Research
How teacher educators hold on to professional judgment and agency when they co-create with generative AI, and the instrument I am building to watch it happen.
From August 2026 I am a doctoral researcher (doktorand) in GenAI and human-AI interaction at the University of Gothenburg, at the Department of Education, Communication and Learning (IPKL). My position belongs to GRAITE, Sweden’s national graduate school for AI in Teacher Education: a VR-funded collaboration across nine universities.
The questions
Plainly, without the apparatus:
- What do teacher educators actually do when a model hands them something plausible? Not what they say they did afterwards, but what happens in the moment. Where do they push back, where do they let it pass, and on what grounds?
- Is professional judgment something you keep by exercising it? If a capable model absorbs the routine parts of planning and drafting, the interesting question is not whether the output is good. It is what becomes of the judgment that used to be spent producing it.
- What does teacher knowledge look like when part of the work is delegated? I use TPACK, the braid of technological, pedagogical and content knowledge, as my frame: it is the vocabulary the field already argues in, and generative AI puts real strain on its joins.
Beneath all three sits agency. Not as a slogan, but as something you can watch a person exercise, hedge, or quietly hand over.
Why it matters beyond the seminar room
Teacher educators train the people who will teach everyone else’s children. If professional judgment quietly migrates into a model that a school did not choose, cannot inspect, and does not own, the issue stops being one of craft and becomes one of accountability: who answers for what gets taught?
Where this comes from
This is not a swerve. In 2024, while building AI products at Volvo Cars, I ran an internal study on epistemic ownership in AI-assisted coding, combining behavioural traces of what people accepted, rejected and edited with survey data (n=127).
What stayed with me was methodological rather than any particular result. Asking people about their own reliance and observing what they actually do are two different measurements, and the gap between them is where the interesting question lives. That is why I build my own instrument instead of borrowing one (see Artefacts). Judgment is precisely the thing that goes quiet in self-report; nobody remembers the suggestion they let straight through. You have to watch the work.