I came back from Bergen with more than conference notes. I found a generous scientific community, open to dialogue, critical discussion, and mutual support. I am grateful to all the people I met there, for the ideas and conversations we shared, and for the informal, enjoyable moments that helped shape this post.
From AI tools to learning conditions
Artificial Intelligence (AI) in education is not new. It has developed from early computer-based systems to data-driven online learning environments and, more recently, to generative applications. This latest phase is bringing many new tools into universities. Yet the discussions I followed during the 16th International Conference on Learning Analytics and Knowledge (LAK26), held in Bergen, Norway, from 27 April to 1 May 2026, brought me back to a deeper issue: the organisational conditions that make learning possible as these technologies become part of everyday teaching and learning.
This reflection is part of my PhD project at the University of Camerino, Italy, where I collaborate with Michele Loreti (Computer Science) and Andrea Perali (Physics). It brings together ideas I encountered during the conference, across sessions, workshops, keynotes, and informal conversations, and relates them to my work on generative AI, learning analytics and formative assessment. I offer these thoughts as a personal contribution to a wider discussion within the research community.
One message I took from Bergen is that generative AI is becoming part of the educational infrastructure itself. It is entering many areas of university life, from writing, assessment, and feedback to tutoring, learning design, dashboards, study planning, and institutional decision-making. This transformation does not make teachers less necessary. On the contrary, it makes their pedagogical role more important.
Beyond the final product
Education cannot focus only on the final product. A polished essay, a correct solution, or a well-structured report may not be enough to show that learning has taken place. The question is what happened during the learning process: what did the student understand, verify, question, revise, reject, transfer, and internalise?
Several discussions pointed to an important distinction: better performance is not necessarily deeper learning. With AI support, a student may produce better work without necessarily developing a deeper understanding or transferable competence.
This distinction makes learning analytics even more relevant. By using data and evidence to understand and improve teaching and learning, the field can help teachers and students look beyond final outputs and focus on how knowledge and competence develop over time.
But making learning processes visible is not enough. Data do not improve education by themselves. They need to be interpreted in context, through pedagogical understanding and shared responsibility. If learning analytics aims to support learning, universities need organisational conditions that connect evidence, teaching practice, student agency, and continuous improvement.
From tutorship to stewardship
These considerations lead to a central hypothesis: universities need to move from tutorship to stewardship.
By tutorship, I mean fragmented forms of support for students, such as occasional tutoring, separate services, and isolated digital tools that intervene mainly when difficulties have already become visible. These forms of support can help individual students, but they rarely help the university learn from its own practice.
By stewardship, I mean a broader organisational logic that connects students, teachers, staff, data, technologies, and institutional decisions within continuous learning processes. The aim is not only to support students, but also to help the university learn, adapt, and improve.
Human Learning Systems as a lens
The argument draws on the Human Learning Systems approach developed by Toby Lowe and colleagues. Human Learning Systems questions some assumptions of New Public Management, especially the idea that complex public services can be governed mainly through targets, indicators, standardisation, and performance control. Instead, it starts from the idea that human outcomes depend on relationships, contexts, constraints, motivations, histories, and interactions among many actors. For this reason, it proposes learning as a management strategy.
In practical terms, this approach works through learning cycles. These are iterative processes through which organisations seek to understand their systems, test small changes, interpret evidence, and adapt their practices. Stewardship teams support these cycles by connecting people, evidence, and action, while protecting the conditions for organisational learning.
This perspective is particularly relevant for higher education. Student success is not produced by a single teacher, office, platform, tool, or policy. It emerges from the interaction of students, teachers, staff, learning environments, institutional rules, data infrastructures, and personal trajectories. In this sense, universities can be understood as complex public service systems, where improvement depends on the ability to learn across the whole organisation.
Bringing stewardship into higher education
What I find promising is the possibility of bringing together Human Learning Systems, learning analytics, generative AI, and formative assessment within a shared organisational perspective. The question is not simply which tools universities should adopt, but how these technologies change the conditions of learning and teaching.
It also means asking who interprets educational data, who decides what improvement means, and how students and teachers can remain active participants rather than passive users of automated systems.
As part of my doctoral research, I intend to explore a Human Learning Systems-inspired approach through small-scale learning cycles in degree courses. A stewardship team would coordinate this work and connect it to communities of practice involving teachers, students, and staff.
The aim is to study how AI tools and learning analytics can support formative assessment and feedback in real educational settings. Artificial intelligence is not the centre of the model. It is one component of a broader learning environment that needs human stewardship and pedagogical judgement.
Feedback, interpretation, and responsibility
Feedback is central to this shift. It cannot remain only a comment at the end of an assignment. It should become part of the institution’s learning culture.
Students need to question, interpret, and use feedback, including AI-supported outputs. Teachers remain central because they design meaningful tasks and feedback processes, support metacognition, sustain relationships, and understand when students need support and when they need productive challenge. Institutions, in turn, need to use learning data to improve education without turning it into surveillance or mere performance control.
The move from tutorship to stewardship is therefore more than a technical question. It is a way of rethinking educational organisations in the age of AI.
If generative AI becomes part of the infrastructure of teaching and learning, universities need an equally robust human, pedagogical, and organisational infrastructure. The challenge is not only to ask how AI can help students perform better, but also how universities themselves can learn better.
Stewardship means protecting and strengthening the human, pedagogical, and organisational conditions that make learning possible as AI becomes part of university life.
An invitation to the community
I share this reflection as an invitation to continue the conversation with researchers, educators, learning analytics scholars, and practitioners working on AI in education.
If AI is becoming part of everyday teaching and learning, what kind of stewardship should universities build? How can AI support learning, rather than only improve performance? And what should remain in the hands of teachers, students, and educational communities?
Declaration
AI assistance was used to support language editing and improve clarity. I retain full responsibility for the ideas, interpretations, and final text.
References
Lowe, T., French, M., Hawkins, M., Hesselgreaves, H. and Wilson, R. (2020). New development: Responding to complexity in public services: the human learning systems approach. Public Money & Management, 41(7), 573–576. https://doi.org/10.1080/09540962.2020.1832738
Lowe, T., Brogan, A., Eichsteller, G., Hawkins, M., Hesselgreaves, H., Jennions, B. N. and Williams, G. (2021). Human Learning Systems: Public service for the real world. Centre for Public Impact. https://www.humanlearning.systems/uploads/Public-service-for-the-new-world.pdf
Lowe, T., Padmanabhan, C., McCart, D., McNeill, K., Brogan, A. and Smith, M. (2022). Human Learning Systems: A practical guide for the curious. Centre for Public Impact. https://www.humanlearning.systems/uploads/hls-practical-guide.pdf
Society for Learning Analytics Research Learning Analytics Definition Task Force. (2025). Reimagining Learning Analytics. Society for Learning Analytics Research. https://www.solaresearch.org/wp-content/uploads/2025/06/Reimagining-Learning-Analytics-V3-002.pdf
LAK ’26: Proceedings of the LAK26: 16th International Learning Analytics and Knowledge Conference. (2026). Association for Computing Machinery. https://doi.org/10.1145/3785022