Artificial Intelligence Fairness in the Context of Accessibility Research on Intelligent Systems for People who are Deaf or Hard of Hearing

Disabled people are underrepresented in AI training data: The paper emphasizes that AI systems for people who are Deaf or Hard of Hearing often fail because training datasets lack sufficient representation from this population, leading to systems that don't work reliably for their actual needs.

AI systems lack transparency about how they make decisions: The paper highlights that many AI-based accessibility tools don't explain their reasoning to users, making it difficult for people with disabilities to understand why a system failed or to trust it with important tasks like communication.

Current evaluation methods may not catch real-world problems: Standard AI performance metrics don't adequately test whether accessibility technologies are actually ready for deployment or trustworthy enough for users to rely on them in daily life.

AI systems can change what abilities users need to have: The paper warns that as AI becomes more prevalent in accessibility tools, it may inadvertently create new barriers by requiring users to develop different skills or adapt to system limitations rather than the technology adapting to human needs.

Researchers and companies have ethical obligations to disabled communities: Those developing AI-based access technologies bear responsibility for ensuring their systems are fair, transparent, and genuinely beneficial—not just technically functional—for people with disabilities.