Conference Program

This paper examines generative artificial intelligence, and particularly Large Language Models (LLMs), as epistemic infrastructures reshaping contemporary academic knowledge production. Situating the analysis within the broader transformation of universities - marked by performance metrics and pervasive technological integration - the study investigates how AI-mediated research practices intersect with questions of epistemic trust and the organization of academic work. Drawing on Science and Technology Studies, epistemology, sociology of science and computer science, in this paper AI is framed as a sociotechnical system, navigating between overly enthusiastic technodeterminism and ideological rejection.

First of all, we need to consider how LLMs, unlike human cognition, operate through large-scale statistical pattern recognition, generating their outputs through probabilistic correlations learned from vast data corpora; there is no access to semantic understanding, intentionality, causal reasoning or contextual judgment. For this reason, what appears as coherence is often the surface effect of statistical regularity rather than grounded comprehension, not considering LLMs structural vulnerabilities, such as susceptibility to data poisoning, sensitivity to prompt framing and generation of plausible but incorrect statements.

As these systems become embedded in core research practices - literature synthesis, hypothesis articulation, methodological design - two different dynamics emerge. First of all, if traditional academic “trust” is rooted in identifiable authorship, accountability and shared epistemic norms - necessitating agents capable of reflexivity and commitment - LLMs, by contrast, cannot assume responsibility for their claims, nor can they participate in the normative space of reasons that structures scientific debate. Secondly, if LLMs intervene directly on the rhythms of research - compressing tasks into near-instantaneous outputs - they carry significant consequences for the quality and depth of scientific work, because they could foster a culture of “epistemic impatience”, in which speed becomes a proxy for productivity and depth is sacrificed in favour of output volume. This dynamic is further reinforced by the institutional pressures already shaping contemporary universities - publish-or-perish logics and competitive funding regimes - which find in AI-mediated acceleration a seemingly natural ally. These processes reveal not only an epistemological challenge, but also a structural tension: the tools celebrated as researchers' “assistants” may erode the epistemic foundations upon which scientific knowledge is built and sustained.

These architectural and institutional tensions demand a calibrated response. Rather than advocating wholesale adoption or rejection, this paper argues for flexible regulatory frameworks preserving human oversight and responsibility, alongside concrete practices - such as disclosure of AI use and verification protocols - to rebuild trust in AI-mediated research environments. Equally important is how the growing integration of LLMs may itself offer an opportunity to critically interrogate the neoliberal turn of the contemporary university: by making visible the tensions between this system and the epistemic request for genuine scientific production, these tools show what is at stake when productivity displaces knowledge as the constitutive horizon of academic life.