Conference Program

In European (Herodotou et al., 2020; Vaarma et al., 2024) and Italian (Zingaro et al., 2021; Cannistrà et al., 2022; Delogu et al., 2024) higher education contexts, predictive AI models have been employed in institutional initiatives to identify students at risk of dropout and to inform decision-making. When predictive systems are mobilized within governance processes, they participate in constructing academic risk as an object of knowledge and intervention. Once embedded in institutional routines, this construction extends to redefining responsibility and success, shaping how achievement is measured and how support or sanction is allocated. Predictive systems may orient governance toward early support and redistribution, or become aligned with performance-based accountability and competitive positioning. It is this tension that frames the problem of predictive governance examined in this paper. While predictive systems are often introduced under an ethically grounded rationale of early identification and support, their integration into institutional processes extends beyond benevolent intent. To account for this reconfiguration, the paper conceptualizes predictive governance as a dispositif in which anticipatory knowledge production, accountability regimes, and strategic decision-making intersect. By translating educational trajectories into probabilistic risk categories, predictive systems contribute to defining what is considered measurable, actionable, and institutionally relevant. Drawing on critical pedagogy (Freire, 1970; Giroux, 2014; Apple, 2006) and analyses of neoliberal governance in higher education (Ball, 2003; Brown, 2015; Slaughter & Rhoades, 2004), the analysis situates predictive AI within governance environments characterized by performativity and metric-based evaluation. When predictive metrics are incorporated into funding and assessment regimes, they may influence institutional agendas and research priorities, privileging quantifiable outputs and potentially narrowing the pluralism of knowledge. Predictive governance thus intersects with academic freedom and the democratic mission of the university. To make these dynamics analytically visible, the paper develops a theoretically derived analytical framework from a governance-oriented conception of predictive systems as dispositifs (Foucault, 2008; Ball, 2003; Brown, 2015), within which the epistemological conditions of prediction (Shmueli, 2010; Douglas, 2009) are constitutive of how risk becomes knowable and governable. From this perspective, predictive governance can be decomposed into three interrelated moments: the epistemic production of risk, its institutional mediation, and its political-operational translation into decision-making. The taxonomy specifies the informational regimes, modelling architectures, predictive quality and inferential robustness through which risk is constructed; the interpretative and evaluative mechanisms through which it becomes actionable; and the strategic purposes, governance structures, temporal orientations, uses and claims of generalizability through which it is operationalized. Rather than empirically classifying models, the framework renders explicit the normative assumptions embedded in their design and deployment. Drawing on an empirical experience conducted in an Italian university (Zanellati, Zingaro & Gabbrielli, 2024), the paper uses this case as a reflective context to illustrate how configurations of predictive governance may orient systems either toward preventive and redistributive support or toward classificatory practices that individualize structural vulnerabilities. The central question is therefore not only how to use predictive systems responsibly, but which rationality they reinforce: a performance-optimized organization or a public university committed to critical inquiry, equity, and democratic responsibility.