Fairness Issues in AI Systems that Augment Sensory Abilities

AI systems designed to help disabled people see or hear may inadvertently exclude them from the training process: The paper highlights that accessibility issues with data and models mean disabled users are often underrepresented in the datasets used to train these systems, potentially making the tools less effective for the very people they're meant to help.

Developers must make difficult ethical choices about which sensory information to share: These augmentation systems don't just translate reality neutrally—they decide what information matters enough to convey to users. The paper notes this raises important questions about who decides what's important and whether those decisions reflect the actual needs and preferences of disabled users.

Privacy concerns extend beyond the primary user to bystanders: Sound awareness tools for d/Deaf users and scene description tools for blind users may capture and process information about other people in the environment, raising privacy questions that go beyond typical AI fairness discussions and affect both users and those around them.

These systems present a unique fairness challenge because they provide access to information already available to non-disabled people: Unlike other AI applications, sensory augmentation tools aim to level the playing field by giving disabled users access to information that sighted or hearing people take for granted, making fairness considerations fundamentally different from other AI contexts.