Are crash-avoidance detectors less accurate with people of color? |
Are automated crash-avoidance systems, critical to self-driving vehicles, worse at detecting people with darker skin?
Research at the Georgia Institute of Technology suggests the answer is yes, and now U.S. Rep. Bobby Rush, D-Ill., says he's making progress with a bill to mandate a federal study of whether systems indeed are a particular threat to people of color.
Under Rush's bill, the U.S. Department of Transportation would be directed to evaluate how accurately crash-avoidance systems detect pedestrians, bikers and others with darker skin compared with those with lighter skin.
"I have long supported the development of autonomous and semiautonomous vehicles," Rush said in a statement. "But we have to be absolutely certain on a federal level that these vehicles are equipped to protect all Americans equally" as more self-driving vehicles hit the road.
"The Crash Avoidance System Evaluation Act will bolster our understanding of how lifesaving, crash-avoidance technologies can be better implemented," said U.S. Rep. Larry Bucshon, R-Ind., who is co-sponsoring the bill.
The Georgia Tech study concluded that standard technologies "appear to exhibit higher precision" in detecting people with lighter skin shades. "We hope this study provides compelling evidence of the real problem that may arise if this source of capture bias is not considered before deploying these sort of recognition models," the report said.
Rush said his bill is backed by groups including the Center for Auto Safety, the National Safety Council and the League of American Bicyclists.
In a statement released by Rush's office, Jason Levine, executive director of the Center for Auto Safety, said the CASE Act "will help provide objective information to ensure that when this technology becomes a standard feature of all new vehicles, it can effectively spot everyone on the road and thus prevent unnecessary tragedies."
— Greg Hinz, Crain's Chicago Business