Researchers spent two weeks training the algorithm to go through the roughly 22 million cars that were pictured in 50 million Google Street View images. Then, computers were able to file each into one of nearly 3,000 categories -- broken down by make, model, and year, researchers said.
If a person were doing the same work, the study said, it would have taken about 15 years to complete (assuming it took 10 seconds to catalog each image.)
On the demographic side of the equation, the study found that Volkswagens and Aston Martins tend to be found in predominantly white areas. African-American neighborhoods, meanwhile, are more like to have Chryslers, Buicks and Oldsmobiles driving around or parked on the street. Asian neighborhoods were more likely to have Hondas or Toyotas, the study found.
And make and model weren't the only useful data points researchers identified.
"If you walk around a neighborhood looking at cars, the density of traffic sometimes tells you things as valuable as the types of cars you see on the streets," Timnit Gebru, a study author, said in a statement. "We can use all this information in our algorithms."
Gebru hopes the algorithm used in the study could someday help monitor carbon dioxide levels, or even improve traffic on congested streets.
(c)2017 McClatchy Washington Bureau
Visit the McClatchy Washington Bureau at www.mcclatchydc.com
Distributed by Tribune Content Agency, LLC.