It doesn’t take extra sensory perception to know that American cities are changing — and fast. On any given street in New York City, there’s something opening, another thing closing, and a third being totally redesigned. The same is true for urban spaces stretching from Seattle to Atlanta.
But according to Nikhil Naik, an expert in computer vision with a Ph.D. from the Massachusetts Institute of Technology, no one’s really been collecting good, hard data on the physical changes cities are undergoing. At least not until this week when his work was published in a new study.
That publication is the Proceedings of the National Academy of Sciences, wherein Naik and his fellow MIT researchers explain the results of a machine-learning-aided analysis on a decade’s worth of Google Street View data.
Under the project name Streetchange, the interactive maps chart changes between 2007 and 2014 in New York, Detroit, Boston, Baltimore, and Washington, D.C.
Within those cities, images of each census tract were run through a computer. The machine had been taught to “see” differences between safe-looking areas, like Brooklyn’s brownstones, and traditionally seedier areas, like waterfronts and barren weed-filled back parking lots. With this knowledge, an algorithm could assign a “Streetscore” to each place, quantifying changes to its perceived safety. To ensure the accuracy of the study, the street bot’s suggested changes were confirmed by humans.
The results were striking, as the interactive maps and side-by-side Google Street View images show:
Below is a video showing areas of high change between 2007 and 2014 in New York City.
Naik tells Inverse he hopes the data will ultimately find its way into the hands of urban planners and political scientists. Until Streetchange, “there [wasn’t] good data on physical changes,” Naik says. “Unlike surveys for demographics … it’s kind of hard to quantify physical change.” Now that there is, he thinks experts will be surprised to find the impact physical features have on how people live.
For example, the MIT team found evidence of a “spillover effect” between neighborhoods, validating an old urban planning idea called the “invasion hypothesis.” As common sense would predict, when one part of the city is changed, the outlying neighborhoods really were affected.
This phenomenon was seen across multiple cities but was perhaps most stark near New York City’s High Line, an old elevated railroad that was recently turned into a public park. Greening that neighborhood led to a whole host of new developments in an area many once avoided. Naik hopes this confirmation will encourage planners to think about what development and policy changes do to peripheral communities.
Additionally, the MIT team found it further validated the idea that areas with a large number of college-educated adults are more likely to undergo physical improvement, and that neighborhoods that were already aesthetically pleasing are more likely to be improved than truly run-down sections of the city.
The Streetchange team wants other researchers to take its methodology and apply it to other rapidly changing cities around the world, from Jaipur, India, to Casablanca, Morocco. “Currently, there’s so much available about cities in terms of images from Street View or satellites,” Naik says. “And we also have really good algorithms with A.I. that can analyze the data. And we were thinking, ‘What are these things good for?’”
In recent years, artificial intelligence and machine learning have been relied on more and more to assess real-world environmental issues. Recently, Carnegie Mellon University teamed up with a satellite company to show that poverty can be predicted from space. Now, it’s just a question of how this information will be used on the ground.