Why it is much harder for AI to overcome ableism

Nov 29, 2018, 7:20 AM EST
(Source: Matthias Ripp/flickr)
(Source: Matthias Ripp/flickr)

The race and gender prejudice of artificial intelligence are an open secret. To be fair, a respectable degree of attention, research and development is being reserved to address this problem. But what may have missed a wider scrutiny is the ableism in artificial intelligence, which is far more deep-seated, subtle, ingrained and tough to weed out than racism and sexism.

IBM researcher Shari Trewin explains that the bias against disability is too intricate and diverse to be modeled as a simple variable as in case of gender or race, writes MIT Technology Review. The forms of disabilities, their varying degree of severity and their dynamic existence, all these parameters combined, don’t fit in with a simple variable for the algorithm to work.

Besides, the sensitive nature of disability information keeps many from disclosing it, resulting in AI systems that are starved for true data and hence, unwittingly biased against people belonging to this group.