Data-Driven Disparities: How AI Applications in Education May Perpetuate or Mitigate Inequality

Against the phenomenological background of multiple experiences of educational access and inclusion, the study focuses on the role of artificial intelligence and asks how algorithmic decision-making can marginalize learners or reduce these exclusionary processes. In view of this, the aim is to determine how AI can provide marginalized and underrepresented learner groups with adaptive access options, taking into account the requirements of DIN EN ISO 9241, the Accessibility Reinforcement Act, the European Accessibility Acts (EAA), the Web Content Accessibility Guidelines (WCAG 2.2), and the German Barrierefreie-Informationstechnik-Verordnung (engl. Barrier-free Information Technology Ordinance, BITV 2.0); which attributions educators, learners, and educational policy makers attribute to the promise of artificial intelligence in inclusive learning environments; and which socio-economic and infrastructural restrictions prevent the standard-compliant integration of these technologies. In this theoretical consequence, a research design with quantitative and qualitative methods is applied, which combines a meta-analysis and a thematic evaluation for the triangulation of results. Based on empirical research, the findings indicate that artificial intelligence can produce learning gains and increased participation in learners with disabilities who use adaptive technologies and assistive devices. However, it can also be exploratively deduced and explicated that these advantages are currently unevenly distributed, are primarily reserved for resource-rich learning environments, and depend on an impartial algorithmic design as well as the available infrastructure and the professional qualifications of teachers. In addition, stakeholders have acknowledged the ability of artificial intelligence to improve accessibility through language and translation tools but have also raised concerns about algorithmic bias, institutional compartmentalization, and a lack of standardized accessibility compliance. It is also clear that financial, technical, and political obstacles are widening the digital divide, especially in rural and resource-poor regions. In this context, it can also be stated that artificial intelligence, if configured adaptively and according to the taxonomy of learning objectives, can compensate for or exacerbate educational gaps, depending on how it is implemented. Last but not least, it is essential to consider that the fair integration of artificial intelligence requires inclusive design principles, ethical governance mechanisms, resilient infrastructure, and the cooperation of all relevant stakeholders, thereby contributing to goal four (SDG 4) of the United Nations 2030 Agenda. Based on this premise, it can be concluded that artificial intelligence should act in a context-sensitive manner by also taking into account processes of enculturation, acculturation, and inculturation. In addition, it should comply with the normative requirements of DIN EN ISO 9241-11, DIN EN ISO 9241-110, and DIN EN ISO 9241-210, as well as the Barrier-Free Information and Telecommunications Act (BITV 2.0), the European Accessibility Act (EAA), and the Web Content Accessibility Guidelines (WCAG 2.2), especially since the Barrier-Free Accessibility Act (BFSG) has been binding in Germany since June 28, 2025. However, current implementations still only partially meet this standard, especially in the tertiary education environment with digitized learning offerings. In this context, it must also implement ethical control mechanisms to ensure that all learners are catered to in a manner appropriate to their target group.