Abstract This paper conceptualizes engineered exclusion —the predictable sidelining of disabled users resulting from choices about data provenance, model objectives, and evaluation practices within AI systems. We examine engineered exclusion through the lived experiences of minimally and nonspeaking autistic people whose communicative profiles challenge the speech-centered defaults embedded in contemporary AI pipelines. Here, “voice” refers not only to speech but to the broader architecture through which embodied, multimodal communication—spanning AAC text, gesture, movement, and partial vocalizations—becomes legible within AI systems. “Nonspeaking” is therefore not the absence of language but a heterogeneous spectrum in which communication is often state dependent, varying with fatigue, anxiety, sensory load, and motor planning demands—forms of variation that design abstractions routinely erase. Tracing exclusionary mechanisms across speech recognition, text-to-speech, plain-language systems, and interface design, we introduce measurable designed-dignity metrics for technical evaluation (Table 1) and a governance framework mapping accountability across the AI lifecycle (Table 2). We argue that accessibility must be treated as a core dimension of AI ethics—on par with fairness, privacy, and safety. Re-engineering AI for designed dignity requires systems that recognize embodied, multimodal, and fluctuating forms of communication, expanding what counts as valid signal and responsible innovation.
Hari Srinivasan (Sat,) studied this question.