Generative artificial intelligence and the marginalization of minoritized knowledges in higher education

Synthesis: This paper argues that Generative AI systems in Higher Education are not epistemically neutral — they actively marginalize non-hegemonic ways of knowing. Drawing on educational sciences, critical technology studies, and disability studies, Tali-Otmani demonstrates how predominantly Anglophone and Western-centric training data reinforces epistemic coloniality. The situation of persons with disabilities provides a particularly clear illustration, where technological architectures confine them to reductive stereotypes or exclude them from the design process entirely.

Double Marginalization

The paper identifies a double marginalization for disabled learners: first, their epistemologies are underrepresented in AI training data; second, they are excluded from the design process that shapes AI tools used in education. This compounds existing Equity challenges documented in The Illusion of Competence: Self-Perceived Digital Literacy and AI Readiness Among European Secondary Students research, where surface-level AI proficiency masks deeper exclusion. The concept extends Equity frameworks by focusing on knowledge production rather than just access.

Epistemic Coloniality

Training data predominantly sourced from Anglophone, Western academic traditions means that Generative AI outputs reflect and reinforce those epistemic frameworks. When deployed in higher education without critical awareness, these systems can present non-Western or disability-centered knowledge as less legitimate. This connects to concerns in Generative AI as a Design Variable: An Evidence-Centered Framework for Principled Governance in STEM Assessment about whose standards govern AI use in educational settings.

Proposed Hybridization

Tali-Otmani explores whether a researcher-machine hybridization could preserve epistemic plurality. Rather than rejecting AI tools outright, she examines whether collaborative human-AI processes might surface marginalized perspectives. However, she warns against treating Bias Mitigation through algorithmic correction as a purely palliative strategy — structural limitations persist when the underlying training data and design processes remain unchanged.

What this means for practice

• Instructors. Treat generative AI output as one epistemic voice among several in your course: name which traditions the tools are trained on and require students to bring marginalized knowledge into the discussion rather than defer to the model's framing.
• Instructors. Audit your own AI-supported work — feedback, peer review, course materials — for which literatures and methods the tools privilege, and make that bias an explicit object of instruction instead of leaving it implicit in the workflow.
• Instructors. Go beyond Accessibility when choosing tools for disabled learners: accessibility of the interface is not the same as legitimacy of their knowledge, so check whose epistemologies the system can represent and validate.
• Learners. When the model cannot account for your experience, read that as a gap in the training data rather than a gap in your knowledge, and say so in the record — as the paper puts it, human judgment must remain primary in the knowledge production chain.
• Instructors. Do not treat bias correction as sufficient: the paper argues that algorithmic remedies remain palliative while the underlying corpora and design processes exclude disabled and non-Western epistemologies.

Limitations

• The paper is a theoretical analysis that draws on educational sciences, critical technology studies, and disability studies; it presents no participants, dataset, or empirical measurement, so it cannot establish how often epistemic marginalization actually occurs in any given tool or institution.
• Its argument runs through three analytical levels — training corpora, algorithmic mechanisms, and evaluative practices — but each is supported by the cited literature rather than by an audit performed here, and disability serves as an illustrative case rather than a sampled population.
• The hybridization between researcher and machine is proposed as a possibility with acknowledged structural limits, not as an intervention evaluated for outcomes; the authors explicitly call for future longitudinal and comparative studies to test the cumulative effects on epistemic diversity.

Citation

Tali-Otmani, F. (2026). Generative artificial intelligence and the marginalization of minoritized knowledges in higher education: The case of disability.