Articles
Original articles on digital accessibility, research, and practice.
4 results found tagged large language models.
-
Trends in Digital Accessibility Research: What Researchers Are Investigating, Finding, and Flagging (2020-2025)
Six years of peer-reviewed accessibility scholarship reveal shifting priorities, emerging populations, and uncomfortable gaps. This article examines trends in what accessibility researchers are choosing to study — from generative AI and neurodivergent experiences to intersecting identities and the Global South — what they are finding, and what they say is still missing.
-
Seeing Data Differently: How Accessibility Research Is Rethinking Charts, Graphs, and Visual Information
Data visualisation is one of the most persistent accessibility challenges — and researchers are moving beyond alt text to multimodal, interactive, and community-driven approaches. This article examines approximately 80 papers spanning the screen reader gap, sonification that achieves performance parity with sighted users, tactile graphics from embroidered textures to data comics, dashboard navigation, AI-assisted interpretation, the overlooked needs of low-vision magnifier users, and the shift from data consumption to data creation by blind users.
-
Machine Learning and Digital Accessibility: What Works, What Fails, and What Gets Lost in Between
AI is simultaneously the most promising tool and the most documented source of harm in recent accessibility research. This article examines approximately 235 papers revealing where machine learning helps — captioning accuracy improved by 5.6%, web navigation 52% faster, audio description production halved — and where it fails: fabricating image descriptions, reproducing ableist stereotypes in career chatbots, homogenising personalised AAC tools, and systematically excluding anyone whose body or speech falls outside training data norms.
-
AI Safety and Disability: What Accessibility Research Reveals About Hallucination, Bias, Privacy, and the Protocols That Do and Don't Exist
Disabled people depend on AI systems they cannot verify, trust AI outputs that fabricate content, and share intimate data with platforms that offer no disability-specific protections. This article examines approximately 135 papers documenting 79 AI harm incidents, the 4.9x improvement in unreliable claim detection when AI uncertainty is surfaced, zero disability-specific privacy protections across 18 AT policies, sycophantic AI behaviour, age stereotyping, career chatbot fabrication, and the protocols that do and do not exist to protect disabled users.