Generative AI for Teachers with Vision Impairments in the Global South: A Bridge Too Far?
Manohar Swaminathan, Tarini Naik · 2025 · ASSETS 2025: 27th International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/3663547.3746392
Summary
This mixed-methods study examines the readiness and capacity of teachers with vision impairments (TVIs) in India to leverage generative AI technologies in their teaching practice. The research combines in-depth interviews with 15 TVIs from 10 schools in Karnataka and a nationwide survey of 105 TVIs across 15 Indian states. India is home to the world's largest population of children with vision impairments (CVIs), many of whom attend under-resourced residential schools for the blind where a significant proportion of teachers are themselves blind or have low vision. The study documents TVIs' everyday teaching practices across three phases — preparation, in-class teaching, and revision/exam preparation — revealing a heavily manual, labor-intensive workflow constrained by delayed Braille textbook delivery, lack of supplementary materials, limited brailler access, and institutional restrictions on technology use. TVIs creatively bridge gaps using YouTube, Google, WhatsApp, and voice assistants, but these efforts remain informal, fragmented, and unsustained. The study then examines TVIs' initial encounters with ChatGPT and Be My AI, finding mixed results: some teachers showed enthusiasm and curiosity, but accessibility barriers (inaccessible login processes, screen reader incompatibility, English-only interfaces), usability challenges, and lack of local language support led to frustration and abandonment for many. The paper positions TVIs not as passive recipients of technology but as expert educators and potential co-designers of accessible educational AI systems.
Key findings
The study reveals a fundamental paradox: TVIs express enthusiasm and commitment toward using technology, yet systemic and institutional barriers significantly limit adoption. Key findings include: (1) Braille textbooks arrive late in the academic year, forcing teachers to rely on personal knowledge and external digital resources to prepare lessons; Braille documents degrade over time requiring constant rewriting. (2) Only 1 of 15 interviewed teachers had access to a functioning brailler (cost: ~INR 45,000/550 USD). (3) Institutional norms actively discourage technology use — using YouTube in class is perceived as "not teaching," and teachers must seek special permission for even basic technology integration. (4) TVIs teach an average of 6.5 grade levels simultaneously (some up to 12-13), across multiple subjects they may not be trained in, creating enormous workload burdens. (5) TVIs depend on sighted colleagues for STEM subjects requiring visual explanations (diagrams, experiments), creating inequitable power dynamics. (6) India's Right to Education Act mandates automatic student promotion regardless of mastery, leading to students in higher grades without basic Braille literacy. (7) NGO interventions (like Vision Empower) have proven transformative for schools they reach — providing tactile STEM materials, teacher training, and digital onboarding — but coverage remains fragmented and limited. (8) The survey across 85 teachers in 4 states confirmed that infrastructure gaps (internet access as low as 17% in Odisha), overwhelming workloads, and institutional barriers are consistent nationwide, with no clear relationship between salary levels and technology adoption. (9) When introduced to ChatGPT, several teachers attempted to use it but encountered accessibility barriers that led to abandonment; Be My Eyes/Be My AI saw slightly better traction for immediate daily tasks but limited classroom integration.
Relevance
This study makes a critical contribution by centering disabled educators — an almost entirely overlooked group in HCI, accessibility, and educational technology research. For the accessibility community, the key insight is that GenAI's inclusive potential cannot be realized through technology alone; it requires systemic change at institutional, infrastructural, and policy levels. The finding that school cultures actively resist technology use is particularly important — it means even perfectly accessible AI tools would face adoption barriers without broader institutional buy-in. The paper's argument for moving from "assistive technology retrofits" to co-designed, contextually grounded systems resonates with broader disability justice principles. For practitioners developing GenAI tools for education, the study highlights the need for local language support, screen reader compatibility from the start, offline functionality, and designs that account for the multi-grade, multi-subject teaching reality of Global South schools. Limitations include the Karnataka-focused qualitative sample, potential selection bias toward more digitally engaged TVIs in the survey, and the absence of school administrator perspectives.
Tags: generative AI · vision impairment · education · Global South · India · teachers · inclusive education · digital divide · assistive technology · schools for the blind · STEM accessibility
Standards referenced: Right to Education Act (India)