An AI Algorithm Can Predict How People Will Vote Just by Looking at the Cars on the Street

Researchers at Stanford have developed an artificial intelligence algorithm that can make estimations on the voting preferences of neighborhoods by analyzing visual data from Google Street view.
Jessica Miley

Do you think the car you drive expresses your political leanings? On an individual level maybe not, but researchers have found that by analyzing images of North American suburban streets and comparing the model, and age of the parked cars, they can determine who that neighborhood is most likely to vote for. The images come from openly available Google Street View and the data analysis is done by self-teaching robots. The study comes from Stanford University's Artificial intelligence Lab and the Stanford Vision lab. Fei-Fei Li, director of the labs explains the research: “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.”

Algorithms predict if a precinct votes Republican or Democratic

The algorithms were trained to ‘look’ at Google Street View images and recognize the make, model and year of every car produced since 1990. This data was then compared to a demographic data obtained from the American Community Survey, and against presidential election voting data, the algorithms could then synthesize this information to estimate the neighborhood's overall voting preferences. For example, the algorithms showed that if the number of pickups is larger than the number of sedans there is an 82 percent chance that neighbourhood will vote Republican. In the same vein, if you switch those numbers and there are more sedans parked in neighborhood streets than pickups there is an 88 percent chance that the precinct will vote Democratic.

Research could have implications for other data collections

Li and her team have published their results in the Proceedings of the National Academy of Sciences where they describe the relationship between cars, demographics and political persuasion as “simple and powerful”.


The algorithms will obviously be of great interest to political analysts who use this types of data to plan campaigns. But the researchers have other thoughts about how it could be useful in gathering important data about how and where we live. The American Community Survey is an ongoing survey of the United States population that samples communities, asking them questions about ancestry, educational attainment, income, language proficiency, migration, disability, employment, and housing characteristics. The survey is currently conducted through costly and labor-intensive door-to-door canvassing but research suggests that parts of what the survey sets out to achieve could be done by the algorithms using the freely available Google data.

Timnit Gebru, first author of the paper and former member of Li’s lab describes how the algorithms could assist the traditional survey methods: “I don’t see something like this replacing the American Community Survey, but as a supplement to keep the data up to date.” The research team is excited to see where this type of application of machine learning can go. Li sums up by saying: “There is great potential to use computer vision technology in a constructive and benevolent way.”