In a context marked by multiple and entangled crises — social, ecological, democratic and epistemic — education research is increasingly challenged to interrogate not only what counts as knowledge, but also how knowledge is produced, mediated and mobilised. As Barnett (1997, 2007) argues, higher education is no longer a stable site of knowledge transmission, but a space of uncertainty in which learners must develop judgement rather than certainty. The rapid expansion of artificial intelligence (AI) in higher education intensifies these challenges, reshaping the conditions under which knowing and acting take place and amplifying tensions between automation, performativity and educational responsibility (Biesta, 2010).

This paper addresses these concerns by presenting a pedagogical experiment that reframes AI not as a tool for knowledge optimisation, but as a contrasting device for critical inquiry within a situated learning framework grounded in University Social Responsibility. Rather than enhancing efficiency or replacing interpretative work, AI is mobilised to expose the limits, assumptions and reductions inherent in algorithmic forms of representation.

Developed within the master’s course Sociology of Physical Activity and Health at the Faculty of Sport, University of Porto, the experience engaged students in a service-learning project connected to Turma do Mar, a community-based surfing initiative involving participants in situations of social vulnerability. The pedagogical design deliberately combined theoretical instruction with embodied participation and social intervention in a real-world context, foregrounding the entanglement between academic knowledge, lived experience and ethical responsibility.

The experiment responds directly to contemporary concerns within education research regarding the growing authority of data-driven and visual forms of knowing (Beer, 2019). Rather than positioning AI as a neutral or authoritative source of interpretation, generative AI tools were introduced at a later stage of the process to produce narrative descriptions based solely on photographic records of the fieldwork. These AI-generated narratives were explicitly framed as partial, decontextualised and provisional accounts, intended to be critically examined rather than accepted.

Students were invited to contrast AI-generated descriptions with their own reflective logbooks, interviews and collective discussions. This comparative work enabled them to identify tensions between visual appearance and lived experience, between algorithmic description and situated understanding. Through this process, AI functioned as a means of making epistemological assumptions visible, exposing the limits of computational mediation in capturing relational, embodied and ethical dimensions of social practice.

From the perspective of “knowing and acting”, the experience highlights how critical judgment emerges through friction rather than coherence. Learning was not oriented towards producing definitive interpretations, but towards cultivating students’ capacity to navigate uncertainty, question representational authority and recognise the interpretative labour involved in educational research and professional practice.

The findings suggest that AI can contribute to critical educational aims when its use is carefully bounded, pedagogically mediated and embedded in socially responsible practices. In a time of polycrisis, this approach offers a way of engaging with technological change while reaffirming the public role of higher education as a space for ethical discernment, reflexivity and situated knowledge production.

The study adopts a qualitative, interpretive and pedagogically embedded methodological approach, aligned with traditions of situated learning, service-learning and critical educational research. The pedagogical experiment was conducted within a master’s-level course and involved students as co-participants in both learning and inquiry processes, recognising professional learning as a situated and relational endeavour rather than a linear acquisition of competences. The intervention unfolded across three interconnected stages. First, students engaged in theoretical seminars addressing sociological perspectives on physical activity, health, inequality and social responsibility. These sessions established analytical frameworks for understanding social intervention as a site of knowledge production, interpretation and ethical decision-making. Second, students participated in fieldwork through the community-based project Turma do Mar, combining participant observation with direct involvement in surfing sessions designed for socially vulnerable groups. This phase prioritised embodied engagement, relational learning and attentiveness to context. Data collection followed a multimodal qualitative design (Denzin, 2009) and included photographic records from the field, individual reflective logbooks maintained throughout the process, and semi-structured interviews and focus group discussions conducted after the intervention. In the final stage, generative AI tools were used to produce structured narrative descriptions based exclusively on the photographic material. Methodologically, AI outputs were treated neither as data nor as analysis, but as analytical provocations. Students systematically compared AI-generated narratives with their own experiential accounts and qualitative materials, identifying absences, misalignments and interpretative reductions. This triangulation strategy made visible the conditions under which knowledge claims are constructed and authorised. Ethical considerations were central throughout the project, particularly given the involvement of vulnerable populations. Informed consent, careful use of visual materials and ongoing pedagogical supervision ensured that AI use remained transparent, limited and critically oriented.

This pedagogical experience contributes to current debates within education research by demonstrating how AI can be mobilised to support, rather than undermine, critical knowing and responsible action. By positioning AI as a contrasting device, the project disrupted assumptions of objectivity, neutrality and efficiency that often accompany algorithmic technologies in educational contexts (Beer, 2019). The confrontation between AI-generated descriptions and students’ situated experiences revealed the epistemic limits of decontextualised forms of knowing and highlighted the centrality of human judgement in interpreting social practice. Rather than resolving uncertainty, the learning process rendered uncertainty pedagogically productive, fostering reflexivity, ethical awareness and professional responsibility — capacities that Barnett (2007) identifies as central to education in conditions of uncertainty. In line with the ECER 2026 theme, the study underscores the importance of examining the changing conditions of education research, particularly the growing influence of data-driven technologies and performative regimes of knowledge production (Biesta, 2010). It argues that responding to these changes requires pedagogical and methodological designs that foreground context, responsibility and the public role of the university. Ultimately, the paper suggests that the educational potential of AI lies not in its capacity to produce knowledge, but in its ability to make visible the conditions, assumptions and limits of contemporary knowledge production. Such approaches are essential if education research is to meaningfully contribute to knowing and acting within contexts of polycrisis.

Barnett, R. (1997). Higher Education: A Critical Business. Society for Research into Higher Education; Open University Press. Barnett, R. (2007). A Will to Learn: Being a Student in an Age of Uncertainty. Open University Press. Beer, D. (2019). The Data Gaze: Capitalism, Power and Perception. SAGE Publications. Biesta, G. J. J. (2010). Good Education in an Age of Measurement: Ethics, Politics, Democracy. Routledge. Denzin, N. K. (2009). The Research Act: A Theoretical Introduction to Sociological Methods. Transaction Publishers.