Are self-driving cars risky for black pedestrians?

Mar 07, 2019, 2:17 AM EST
(Source: Michael Coghlan/flickr)
(Source: Michael Coghlan/flickr)

A recent study warns of a new blind spot with self-driving cars, which already seem to be way off the mark over their safety features, potential impact on traffic congestions and handling of tricky on-road situations.  

The study at the Georgia Institute of Technology cautions that autonomous vehicles may not spot dark-skinned pedestrians with the same accuracy as they detect their white peers, an infirmity that stems from algorithmic bias, which in turn originates from implicit biases in human developers, notes Vox.

For the study, the researchers ran a trial with a dataset, comprising of images of pedestrians from diverse demographic groups. The object-detection models, similar to the ones used in autonomous cars, then tried to detect people in these images, which were sorted according to the Fitzpatrick scale that classifies human skin tone from light to dark.  

The results bare open a flaw, which researchers believe, contributes to higher error rates by self-driving cars in certain demographics, writes The Independent. The image-recognition systems showed a five percentage point decline in detecting people with average dark-skin tones.