1. médialab Sciences Po
  2. News
  3. L'IA générative dans le métier d'étudiant. Une analyse ethnométhodologique des régimes d'accountability

L'IA générative dans le métier d'étudiant. Une analyse ethnométhodologique des régimes d'accountability

Bilel Benbouzid (LISIS, Gustave Eiffel University) will present his work on how student's work is changing with the rise of generative AI during the médialab seminar.

Event, Research Seminar

⇥

Salle du Conseil, 13 rue de l'Université, 75007 Paris

Abstract

Most research on generative AI in educational context uses the brain as its unit of analysis. The aim is to measure the cognitive impact of these tools on students and to assess what they gain or lose in terms of memory, attention, metacognition, and cognition. Ethnographic approaches, which focus on situated uses and the practical meanings that users gives them, remain rare, with the notable exception of the work by Ricci and Alcaras (2025). This paper follows this minority approach by adopting an ethnomethodological framework that, by examining what students do with ChatGPT, analyzes how they account for their actions, both to themselves and to a student researcher.

The study is based on a corpus of nearly 200 qualitative interviews conducted with students from a variety of academic disciplines (social sciences and humanities, physics and chemistry, mathematics and computer science, architecture, urban engineering, literature, and foreign languages). Their analysis employs a hybrid methodological approach that itself constitutes a reflexive issue: each interview was first processed using a system of structured data sheets generated by an LLM, before being fed into a multi-agent architecture tasked with constantly comparing interviews to identify interpretive clusters. The paper also aims to address the question of what delegating part of the analytical work to agents means for qualitative sociology.

The analysis does not focus on the opinions expressed, the representations invoked, or a mere inventory of practices, but rather on the intensity and structure of the interactional work undertaken by the student to navigate this interview situation. How do the difficulties arising from the obligation to account for the use of IAGs manifest themselves, and how are they addressed? The interview emerges as a scenario that students navigate in different ways.

Five accountability regimes have emerged from this analysis, with an accountability regime referring to the overall interactional tone of an interview, that is the way in which a respondent manages, throughout the exchange, the fact of having to account for their use of IAGs before a peer: The regime of moral scrutiny treats usage as a transgression to be managed. That of instrumental normalization proceeds through pragmatic trivialization. Sovereignty asserts the maintenance of cognitive control over the machine. Defensive advocacy reveals the presence of an implicit accuser, often the university institution, incapable of a clear policy, before whom the respondent must justify themselves from the outset. Finally, the claimed optimization turns the stigma into an asset, efficiency becomes competence, and usage becomes proof of professional adaptability.

What these frameworks help us understand is how the role of the student is being reshaped under the pressure of a technology that blurs the lines between independent intellectual work and outsourced assistance. This presentation will seek to illustrate these frameworks through fragments from interviews and to discuss the implications of this approach for the sociology of digital practices.

What the analysis of these regimes also shows is that conversational agents (primarily ChatGPT and Gemini) are becoming, for many students, a discreet space into which certain tasks, previously handled to varying degrees by the teacher-student relationship, group work, or standard university exams, are shifting. We will show that this intimacy dividend (a gain in intimacy) based on a student-machine hybridization can become an intimacy trap.

Biography

Bilel Benbouzid is an associate professor at Gustave Eiffel University and a researcher at the Interdisciplinary Laboratory for Science, Innovation, and Society (LISIS). His work focuses on the regulation of digital technologies, particularly policies related to artificial intelligence. He is particularly interested in issues of fairness and accountability in machine learning, evaluation techniques for large language models, and European regulation of artificial intelligence. His current research includes an in-depth investigation into the uses and practices of generative artificial intelligence in higher education. His work also draws on computational sociology methods applied to the study of the economic and social dynamics of digital platforms, particularly YouTube. Involved in several collaborative research projects, he works at the intersection of the social sciences, computer science, and public policy.

Practical informations

This seminar will be held in person and in French, on Tuesday, June 2, 2026, from 2:00 PM to 3:30 PM, in salle du Conseil, 13 rue de l'Université, 75007 Paris.

Registration is mandatory via this link.