Advances in artificial intelligence show promise in making the world more accessible to people who are disabled. However, it still faces several major hurdles.
A machine learning model can only be as good as its training dataset. The lack of training data representing disabled people has been coined the “disability data desert.” Some companies are specifically trying to address the data desert by recruiting disabled people to generate data, but there are many ethical considerations.
Despite these ongoing challenges, some accessibility companies are overselling their capabilities, selling packages of “accessibility” solutions powered by artificial intelligence.
An increasing amount of regulation in the US and Europe requires companies to make their products accessible, and the number of digital accessibility lawsuits has been increasing. However, when companies look to bring in AI solutions, very little information is made available so they can determine whether it meets regulatory requirements and improves experiences for people with disabilities.
Machine learning has been good at categorising people into buckets and throwing out “outliers.” Some people say that it’s impossible to capture the wide spectrum of people with disabilities. However, the issues around representing disabled people are not outliers. The Financial Times' investigation on the shortcomings of AI toward the disability community provides a lens into the wider issues and challenges that our society and regulators need to address — and soon.
