowledge, but also a critical examiner and a source of motivation and social support.

To summarize, where is it sensible to use AI in schools and universities, and where should it be avoided?

Weidlich: AI is useful when it supports learning and activates learners. It becomes counterproductive when students outsource key thinking and learning processes without doing any reflection. They may get a feeling of having learned something, but the critical learning processes didn't actually take place. You often only notice this in an exam, when AI support isn't available. That's when it becomes clear what students actually learned. This is exactly why we need pedagogical frameworks, clear learning objectives and people who can provide critical guidance on how to use AI.

Interview: Thomas Gull und Roger Nickl/Translation: Gena Olson

More about the interviewees

Digital transformation in teaching and studies - using AI innovatively