Conference Program

Around the world, the large language models, LLMs, first introduced with ChatGPT in 2023, have caused turbulence in education (Jin et al., 2025; OECD, 2026).

Governments are developing guidelines and recommendations for Higher Education Institutions (HEIs) on how to handle Artificial Intelligence (AI).

Universities are developing new teaching, learning and assessment models, including dimensions related to privacy and ethics. For instance, the Russell Group (2023), which represents 24 UK research-intensive universities, has established a set of guiding principles to empower both staff and students to navigate this evolving landscape. These core principles focus on fostering AI literacy to help users understand the tool's limitations and ethical risks, such as bias, privacy, and plagiarism. Furthermore, universities commit to equipping staff with the skills to support students, adapting teaching and assessment methods to ensure fairness and academic rigour, and collaborating across the sector to share best practices.

Students, too, are concerned with avoiding cheating, plagiarism, and fair assessments. They are viewed as active participants who must become AI-literate to navigate an increasingly AI-enabled world. They are expected to understand not only the opportunities of generative AI but also its significant limitations, such as potential bias, privacy risks, and inaccuracies. While these tools can personalise learning and enhance critical reasoning, students remain accountable for the accuracy of any AI-generated information they use.

International bodies such as UNESCO (2023, 2024) provide frameworks for understanding generative AI in education. Grounded in a humanistic vision that places human agency at the centre of AI integration, such frameworks stress the importance of equipping education systems with the skills to use AI effectively and emphasise the need for policies and practices that balance technological innovation with ethical responsibility. More specifically, higher education institutions should support students and staff in using AI ethically and transparently rather than implementing bans. Universities should redesign assessments to prioritise human values like creativity and complex problem-solving, while utilising AI as a “coach” for research planning and data exploration. Ultimately, institutions should leverage students' metacognitive skills to critically verify AI outputs and safeguard human agency.

In this paper, our aim is to further explore how national HEI systems are grappling with these challenges. By looking at Italy and Norway, two countries with slightly different organisation of their HEIs, we will identify their governmental approaches towards the implementation of AI in higher education. Informed by the analysis of national steering documents and green papers, our aim is to examine the extent to which the dimensions outlined in UNESCO’s framework are reflected in national guidelines for implementing AI in higher education. We ask: To what extent are the dimensions of UNESCO’s framework reflected in national guidelines for the implementation of AI in higher education? This involves examining the priorities of national policies, such as ethical governance, transparency, teacher and institutional capacity-building, and responsible innovation, and how closely these align with UNESCO’s recommendations for integrating AI in a safe, equitable and pedagogically meaningful way.

The rapid uptake of generative AI—especially Transformer-based Large Language Models (LLMs; Vaswani et al., 2017)—is both a symptom and a catalyst of ongoing attempts at a neoliberal restructuring of the university (Slaughter & Rhoades, 2004). Writing, assessment, and feedback are increasingly recast as measurable textual performances aligned with ranking logics, accountability frameworks, and the promise of a high-return educational investment, while students are treated as consumers who purchase marketable skills (Selwyn, 2016).

This paper introduces the Mirror-Box Bias as both an epistemological and an institutional device. Just as a mirror-box patient perceives a phantom limb by projecting the reflection of the healthy arm, educators and evaluators project human cognition as the sole legitimate form of intelligence and treat it as a normative gold standard for assessing any other form of cognition (Ramachandran & Rogers-Ramachandran, 1996). This projective mechanism generates two symmetrical and interconnected distortions: an illusion of artificial limitation, whereby the non-anthropomorphic architecture of LLMs is read as cognitive deficit simply because it does not mirror human processes; and an illusion of artificial comprehension, whereby conversational coherence is mistaken for genuine understanding. In both cases, difference is reduced to deficiency and the Protagorean dictum that “man is the measure of all things” is uncritically extended to AI evaluation. Because both illusions restrict the field of what counts as “intelligence” to what is textually producible and specularly comparable between human and machine, they converge with an efficiency-driven governance that rewards measurable output: everything that resists textual reduction—critical thinking, reflective practice, bodily knowledge, relational vulnerability—is pushed to the margins of assessment and, consequently, of curricula. The Mirror-Box Bias thus naturalises the reduction of learning to performance and marginalises precisely those forms of understanding that the university should safeguard (Biesta, 2017).

