Refocus on the student, rather than on the testing
Generative AI is profoundly changing higher education, but an excessive focus on secure testing can be harmful, writes Nicos Starreveld. “The real danger lies in constantly emphasizing and testing purely content-based performance and knowledge.”
I still clearly remember experimenting with ChatGPT for the first time in November 2022. I was very curious about how this new program worked and wanted to see how it handled mathematical problems. Since then, this technology has only improved. Today, AI models can generate code, text, images, and even mathematical solutions – currently at an Olympiad level. With generative AI (GenAI), I could do everything faster and better.
Yet I noticed that, as ChatGPT got better, I increasingly longed for independence and control. At some point, I also realized that after a few weeks, I had retained very little of what I had created with the help of GenAI. This feeling is supported by recent research and this alarming article in Trouw. After four years, I mainly ask myself: am I more productive or just more rushed? This led to a period of reflection on my use of GenAI. Reflection without ChatGPT.
UvA’s positions
To determine my own stance as a lecturer in the AI discussion, I first familiarized myself with the UvA’s positions. The Policy Framework and GenAI Guidelines in Education provide initial guidance for lecturers and students regarding AI. They also give a preview of what we can expect in the new educational vision of the UvA, which will be published in September.
If you – by comparison – read the current 2017 educational vision, you mainly see an emphasis on an ambitious learning environment, research-based teaching, and programs developed around the intended learning outcomes for students. These are important positive steps with a significant impact on how programs are structured. At the same time, I was surprised by how little attention that vision paid to building a trusting relationship between lecturers and students. Mutual trust and getting to know one another have always been central to the way I design my own teaching.
Concerns
There are currently concerns that students are not using AI tools responsibly. Examination boards want to be sure that students cannot commit fraud. Students, in turn, want to see that programs make every effort to safeguard the value of their degrees. All of this is reflected in the policy framework. Much attention is given to students’ responsibility for the work they submit. Take-home assignments are on a blacklist. There are also many concerns about theses, which in some cases have been written by ghostwriters.
Of course, we should not deny that these challenges exist. But as a mathematician, I always try to identify the axioms: the core principles on which a theory or discussion is based. Is secure testing the main axiom in discussions about the use of GenAI? Or is it perhaps the learning and well-being of lecturers and students? Or maybe a more efficient use of resources and staff? If it’s a combination of all three, where is the golden mean?
After reading various reports, my impression is that—at the institutional and policy level – most attention goes to AI-secure testing. As a lecturer, this worries me greatly. Placing secure testing at the center signals that we do not and will not trust our students – a mindset that will be highly detrimental to higher education.
Knowledge institution
Of course, the quality of a program’s assessment is very important. But we must not forget that the university is, first and foremost, a knowledge institution – not a competitive testing center. Secure testing is not an end in itself! The student as a person, their learning, and their well-being should always be central. The new educational vision offers a unique opportunity to put trust back at the center, rather than GenAI or assessment.
The University of Amsterdam has a wealth of expertise and support structures, providing room to develop an educational system in which trust is central. Educational literature also offers guidance on how to structure both the learning environment and assessment so that trust between lecturers and students grows. A useful tool is Bloom’s taxonomy, which divides learning objectives into three domains: knowledge and thinking skills (cognitive), practical skills (psychomotor), and attitudes, motivation, and values (affective). Many programs currently focus primarily on the cognitive domain, meaning that students are mainly assessed on knowledge and technical skills. We need to collectively – and at the curriculum level – pay more attention to the affective domain.
AI-secure assessment
The film Freedom Writers provides a compelling example of what the affective domain entails. In the film, Erin Gruwell (Hilary Swank) – a young teacher at a racially divided school in Los Angeles – motivates a class of at-risk students, who are often deemed unfit for learning, to develop tolerance, commitment, and the ambition to continue their studies after high school.
How does she do it? She gives her students a journal in which they are completely free to write down their thoughts and feelings. They write and leave their journals in the classroom. Something every day, with pen and paper. It can be that simple – and so AI-safe. AI-secure assessment is ultimately not about GenAI itself. The real danger lies in constantly hammering on – and testing – purely content-based performance and knowledge.
Most recommendations for AI-secure assignments stem from the affective domain. These are interventions that should have been implemented years ago, long before GenAI arrived – simply because they help students learn better and strengthen the trust between lecturers and students. This step requires resources and educators willing to make not only an intellectual but also an emotional effort to get to know their students. But we should take that step and use the new educational vision to put the student back at the center, rather than the thesis.
Nicos Starreveld is a lecturer in the bachelor's program in Mathematics.
