From Digital Competence Gaps to AI-Enabled Inclusive Learning: A Learner-Centered Framework for Persons with Disabilities in Mongolia

Abstract

Artificial intelligence can widen access to learning, but it can also reproduce digital and accessibility gaps when systems are built for an assumed average user. This design study asks which learner needs should shape an inclusive AI-enabled learning environment in Mongolia and how those needs can be translated into an auditable architecture. The empirical base is a mixed-method study of 158 adult (16-35 years old) with vision impairments (n=67), hearing impairments (n=47), and speech impairments (n=44) in Ulaanbaatar. It combined a questionnaire, observation, focus-group inquiry, a five-area DigComp assessment, and an accessibility review of 85 websites and five mobile applications. Aggregate results show that smartphones were the main device for 53%-67% of the three groups, while mean digital-competence scores remained between 2.20 and 2.54 on a four-point scale. Independent participation in online learning ranged from 29% to 49%; average accessibility was 41% for visual access, 47% for hearing access, and 29% across the evaluated mobile applications. Using a design-science process, these findings were converted into requirements for a mobile-first, offline-capable, WCAG 2.2 AA learning environment. The proposed framework combines a consent-based learner model, DigComp 3.0 learning paths, multimodal content transformation, bounded generative AI over approved sources, accessible interaction, human support, and privacy-aware learning analytics. The contribution is a traceable link between disability-specific evidence and a responsible learner-centered AI architecture.