Artificial Intelligence Predicts Neighborhood Politics from their Cars

Is this amazing or what? Artificial intelligence uses cars to predict the politics of a neighborhood. Brand new algorithms used by computers are able to take images that are available publicly from the street view in Google to predict political leanings of neighborhoods by simply observing the cars parked on their streets.

Through the use of these algorithms which are able to observe and learn, scientists were able to evaluate millions of images for this new study, which appeared in the publication, Proceedings of the National Academy of Sciences.

Learning about Communities

“Using easily obtainable visual data, we can learn so much about our communities, on par with some information that takes billions of dollars to obtain via census surveys. More importantly, this research opens up more possibilities of virtually continuous study of our society using sometimes cheaply available visual data,” said Fei-Fei Li, who is a computer science associate professor from Stanford University and also the director of Stanford Artificial Intelligence Lab, where this study took place.

Computers that are Able to See and Learn

Li is quite an expert in deep learning and computer vision, which is actually a type of artificial intelligence where computers are able to teach themselves how to identify and recognize three-dimensional objects within two-dimensional images—computers that are to see, as she likes to describe them.

These researchers taught these algorithms—or, rather they taught themselves—just how to recognize the model, the make, and the year of every automobile built since 1990 from over some 50 million Google Street View photos located within about 200 American cities.

The data collected on different car types and their locations were compared with the most current and comprehensive demographic data that is being used today from the American Community Survey. After that, data from the presidential election was used to estimate demographics from factors like race, income, education, and even the preference of the voters.

Li and her research discovered a very simple linear correlation that existed between cars, demographics, and their political persuasion. The verbiage in the study described these societal associations as “simple and powerful,” in their final report.

For example, if the number of sedans within a neighborhood was greater than say the total number of pickup trucks, then there was about an 88% chance that it was a Democratic community. If you were to reverse those numbers where there were instead more pickups than sedans, then there was about an 82% chance it would vote for a Republican.

Rapid algorithms

To extend past these political ramifications, these scientists feel that these algorithms might even help provide even more timely and ongoing supplements to future demographic surveys. The American Community Survey is in the process of conducting through expensive and very labor-intensive door-to-door surveying and canvassing which costs the United States over $250 million every year.

Even at those costs, the worse part about it is the horrible lag time between gathering data and then the publication, which could easily stretch beyond two years or even longer; this is especially true for smaller cities and remote rural areas.

The work done by Li can piggyback upon a database that is always publicly available and also provides an image database that is regularly updated. And it’s built and also paid for by Google Street View, and the data is generated in practically real-time.

“I don’t see something like this replacing the American Community Survey, but as a supplement to keep the data up to date,” said Timnit Gebru, who was the first author of the research paper and was a former member of Li’s lab. Gebru is currently a postdoctoral researcher with the Fairness Accountability Transparency and Ethics (FATE) within the Artificial Intelligence group located at Microsoft Research.