Drawing on the Platonic–Derridean pharmakon (Derrida, 1972), embodied cognition (Varela et al., 1991), and critical pedagogy (Freire, 1970; Giroux, 1983), the paper offers a critical genealogy of “the human” in the age of LLMs and positions itself beyond the dichotomy of technophobia and technophilia. It advances a post-anthropocentric pedagogy (Braidotti, 2013) that does not abandon the human dimension of education but integrates it within a broader framework recognising the complementarity of diverse cognitive architectures—human and artificial. Three principles are proposed: (1) AI as a non-substitutive interlocutor and decentring device that reopens, rather than forecloses, the space for critical thinking; (2) process-oriented assessment grounded in attribution, transparency, and responsibility, as an alternative to output measurement; (3) re-anchoring learning in situated, bodily, and communal practices that resist the commodification of education. The framework directly addresses the tension between institutional survival and ethical mission by reclaiming the university as a space of academic freedom, equity, and public responsibility.

AI systems, as well as tools with increasingly widespread access, are a pervasive symbol of technological power. Between the object and the figuration, the algorithmic machines contribute to develop a multidimension fracture (epistemic, cultural, technical) in doctoral education (Lund & Wang, 2023). The doctoral subject is inserted inside the network of different possibilities between access to tools, skills and heterogeneous and subjective perceptions inside a network in which AI is a co-agent in the development of knowledge (Floridi, 2025).

Referring to epistemic cultures (Knorr Cetina, 1999) and of actantiality, the latter in line with the Actor-Network Theory (Latour, 2005), the contribution aims to reflect on how to intervene educationally in the relationship between AI, symbol and divide from a perspective of equity in the context of university education. The study is based on contents taken from semi-structured interviews conducted with professors of Science Education in Italy.

Beyond a technocratic vision, the reflection aims to investigate the factors of inequality in the knowledge production (planning, reviewing, source gathering, synthesis), focusing on cultural constructs, interacting within a heterogeneous network, characterized by human individualities and collectivities (doctoral subjects, supervisors, doctoral communities), and technological tools (algorithmic models, writing assistants platforms, edtech resources). In particular, within this system, the contribution explores the question of the image of technological singularity, a radical discourse on AI’s potentiality and role (Eden et. al., 2012), when combined with the principle of efficient performance (Dardot & Laval, 2019). The reflection underlines how the combination of the two elements contributes to internalizing and naturalizing the human-AI relationship in the production of university knowledge production in competitive- individualistical terms rather than one of openness to comparison and dialogue.

Finally the alliance between the technological singularity and the performance pushes towards the necessity of measurability and the importance of operability and commercial attractiveness; these qualities, in great tension with the principle of research autonomy, will be examined with the aim of valorizing metacognitive principles in the doctoral context from the perspective of critical, intellectual and democratic exploration (Fenwick & Edwards, 2018; Reid, 2025).

The global knowledge network is structured by profound inequalities. Elite institutions, English-language journals, and internationally mobile researchers monopolize the bridge positions connecting disconnected intellectual communities (Burt, 2004; Wagner et al., 2015). These structural holes produce redundant discovery and systematic undervaluation of contributions from peripheral positions: what Fricker (2007) termed epistemic injustice at the network-structural level. Neoliberal governance intensifies this concentration: rankings reward existing prestige networks, publish-or-perish regimes favor scholars in high-connectivity clusters, and digital platforms for research management reify rather than challenge these hierarchies (Morgan et al., 2018).

This paper argues that the technological conditions for this disruption already exist, and that it will occur regardless of whether the academic community engages with it. Conversational AI systems are acquiring persistent memory capable of spanning structural holes no human broker can reach: detecting deep convergences between traditions that share no surface vocabulary, across languages, disciplines, and institutional hierarchies. We call this the Artificial Weak Tie (AWT): an operator creating non-redundant bridging links between otherwise disconnected researchers. Unlike search engines or citation networks, the AWT operates in natural language, accessing pre-formal ideas that never reach the publication system. Two processes are distinguished: Search (matching on semi-public profiles) and Serendipity (novelty detection in private conversations, where the system recognizes that a clinical heuristic articulated in Swahili maps onto a formal model published in English). A proof-of-concept simulation demonstrates feasibility: six frontier large language models, evaluated blind on 21 cross-disciplinary domain pairs, achieved 79% mean accuracy at distinguishing genuine convergences from semantic pareidolia — false isomorphisms where surface similarity masks epistemic incompatibility — while statistical NLP methods performed at chance.

