Study: Generative AI marginalizes disabled knowledges in higher education
New research exposes how AI training data reinforces epistemic coloniality against disabled communities.
Get AI news that actually matters
One email a day. Zero fluff. Join 10,000+ professionals.
A new paper by Fatiha Tali-Otmani (EFTS, Grhapes), published on arXiv (2605.26769), examines how generative AI systems in higher education systematically marginalize minoritized knowledges, with a focus on disability. The research, grounded in educational sciences, critical technology studies, and disability studies, argues that AI training datasets—predominantly Anglophone and Western-centric—reinforce epistemic coloniality. These systems are not neutral; they actively restructure how scientific knowledge is produced and validated, often excluding non-hegemonic perspectives.
Using the situation of persons with disabilities as a case study, the paper demonstrates that technological architectures confine these individuals to reductive stereotypes or exclude them entirely from design processes, creating a double marginalization. The author questions whether hybridization between researcher and machine could preserve epistemic plurality, while acknowledging that purely palliative algorithmic corrections are structurally limited. The study calls for deeper integration of marginalized voices in AI development to avoid further entrenching inequality in academia.
- Training datasets are predominantly Anglophone and Western-centric, reinforcing epistemic coloniality in AI systems.
- Disabled individuals face double marginalization: reductive stereotypes and exclusion from AI design processes.
- Paper proposes hybrid human-machine collaboration as a potential remedy, but cautions against purely algorithmic fixes.
Why It Matters
For professionals building AI in education: inclusive training data and design are critical to avoid marginalizing disabled communities.