Using Artificial Intelligence to Balance Agency and Efficiency in AAC Devices for Speech-Impaired Users

The speech-impaired population often encounters barriers to effective self-expression through speech, which can limit social participation. Many use Augmentative and Alternative (AAC) devices to facilitate communication and interaction. Recent advances in artificial intelligence (AI) have introduced features such as predictive word and phrase suggestions to improve communication efficiency. However, while these systems enhance efficiency, they can constrain users’ sense of agency and authentic self-expression. Prior work has largely examined efficiency and agency as separate design concerns, leaving a gap in approaches that address both simultaneously. To address this gap, this study investigates how AI can be designed to balance efficiency with user agency in AAC systems. We adopt a research-through-design approach, using iterative prototyping informed by human factors principles of controllability, self-representation, and user override. The proposed multimodal system enables users to capture contextual input via images, specify emotional tone, and generate keyword-guided sentence suggestions. Users can review, modify, accept, or reject AI-generated outputs at every step, thereby preserving user control. This work explores how multimodal AI could be designed to support communication efficiency while preserving user agency. It advances a design perspective that frames efficiency, agency, and expressivity as interconnected human factors, informing the design of future AI-assisted AAC systems.