Penny is an artificial intelligence (AI) program built by Stamen Design and researchers at Carnegie Mellon University on top of GBDX, DigitalGlobe’s analytics platform, to predict neighborhood income levels based solely on satellite imagery. Penny analyzes the shapes and structures that make up city blocks and labels neighborhoods as high, median, or low-income areas.
To train Penny, the team began by using financial data from the U.S. Census Bureau to color-coordinate tiles around cities, essentially creating citywide income maps. Those maps were laid over corresponding DigitalGlobe satellite imagery to create a correlation between urban building types and income levels. With those two datasets, the AI’s neural network learned over time how to distinguish features of wealth from features of poverty.
After running through these datasets, Penny can predict income levels with an average accuracy of 86 percent. Baseball diamonds, parking lots, and large, identical buildings like housing projects indicate lower income areas. Single-family homes and apartments are found mostly in middle-high or middle-low income areas, depending on other factors like freeways and trees. High-rise buildings, homes with large yards, and parks mark high-income neighborhoods.
Penny has implications beyond simple pattern recognition. It’s intended to challenge our understanding of how place and people’s well-being relate, and is a bold step toward answering large-scale, abstract questions using non-human intelligence. In the future, AI programs inspired by Penny could help urban planners build smarter, more efficient, and more beautiful cities.
Users can test Penny for themselves, dragging and dropping individual features (a pool, an airplane, an apartment high-rise) into a small region of the virtual cityscape to see how they affect the algorithm’s understanding of the area’s income. The AI is currently available for free use with imagery of New York and St. Louis.
Photo Credit: DigitalGlobe