A research team from Stanford University has made a demographic analysis of US cities by using artificial intelligence (AI) on Google images of street views across the United States, the New York Times reported.
The study, led by computer vision scientist Timnit Gebru, analyzed 50 million images of street views from Google Street View, where 22 million images of cars were identified.
These vehicles – located in more than 3,000 US postal codes and 39,000 voting districts – were classified into more than 2,600 categories such as their make and model.
Researchers then linked up the findings with other data sources to predict factors such as income, race, education, pollution and even voting patterns at the neighborhood level, across the country.
The research results reveal that New York has got the most expensive cars, San Francisco has the largest percentage of foreign cars, and Chicago is the city with the greatest income disparities, having plenty of expensive and cheap cars in various neighborhoods.
The greenest American city, according to car attributes, is Burlington, Vermont, while the city with the biggest per-capita carbon footprint is Casper, Wyoming.
The first step of the project was to train the AI software to understand the images. Prior to this, a database had to be built.
The researchers enlisted hundreds of people to classify cars among millions of images. Some of the contractors simply identified the cars in the images, while car experts were able to spot the subtle differences among the vehicles.
After the car-image database was formed, the AI was able to classify the cars in those 50 million images in two weeks. For the same task, it would take a human expert about 10 seconds on an image or more than 15 years to complete the entire project.
This article appeared in the Hong Kong Economic Journal on Jan 3
Translation by Jonathan Chong
[Chinese version 中文版]
– Contact us at [email protected]