Toward Fairness in AI for People with Disabilities: A Research Roadmap

AI systems often fail for people with disabilities despite being designed to help them: The paper identifies a critical gap where widely deployed AI technologies—including speech recognition for deaf and hard of hearing users and communication aids for people with speech disabilities—may not function properly for their intended users or could actively discriminate against them, even when disability inclusion is a stated goal.

Fairness in AI for people with disabilities has been largely overlooked: The authors note that considerations of fairness specifically for people with disabilities (PWD) "have thus far received little attention" in AI research, suggesting this is an underexplored area that requires urgent focus from the research community.

Different disabilities require different AI safeguards: The paper presents a risk assessment framework mapping how various classes of AI technology may interact differently with various classes of disability, indicating that one-size-fits-all fairness solutions are insufficient and that disability-specific considerations are essential during AI design and testing.

Systematic testing and data collection are needed: The authors call for future research to "gather data, test these hypotheses, and build more inclusive algorithms," suggesting that responsible AI development for PWD requires deliberate empirical investigation rather than assumptions about how systems will perform.