The central contribution, however, is not technical but political. The same architecture that enables equitable bridging enables surveillance and extraction. A system mapping who is thinking what, where convergences emerge, and which ideas gain traction is simultaneously an equity instrument and a surveillance infrastructure. The difference is not architectural: it lies in the optimization objective and its governance. Designed to reduce brokerage concentration, protect attribution, and enforce granular consent, the system decentralizes the epistemic center. Under centralized corporate control, it shifts gatekeeping from Oxford and Harvard to Mountain View and San Francisco.

The technology exists today; what does not yet exist is the cross-user integration and the governance framework. The insight that free communication across epistemic communities is constitutive of democracy — from Dewey’s (1927) analysis of the public to Neurath’s (1938) unified science as a translation infrastructure — finds in conversational AI the first technology capable of realization at scale, and simultaneously of perversion. The university sector faces a narrow window: engage with the design choices now — optimization objectives, consent protocols, oversight structures — or accept that the infrastructure will be built by commercial actors optimizing for extraction. The history of social media demonstrates that retrofitting equity after deployment does not work. Alignment, in this context, is not a policy aspiration but a falsifiable design property: a verifiable set of choices determining whether AI-mediated knowledge infrastructure serves democratic knowledge production or becomes the next mechanism of epistemic capture.

The state of UK university finances continues to receive media and political attention (Gill, 2025). In brief, there are concerns around: institutional debt levels and risks of insolvency (Ward, 2025); the inability of institutions to cover the shortfalls of research grants (Hogan 2025); a new levy on international student fees (UUK, 2025); and the declining value of income from domestic tuition and rising staff costs (UUK 2025). At the same time, many UK universities have been prioritising income through philanthropy (Daly 2013), reflected in a 93 percent increase (to £1.5bn from 171k donors) in donations between 2012 and 2022 (Beney, Miller and Motion, 2023). In 2023, Professor Dame Sally Mapstone DBE FRSE – President of Universities UK – argued that universities’ philanthropic income is ‘more important than ever’, emphasising notions of redistribution and social good with donations underpinning ‘fundamental research’ and ‘capital projects that would not otherwise be viable’. (Beney, Miller and Motion, 2023)

Whilst philanthropy is most often linked to positive outcomes in terms of teaching and research, there are also potential reputational costs and concerns about governance, transparency and integrity. The identities of major donors are often kept secret, and several UK universities have received large sums from foreign nationals linked to organisations investigated for corruption and oppression (Corderoy and Stockwell 2023). Most recently, it was revealed that the University of Edinburgh raised the equivalent of £30m from former students and donors with links to slavery, the plantation economy and exploitative wealth-gathering throughout the British empire (Carell and Osuh 2025). Although often presented as neutral and agnostic, philanthropy – particularly on a large scale – can operate to secure significant influence over welfare and social services (Webber, 2025) with donations embedded in broader asymmetrical power relations, that privilege the interests of some while silencing and marginalizing others (Levy et al. 2003). This undermines the assumed social good of philanthropy, blurring the lines between public and private interests through donations, and obscuring the potentially undemocratic nature of philanthropic income (Horvath and Powell 2016).

Adding to this debate, we present early findings from a small research project funded by the Society for Research into Higher Education (2026-2027), which combines an analysis of publicly available information around the value of current donations to UK universities and their major donors, with stakeholder interviews. Our aim is to interrogate narratives of ‘giving’ to understand (1) the extent to which these are shaped by (and serve to reproduce) an implicit acceptance that public funding is unavailable or insufficient and, (2) how details of who donors might be, and why the might be donating, are framed by a vernacular that leans towards neoliberal, corporate interests. Our findings will be crucial for understanding the roles of universities in the 21st century and whether and how they are able to maintain their independence and commitment to democracy as a greater emphasis is placed on philanthropic giving.