Before the Technological Fix: Scoping AI and AAC for Social Futures
Seray B Ibrahim, Tom Griffiths, Michael Clarke, Simon Judge, Petr Slovak, Graham Pullin, Jeff Higginbotham · 2026 · Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems (CHI EA '26) · doi:10.1145/3772363.3798436
Summary
Ibrahim and colleagues - a multi-institution team of researchers working at the intersection of AAC, HCI, and disability studies - argue that the recent surge of HCI work applying AI to Augmentative and Alternative Communication (AAC) is drifting away from where contemporary AAC scholarship has been heading. Modern AAC research treats communication as collaborative, multimodal, and contextually distributed, with conversation partners as co-constructors of meaning; recent HCI papers, in contrast, often optimize for linguistic efficiency (speech rate, prediction accuracy) within an implicit sender-receiver model and a deficit framing of disability. The paper presents initial findings from a 20-year scoping review (2005-May 2025) of the ACM Digital Library and Scopus, following Joanna Briggs Institute and PRISMA-ScR guidance, that asks one question: what design problem did the researchers hope to solve with AI? After screening 855 unique papers and snowballing, 65 papers were included. Three authors independently coded a 15%% sample, refined categories collaboratively, and the first author then coded the remaining papers with weekly verification. The team mapped each paper along two dimensions - technology-centred vs person-centred, and individual-focused vs interaction-focused - producing a landscape of how the AI+AAC literature has framed problems.
Key findings
The dominant framing across the 65 papers was technological factors (43/65) - improving accessibility (17/43, e.g. agrammatic-input correction with LLMs for people with aphasia), language prediction speed (16/43, including Dasher gaze typing and utterance-based generation), expressivity (13/43, synthetic AAC data, emotional TTS, humour, storytelling), and personalisation (5/43). Even when AI dramatically improves prediction, the authors note, the practical ceiling is around 40 wpm - well below the 125-185 wpm needed to keep pace with typical conversation. Bodily factors framed 19/65 papers, split between individual body functions/structures (intelligibility, cognition, physical access) and activity/participation (idiosyncratic communication modes, just-in-time vocabulary, conversational scaffolding). Only 11/65 papers explicitly framed the design problem as temporal-relational - addressing conversational management (turn-taking, pre-emption, flow regulation) or establishing shared meaning/common ground. A small set (3/65) raised methodological concerns about user involvement, AI expertise, and credibility. The authors highlight a few exemplar papers (COMPA, GenieTalk, design probes with humor and aphasia) that align with contemporary AAC's collaborative-interaction view, but find these are exceptions. Even where HCI papers do engage with conversation timing, most still operationalize success as keystroke savings or wpm gains rather than as situated agency, partner relationships, or social participation.
Relevance
For accessibility researchers and product teams working on AAC, this scoping review is a useful corrective: it documents, with numbers, that the field has been answering a narrower question (how do we type faster?) than the question AAC users and clinicians have been asking (how do we participate in social life?). The two-dimensional landscape (technology vs person, individual vs interaction) and the four design-problem categories (technological, bodily, temporal-relational, methodological) form a usable evaluative lens for any AAC+AI proposal: where in the space does this work sit, and is that location intentional? The paper is also a directional argument for the CHI community to borrow conceptual tools from AAC scholarship - ethnomethodology, conversation analysis, multimodality, critical disability theory, the ICF model - rather than re-inventing communication models from inside the technology stack. Caveats: as a work-in-progress scoping review the paper does not yet present the full coding framework or the 12-question analysis (those are flagged for later work), the search is restricted to ACM and Scopus, and the team's own epistemic position - mostly AAC researchers - shapes which framings count as 'aligned' with contemporary AAC. The argument is normative as well as descriptive; readers seeking purely empirical mapping should treat the discussion as advocacy.
Tags: AAC · augmentative and alternative communication · artificial intelligence · scoping review · social interaction · accessibility theory · critical disability studies