For decades, the image of the exhausted teacher buried under a mountain of ungraded papers has been a staple of the education system. But in classrooms across the Netherlands, a digital shift is quietly altering that landscape. No longer limited to the binary simplicity of multiple-choice tests, an increasing number of docenten kijken na met AI, utilizing sophisticated tools to evaluate open-ended responses and even decipher handwritten scripts.
This transition is not merely about speed; it is about a fundamental change in the labor of teaching. By leveraging artificial intelligence to handle the first pass of grading, educators are attempting to reclaim hours of their week. However, as these tools migrate from experimental pilots to standard practice, a sharp divide has emerged between those who see a liberating “teaching assistant” and those who warn of a profound loss in the pedagogical relationship between teacher and student.
The adoption is already measurable. According to company data, the tool Kwizl is currently implemented in approximately 45 secondary schools, whereas ToetsTester is utilized in around 15. These platforms represent a new generation of nakijksoftware that goes beyond simple automation, attempting to mimic the cognitive process of a human grader by analyzing the content of a student’s answer rather than just searching for keywords.
The Rise of the Digital Teaching Assistant
For many educators, the appeal of AI grading is rooted in the sheer volume of administrative pressure. Heleen Ogier, a Dutch language teacher, has integrated ToetsTester into her workflow since May of last year. The process is seamless: handwritten texts are uploaded, converted into digital text, and then evaluated by the AI.
“It is a sort of teaching assistant that does an initial sorting, which leaves you with time for other things,” Ogier says. The software suggests a point value based on the content of the answer, but Ogier emphasizes that the final decision remains human. “I then have time to look at the formulation, and I still determine whether the number of points is justified.”
Beyond the immediate time savings, teachers highlight the power of automated analysis. Joost Verheugen, another Dutch teacher, notes that the data provided by Kwizl offers a level of insight into student skills that would be nearly impossible to aggregate manually. “I can make that analysis myself, but with thirty students, that is a lot of work. We don’t get that time as teachers,” Verheugen says.
This data-driven approach allows teachers to identify systemic failures in their own curriculum. Martin Bakker, a geography teacher, found that ToetsTester acted as a mirror for his own instructional clarity. “You are confronted with your own unclear questions or imperfections in your answer model,” Bakker observes.
The Didactic Risk: Losing the Human Connection
Despite these efficiencies, critics argue that the act of grading is not a chore to be outsourced, but a vital part of the educational process. Felienne Hermans, a professor of computer science didactics at the Vrije Universiteit Amsterdam, warns that outsourcing this task risks stripping the classroom of its “didactic value.”
According to Hermans, the process of reading a student’s open-ended answer is where a teacher truly learns who their students are and how they think. “Grading gives teachers valuable information about their students. If you leave that to AI, you lose an important didactic value,” she argues. She suggests that using AI for multiple-choice questions is acceptable, but using it for open questions creates a dangerous gap in the teacher-student relationship.
Hermans similarly challenges the exceptionally premise of time-saving. She points to the concept of “human in the loop”—a requirement in emerging regulations ensuring that a human maintains control over AI outputs. “If you still have to check everything, it is not more efficient,” she says, suggesting that the overhead of verifying AI accuracy may cancel out the initial speed gains.
A Legal Gray Area
The rapid deployment of these tools has effectively outpaced the law. The European AI Act classifies AI applications in education as “high-risk” because they can significantly influence a person’s life trajectory via grades and certifications. While parts of this legislation were slated to take effect this summer, the implementation of these specific rules was recently delayed until next year.
This delay has left schools and software providers in a period of tentative exploration. The Dutch Ministry of Education has stated that using AI for grading is permitted under current data protection laws and the spirit of the upcoming AI Act, provided that the school remains ultimately responsible for the assessment.
This responsibility becomes particularly sensitive during high-stakes testing. When asked if AI could be used to grade final exams in May, Cito and the College voor Toetsen en Examens stated they do not dictate the specific methods. The general line is that the responsibility for the grading process rests solely with the school and the individual corrector.
| Tool | Estimated School Adoption | Key Capability | Primary Use Case |
|---|---|---|---|
| ToetsTester | ~15 Schools | Handwriting to Text | Open-ended answers |
| Kwizl | ~45 Schools | Skill Analysis | Competency tracking |
| ToetsPers | Unspecified | Handwritten Grading | General assessment |
The Path Toward Standardization
As the academic year progresses, the tension between efficiency and pedagogy remains unresolved. Some teachers, like Heleen Ogier, have opted for a middle ground, using AI only for non-graded assignments to ensure the tool remains supportive rather than decisive. Others, like Martin Bakker, remain in the experimental phase, arguing that it is too early for school-wide adoption.
The consensus among those experimenting with the technology is a plea for centralized guidance. Rather than leaving each school to navigate the ethical and legal minefield alone, educators are calling for a governing body to establish clear protocols on how AI should be used in assessments. “An agency should say: this is what you can do, in this way,” Bakker says. “If you leave it to the schools, it will not work.”
The next critical checkpoint will be the implementation of the European AI Act next year, which will likely introduce stricter transparency and accuracy requirements for “high-risk” educational software. Until then, the classroom remains a living laboratory for the future of grading.
Do you believe AI in grading enhances or diminishes the quality of education? Share your thoughts in the comments below or join the conversation on our social channels.
Disclaimer: This article discusses the use of AI in a legal and educational context; it does not constitute legal advice regarding the European AI Act or national education laws.
