Designing with Tensions: Understanding Professionals' Needs in Integrating AI Chatbots for Wheelchair Assessment Services in Low- and Middle-Income Countries
Wen Mo, Aneesha Singh, Amid Ayobi, Catherine Holloway · 2025 · Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2025) · doi:10.1145/3663547.3746340
Summary
This paper investigates the potential and limitations of integrating AI chatbots into wheelchair assessment services in low- and middle-income countries (LMICs), where approximately 65-95% of people needing wheelchairs lack access to one. The researchers conducted a two-part study with 11 rehabilitation professionals (physical therapists, occupational therapists, and prosthetists/orthotists) from Africa and South Asia (primarily Kenya, plus India, Pakistan, South Africa, and Uganda). Part one consisted of semi-structured interviews exploring current wheelchair assessment practices and challenges, revealing a six-stage workflow: background interview, physical assessment, prescription, fitting, training/rehab planning, and follow-up. Part two used two chatbot design probes—"Ask Wheelie" (a consultation tool guiding professionals through assessments) and "Wheel Care" (a follow-up care chatbot prescribable to clients)—to elicit situated feedback on chatbot integration possibilities. Current practice remains overwhelmingly paper-based, with nearly all participants (10/11) using pen and paper during assessments, generating extensive notes that are time-consuming to digitise and frequently lost during fieldwork. Professionals face significant workload pressure from resource constraints, limited wheelchair and repair component availability, understaffing, and the need to see as many clients as possible. The study surfaces how professionals weighed new chatbot interactions against potential disruptions to their current practice, revealing competing pulls between efficiency gains and workflow changes.
Key findings
The study identified 13 distinct tensions that arise when envisioned chatbot use misaligns with professional practice, organised into three interconnected domains. Anchored Values (practice-based) include tensions around professional autonomy (experienced professionals feeling chatbot guidance unnecessary), adapting in-situ (structured chatbot templates conflicting with fluid, off-script assessment interviews), communicating nuances (chatbot text interfaces unable to convey subtle physical observations), and expressive freedom (chatbot Q&A format restricting professionals who rely on sketches, body maps, and non-linear annotations). Anchored Values (relational) include tensions around establishing rapport (chatbot use during sessions potentially making interactions more transactional) and building trust (clients potentially perceiving phone-based chatbot use as unprofessional). Practice Structure tensions include shifting workflows, evolving responsibilities (chatbot offloading creating new coordination burdens), and transforming data practices (digitisation altering information flow between staff). Contextual Readiness tensions include confidence in AI for local contexts (doubts about AI handling local dialects and contextual nuances), trust in data safety, access to technology (unreliable internet, limited devices), and digital confidence and literacy. These tensions operate not in isolation but as interconnected domains where changes in one area cascade across others.
Relevance
This paper makes an important methodological contribution through its tension-informed design framework, which reframes adoption barriers as design opportunities rather than obstacles to overcome. For assistive technology service providers and digital health designers working in LMICs, the framework provides a structured approach for anticipating and navigating the competing demands that emerge when introducing AI tools into established professional workflows. The study's emphasis on building human capacity before introducing technology—aligned with Toyama's Technology Amplification Theory—offers a critical counterpoint to techno-solutionist approaches. Practical design recommendations include offline-first or low-bandwidth chatbot capabilities, compatibility with basic smartphones and existing communication channels like WhatsApp, customisable chatbot roles (assistant, teacher, note-taking tool), multimodal input supporting voice and images alongside text, and modular onboarding with basic and advanced modes. The finding that simply introducing chatbots is insufficient without adequate training, infrastructure support, and gradual capacity building has broad implications for any organisation deploying AI tools in resource-constrained assistive technology services.
Tags: wheelchair provision · AI chatbot · assistive technology services · low- and middle-income countries · digital health · wheelchair assessment · tension-informed design · rehabilitation
Standards referenced: WHO Wheelchair Service Training Package