From Information Seeking to Empowerment: Using Large Language Model Chatbot in Supporting Wheelchair Life in Low Resource Settings
Wen Mo, Aneesha Singh, Catherine Holloway · 2024 · Proceedings of the 26th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS) · doi:10.1145/3663548.3675609
Summary
This paper explores the deployment of Wheelpedia, a WhatsApp chatbot powered by GPT-4, as a design probe to support wheelchair users and professionals in Nigeria and Kenya over a two-month period. The WHO estimates that approximately 80 million people globally require a wheelchair, yet in low and middle-income countries (LMICs), access to wheelchair service information, professional training, and appropriate equipment is severely limited. The chatbot was built on WhatsApp due to its dominance as a messaging platform in these regions, and could process text, voice (via Whisper transcription), and images (via GPT-4 Vision). The study involved 18 semi-structured interviews (14 wheelchair users and 4 professionals) and analysis of 471 messages from 42 participants. The research identified five overarching themes: how users interacted with the chatbot (question domains, multimedia use, broad versus specific queries), acceptance factors (answer quality, convenience, communication style, novelty), applications beyond information seeking (consultation, education, empowerment, stigma reduction), trust management strategies (testing with known answers, reputation, community leader endorsement, transparency), and drawbacks and concerns (underrepresented local dialects, low digital literacy, limited smartphone access, expensive data, privacy uncertainty).
Key findings
Analysis of 471 messages revealed that 46% related to wheelchair life topics, 31% were user engagement messages (greetings, feedback), and 23% covered other topics. Among wheelchair-related questions, 45% focused on wheelchair service steps (assessment, procurement, maintenance, fitting), 17% on health and wellbeing, and 12% on wheelchair education. About 55% of questions sought advice and recommendations, while 59% were categorized as broad in scope, indicating users struggled to formulate precise queries. Despite being informed about voice and image capabilities, 94% of messages were text-only, reflecting unfamiliarity with multimedia chatbot interactions. Participants expressed overwhelming enthusiasm, viewing the chatbot not just as an information tool but as a means of empowerment — users reported feeling ownership over their wheelchair knowledge and gaining confidence in self-advocacy. Professionals valued it for expanding their consultation capabilities and as training material. Notably, participants saw potential for the chatbot to reduce social stigma around wheelchairs by educating non-wheelchair users. Trust was established primarily by testing the chatbot with known answers first, and was bolstered by endorsement from trusted community leaders.
Relevance
This research provides critical evidence for deploying AI-powered accessibility tools in resource-constrained settings where professional support is scarce. For accessibility practitioners and organizations, the findings highlight that LLM chatbots can serve multiple roles beyond information delivery — as educational tools for caregivers and professionals, emotional support companions, and instruments for reducing disability stigma. The study identifies essential design considerations for LMIC contexts: supporting local dialects and languages, optimizing for low-bandwidth and expensive data environments, providing image-based responses for wheelchair assessment, incorporating FAQ templates to help users with low digital literacy formulate questions, and building trust through community leader endorsement. The paper also raises important cautions about AI tools potentially replacing professional consultation and the need for transparent disclaimers and culturally appropriate content.
Tags: wheelchair · assistive technology · chatbot · large language model · low and middle-income countries · WhatsApp · information access · Global South · empowerment