All current forms of artificial intelligence rely on statistical reasoning to decide and choose. The consequence of this is that AI chooses and decides with the statistical average in available data.
The only common data characteristic of persons with disabilities is sufficient distance from the average population that standard systems do not work. The impact of this heterogeneity and distance from the average is that decisions made by AI are frequently biased against persons with disabilities.
However, AI is also a powerful tool for bridging barriers for persons with disabilities by translating information, performing average tasks and providing access to average content. Unfortunately this results in greater vulnerability to a number of risks.
To optimize the benefits and mitigate the risks requires involving persons with disabilities throughout the AI lifecycle in all roles and functions. Disability is the ultimate challenge for AI and results in the greatest innovation.