"Computer Says No": Disabled Welfare Experiences and Envisioned Futures Under AI Governance
Humphrey Curtis, Adam D G Jenkins, Alistair Gentry, Sioban Zacharek, Sally McVicker, Timothy Neate, Filip Bircanin · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791025
Summary
Curtis and colleagues investigate how people with aphasia — a language impairment most often caused by stroke that affects reading, writing, speech, and comprehension — experience the increasingly digitised and AI-automated UK welfare state, and how they would redesign it. The study responds to a policy moment in which UK welfare is shifting aggressively to 'digital by default' delivery, mass migration to Universal Credit, reformed PIP eligibility, £5 billion in announced welfare cuts, and a July 2025 OpenAI deal, all against the backdrop of widening digital exclusion for disabled citizens. Over six weeks the authors ran a four-stage participatory design study with 42 co-designers recruited through London charity Aphasia Re-Connect, of whom 29 had aphasia and 17 also had hemiplegia. Stage 1 formed an advisory board and ran unstructured interviews (including shadowing a live PIP-to-Universal Credit migration call). Stage 2 used tangible, plain-language HELP and FEELINGS card decks to rank and emotionally map welfare services. Stage 3 used printed journey-mapping toolkits with WHERE and BARRIER card decks. Stage 4 showed five short design-fiction video clips depicting near-future AI-automated welfare scenarios ('Press 1 to be Denied,' 'Just Want to See the Doctor,' etc.) to elicit reactions. All materials were built for aphasia-accessibility by a disabled artist. 2,141 coded instances were analysed using Braun and Clarke's reflexive thematic analysis in Miro.
Key findings
Six themes emerged. (1) The cost of performing disability: applicants with invisible conditions must enact 'eligibility theatre' to be believed, with PIP forms and phone assessments described as 'humiliating,' 'too personal,' and designed to make you feel you don't have a disability when rejected. (2) Geographies of inequity: access depended less on individual capacity than on mobilising family, advocates, and digital literacy; regional NHS disparities left some participants without a GP visit in years. (3) Navigating digital bureaucracy: logins, 2FA, and smartphone-only portals (Evergreen Life, Patchs, Accurx, Swiftqueue) were unusable for those who cannot read or write fluently; 'system veterans' develop elaborate hacks like timing calls to the second to jump phone queues. (4) The accessibility paradox: supposedly accessible systems trapped applicants in limbo cycles of reapplication and appeals (one appeal was 'coming up in two years' time'). (5) Hostile design: assessments were perceived as deliberate deterrence, with lost forms, hidden cameras in waiting rooms, and obfuscated rules. (6) Envisioned AI futures: co-designers wanted AI dialogues that are patient, multimodal, and supportive; empathetic welfare systems that retain human recourse (via a 'radar key' mechanism to summon a human); and open-source, publicly governed, truthful infrastructure — resisting proprietary black-box deployment and AI-as-scapegoat for austerity cuts.
Relevance
For accessibility practitioners, this paper is a sharp warning about how 'digital-by-default' and AI-automated public services compound exclusion for people with communication disabilities. It offers a tangible methodology — card decks, journey-mapping toolkits, design-fiction videos — for running genuinely accessible participatory design with people with aphasia, reusable in any bureaucratic accessibility context. The paper's policy-relevant implications include: mandatory human-in-the-loop escalation, plain-language explanations of automated decisions (echoing GDPR's 'meaningful information' clause), model cards for public-sector AI, robust complaint pathways, and open-source/publicly governed infrastructures to prevent AI being used as scapegoat-by-outsourcing (cf. Horizon/Post Office, Robodebt, MiDAS, SyRI). Limitations: UK-specific, Global North, self-selected charity membership, and speculative rather than evaluative AI prompts. Practitioners working on government services, chatbots, or AAC-adjacent tools should treat this as essential reading.
Tags: AI · artificial intelligence · AI governance · algorithmic decision-making · aphasia · welfare · digital exclusion · participatory design · design fiction · disability rights · critical HCI · government services · hostile design · communication disability
Standards referenced: GDPR