Toward Intersectionally Fair Educational AI for Learning Analytics and Student Guidance

Abstract

Artificial intelligence is increasingly embedded in learning analytics dashboards, adaptive learning environments and student-guidance tools. These systems do more than personalise learning: they rank, filter and prioritise educational opportunities in ways that shape learner visibility, access and future pathways. Fairness evaluations, however, often remain organised around single protected attributes such as gender, ethnicity, socio-economic status or disability. This paper argues that such evaluations are insufficient for educational contexts because they can obscure harms experienced by learners located at the intersection of multiple demographic positions. It therefore advances intersectional fairness auditing as a human-centred requirement for responsible educational AI. The argument is informed by a motivating adjacent-domain case: a study of a transformer-based job-recommendation system found that a system appearing fair attribute by attribute can still conceal substantial unfairness for intersectional subgroups, and that common mitigation methods do not necessarily improve all subgroups evenly. Drawing on recent scholarship on human-centred AI in education, learning analytics auditing, educational recommendation and intersectional fairness, the paper proposes an audit framework for educational AI. The framework extends existing learning analytics audit work by making subgroup visibility, ranking effects, demographic data governance, contestability and post-deployment monitoring explicit. The paper concludes that educational AI should be judged not only by predictive accuracy or personalisation, but also by whether it supports equitable learner futures across intersecting differences and enables learners and educators to question, interpret and contest AI-mediated guidance.