a paper published earlier this year, Stanford computer scientist
Timnit Gebru wrote about how neighborhoods can be evaluated by the
makes and models of the cars parked in their driveways. The paper appeared
in the Proceedings of the National Academy of Sciences and it's an
analyzing the images already available as part of Google
Street Views, the research team was able to identify which
neighborhoods were Republican and which were Democrat as well as
many other characteristics.
determined that in those areas where the number of sedans is higher
than pickup trucks, there’s an 88 percent chance of the district
voting Democratic. Where there are more pickup trucks, there’s an 82
percent chance it’s a Republican-voting district.
project devised an automated methodology that estimated the social
characteristics of regions covering 200 U.S. cities based on
analyzing 50 million images from Street Views. The images were
originally created by Google sending cars through every neighborhood
in the country, capturing images that are then displayed and
accessed on Google Maps. Their automated process took two weeks,
compared to 15 years if the images had been analyzed by hand.
automated process to analyze the images was accomplished using
computers and artificial intelligence software called “convolutional
neural networks” that learned to recognize the vehicles by
identifying unique features on each. That allows the computer to
identify the make and model, year, value, and fuel efficiency of the
characterize the automobiles, they hired Amazon Turk workers to
develop a library of car images from Edmunds.com, Cars.com, and
Craig’s List. Their data came up with 2,657 visually distinctive
categories, covering cars found in the U.S. since 1990.
what else did their analysis show from the automobile information?
They came to the following conclusions, as quoted from the report:
and Toyotas most strongly indicate an Asian neighborhood.
Buicks and Oldsmobiles “are positively associated with
trucks and Volkswagens are associated with white neighborhoods.
are most associated with Democratic voter precincts;
Republican-leaning precincts are most associated with extended-cab
researchers noted how this process could be a supplement or even a
substitute for the way census data is now acquired because they
found good agreement between their findings and those from the
also noted that the U.S. spends more than $250 million each year on
the American Community Survey (ACS) that sends workers door-to-door
to interview the residents in each home in order to gather
statistics relating to race, gender, education, occupation, and
more. The Census Bureau conducts their survey once every 10 years.
While both are more accurate, they each take a long time to analyze
and don't pick up recent trends.
research team even hypothesized that when self-driving cars become
available, they can be used to scurry around our neighborhoods,
quickly accumulating even more data.
a fascinating discovery, it will require a leap of faith and more
validation for us to believe we are what we drive.