Why we need to radically rethink examinations

On September 25, 2025, I will be giving a keynote at the Day of Digital Education in Regensburg. The title: Examinations and AI: From Symptom to System Question.

I will reflect in more detail here in the coming days on my today’s presentation on AI, data protection and examinations – and I will also do so at the keynote in Regensburg. But here’s a preview of what I’ll be talking about in Regensburg.

The starting question is simple and disturbing at the same time: How do we motivate students to learn independently when AI tools are available at any time? After all, individual tutoring is now affordable for everyone. What we observe instead: an increasing externalization of thinking, a declining quality of discourse in courses and – at least in the Bamberg computer science degree programs – sharply rising failure rates in foundational courses.

The intensively discussed countermeasures – such as task assignments that exploit knowledge cutoffs and detailed monitoring through mandatory submission of chat histories – are labor-intensive for instructors but easy to circumvent. They do not provide an answer to the system question posed at the beginning. However, computer science students also need solid basic knowledge in the future so that they don’t become passengers in system development. This certainly also applies to other subjects.

Instructors also face new temptations. The time savings from automatic correction of free-text answers would be considerable, but is legally demanding (AI Regulation, GDPR) and didactically risky (bias, hallucinations). AI-supported optimization of exam questions appears more promising – for example, to identify ambiguities that could disadvantage international students. What increases quality, however, quickly becomes a procrastination trap.

We should radically rethink established formats. I will present three approaches from our practice: First, our booklet system – weekly handwritten study notes that are permitted as personal exam aids. Second, supervised e-examinations in an isolated environment with real tools without AI access, but with unlockable tips and questioning opportunities. Third, supervised multi-hour writing sessions as an alternative to traditional term papers – not perfect, but more authentic than the fiction of independent homework.

My thesis: We need an examination culture that consistently demands performance under controlled conditions like in concert performances or sports competitions. Not only to maintain equal opportunities, but possibly also to protect students from their own weakness of will. The real problem is not AI use by students – it is the illusion that we can assess competencies without performative proof.

After the keynote, I will report here again in more detail about the discussions and new insights. Until then, I look forward to the exchange in Regensburg – also because my positions are not uncontroversial.