Two researchers from the University of Washington have found a way to estimate a U.S. city’s obesity level without having to look at its inhabitants.
The duo trained an artificial intelligence algorithm to find the relationship between a city’s infrastructure and obesity levels using satellite and Street View images from Google. By understanding how city planning influences obesity, health campaigns and new construction can be coordinated to improve a city’s health, the researchers wrote in a paper published in JAMA Network Open.
The algorithm was trained using more than 150,000 satellite images of six cities, as well as 96 categories of points of interest like grocery stores and pet shops. It was then correlated with obesity rates reported from each city. The researchers included the points of interest because they could have an effect on the activity of a neighborhood. An area with pet shops could have more people taking dogs for a walk, for instance.
Unsurprisingly, the algorithm correlated areas with more green spaces to walk and more spacing between buildings with lower obesity rates. This is indicative of wealthier neighborhoods having fewer obese residents. The researchers acknowledged that the entangled relationship between income and health could skew the algorithm. However, through further validation tests, the team found the algorithm actually found a link between the number of buildings and green space and obesity, not just wealth.
The study was singularly based on U.S. data, so the algorithm is unlikely to work in other countries without adjustments. Different approaches to city planning, and more or less active cultures would certainly be misunderstood by the algorithm, as has been shown with bias in so many other applications of artificial intelligence.