Currently reading Kahn et al. (2025) on “Teacher agency and generative artificial intelligence: teaching in higher education as a responsive, cultural activity”

November 26th, 2025
Posted in literature
Tagged artificial intelligenceGenAIrelationships

The wide availability of GenAI has introduced challenges to relationships between teachers and students and within the respective groups (see for example Rachel’s and my work on that!). GenAI being able — and to unknown extent being used — to perform actions that before had to be done by humans creates different perceptions of what should be done by humans vs the black box, and also expectations of how others should be working with (or without) it.

In this context, Kahn et al. (2025) explore teacher agency. Agency for them means “a process of deliberating upon and prioritising concerns, before undertaking projects (that is, sets of actions) that respond to those concerns and that, in time, become embedded as practices“. They find four main issues:

“Uncertainties around student use of LLMs“: teachers cannot necessarily recognise whether work has been created with or without GenAI. Yet, teachers report being hesitant to have conversations with students about their use of LLMs, both because they fear that students would respond with what they think the teacher wants to hear, but also because they feel that having such conversations with students they are directly working with might be difficult relationship-wise because of the power differential. But in the end, that means that teachers have no clear idea of how students are using LLMs, and therefore they cannot advise them on how to avoid specific pitfalls or make better use. And it turns out that teachers think that students use LLMs much more than students think of each other.
“Concerns and actions pertaining to a weak student use of LLMs“: teachers are concerned that students use GenAI in ways that is not supporting, or even hindering, their learning, and lead to lower grades. The typical criteria that teachers use are about the quality of arguments, how students work with literature and evidence, presentation and language, and how they analyse, all of which they feel turn out worse when using GenAI than without — while students generally believe they do better when using GenAI.
“Tensions in the way that expertise and knowledge shape teacher agency“: Several teachers respond in ways that can be classified as “‘fractured reflexivity’ […] characterised by anxiety and worry, that is by trains of thought that are not directly connected with action“, putting the head in the sand. Generally, there are some teachers that state they won’t engage with GenAI, which will make things increasingly difficult for them and their students.
“Challenges that were experienced by staff to the collegial basis for shared action around student use of LLMs“: Many teachers report that they feel uncomfortable discussing GenAI with colleagues, which of course does not help at all, and leads to concerns being downplayed.

Kahn et al. (2025) write later that “It is important […] for staff to maintain a rich understanding of the practices and intentions of students” (which requires both conversations with students, but also an adequate understanding of, and experience with, GenAI yourself to be able to have those conversations), since “[a] shift in the ‘resources and rules’ […] needs to be addressed in a relational manner, rather than being a problem which the different individuals and groups adapt to in an isolated way“. They also describe that independent learning is highly valued in academia and that we tend to leave students alone quite a bit (interestingly, I just wrote about how we don’t really know what students do “when we are not there”, and how many don’t really know how to learn independently!), but that this can become really problematic when we aren’t recalibrating relationships, too.

And GenAI also widens gaps: “LLMs will tend to match any specialised vocabulary from the user input in their responses because they infer user expectations from such input. This means that expert users can ‘unlock’ functionality which those with less cultural capital will be unable to access.”

One sentence somehow stuck with me, and I don’t know if that was from my first reading of the article or from having heard it in another context, but it is that the questions related to GenAI are “casting teachers in the role of local interpreters of abstract policies“. And this is difficult enough as it is, but even more so if there are little to no conversations between peers to support each other in interpreting and making meaning together!

One point Kahn et al. (2025) mention in passing in the introduction is that students who rely on LLMs for their social support isolate themselves from others and report lower levels of belonging, which of course makes intuitive sense, but is still concerning and something we should keep an eye on and also make students aware of — GenAI is not just a tool that people use instrumentally, there are lots of ways in which GenAI use depends on — and influences! — other factors in life.

So moral of the story? We need more and better conversations between actual humans, as always…

Kahn, P., Carrigan, M., Smith, P., Murtagh, L., Liu, R., & Song, F. (2025). Teacher agency and generative artificial intelligence: teaching in higher education as a responsive, cultural activity. Learning, Media and Technology, 1-12.

First dip back in Sweden after the work trip to Norway that I also haven’t blogged about yet… How did I get so far behind??