A recent report has demonstrated that the systems designed for the sake of helping the autonomous cars to recognize the pedestrians might have considerable trouble when it comes to identifying people with darker skin tones. The worrying research has been uploaded to the preprint server arxiv.
There have been pieces of evidence that suggested that some of the facial recognition software struggled heavily when it came to working with darker skin tones. However, the outcomes of the research into autonomous cars imply very dangerous outcomes. A team of researchers from Georgia Tech carried out an investigation of eight AI models that are incorporated into the state-of-the-art object detection systems for the sake of completion of its study. The investigated systems enable the autonomous vehicles for determining pedestrians, road signs, and other objects while navigating roads.
The systems were tested by making use of two categories based on the Fitzpatrick scale. The Fitzpatrick scale is generally used for classifying human skin color. The accuracy of the system decreased by a whopping 5% when it was subject to groups of images of pedestrians that had darker skin tones. As per the published paper, the models displayed ‘uniform poorer performance’ when tested using the three darkest shades on the scale.
The results have been announced after adjusting the outcomes for whether the picture was taken during the night or the day. In short, the report dictates that people with darker skin tones won’t be as safe as people with lighter skin tones close to roads that are filled with autonomous vehicles. However, the report also provides a solution for this worrisome problem; simply increase the number of images of dark-skinned pedestrians in the data sets that are being used for the training of these systems.
The authors of the report state that it provides enough evidence to address this problematic issue before incorporating recognition systems all around the world. It has been noted that the engineers in charge of the development of these systems must lay more emphasis on training the systems to achieve higher accuracy for this group of darker skin tones.
It is high time that the AI development community comes together and addresses this issue before it ends up creating chaos. What do you think?