Driverless cars, which are still under testing, are less adept at detecting children and darker-skinned pedestrians than lighter-skinned adults, thus increasing the risk of the former being incorrectly identified and involved in a collision. After testing eight pedestrian-detection systems on 8,311 images with 16,070 gender labels, 20,115 age labels, and 3,513 skin tone labels, researchers at King’s College, London, found that detection of adults was 19.67% more accurate than children and 7.52% more accurate for pedestrians with light skin than pedestrians with dark skin; the discrepancy between genders was only 1.1%. The likely reason is that artificial intelligence programs are trained using photos that show more people with light skin than dark skin. “The danger that self-driving cars can pose is acute,” said Jie Zhang, one of the authors, in a press release. “People from minority groups can face severe injury.” According to Zhang, driverless vehicle manufacturers and the government need to work together to ensure equal treatment of all pedestrians (ArXiv.org, August 5; King’s College London, August 23; PCMag, August 28).
RepublishTechnology
Greater risk of being hit
Driverless car, less attentive to minority groups
Marco Verch / Flickr
