AI Fairness for People with Disabilities: Point of View

Disability fairness requires a fundamentally different approach than other protected categories. The paper argues that fairness frameworks designed for attributes like gender or race don't directly transfer to disability, because disabilities manifest in extremely diverse ways and people adapt differently to them—making one-size-fits-all fairness definitions inadequate.

Disability data is uniquely sensitive and often hidden, complicating fairness efforts. People with disabilities frequently don't disclose their status due to discrimination concerns, which means AI systems may lack the disability information needed to implement fairness measures, creating a catch-22 for responsible AI development.

Standard fairness definitions may not work well for disability contexts. The paper explores how existing fairness definitions perform in disability applications and suggests they have significant limitations, implying that new fairness frameworks specifically tailored to disability may be necessary.

The extreme diversity of disabilities demands individualized consideration. Rather than treating disability as a single category, the paper emphasizes that the wide variety of disabilities and individual adaptations means fairness approaches must account for this heterogeneity rather than applying uniform solutions.