Aktivitäten 

Freitag, 27.06.2025 |  13:00 - 14:00 Uhr 
3. Lunchtalk mit Andrei Poama

 
Fair Assessment in the Age of GenAI: A Scanlonian Approach

Abstract: Generative AI systems (hereafter, GenAI) are stretching the technical and educational limits of higher-education assessment. In particular, GenAI is automating tasks traditionally undertaken by students, teachers or support staff, relies on big data that goes beyond the information typically required, used or managed for standard assessment processes, and introduces opaque modes of inference through highly complex models that follow a statistical rather than a discursive mode of reasoning which is typically expected, even if not always satisfied in higher-education assessment. These transformations often trigger claims about fraud or deceit, complaints about biased textual output, as well as about the inequalities they introduce or compound between students or universities;  conversely, GenAI is sometimes praised for its tailored treatment of neurodivergent students or its potential for levelling the playing field among different higher-education institutions with respect with assessment quality. Importantly, these claims are either advanced or can be at least prima facie construed as claims about the fair or unfair nature of GenAI uses and applications in higher-education assessment – from generalized cheating to performance inequalities among students to making teachers’ feedback more consistent or less biased. In this paper, we ask which (if any) of these (un)fairness claims about assessment in the age of GenAI can be soundly vindicated. To do so, we put forward principles of fair assessment that can guide and inform higher-education assessment practices and distinguish between cogent and faulty (un)fairness claims made about such practices. Specifically, we start with substantive assessment principles already proposed by higher-education practitioners or practice-embedded scholars and subject these principles to what we call Scanlon’s fair non-rejectability test – namely, we examine whether they can be reasonably rejected or endorsed in light of the benefits and burdens incurred by the parties who are subject to these principles (in particular, students and teachers) or, more precisely, to practices that would be governed by them. Armed with such principles, we explore concrete changes that could render current higher-education assessment practices more rather than less fair in the age of GenAI.