DARPA-Funded AI Predicts Future Crimes In Cities with ‘About 90% Accuracy’

Scientists at the University of Chicago have created a new AI program that’s able to forecast when and where crime will occur in a given city up to one week in advance and “with about 90% accuracy.”

Data and social scientists from the University of Chicago (UChicago) have developed a new AI program that forecasts crime by identifying the temporal and spatial patterns of criminal activity in public data for a given city. The tool has already been “validated” using historical data from the City of Chicago around two broad categories of reported events: violent crimes and property crimes. According to a UChicago press release, the scientists’ program “can predict future crimes one week in advance with about 90% accuracy.”

According to UChicago the scientists’ new program—the research for which has been funded in part by the U.S. Government’s Defense Advanced Research Projects Agency (DARPA) as part of an “AI initiative”—“isolates crime by looking at the time and spatial coordinates of discrete events and detecting patterns to predict future events.” The University notes the program divides a given city into spatial tiles that are approximately 1,000 feet across in order to make its predictions, as “relying on traditional neighborhood or political boundaries” would apparently open up the program to bias.

“We created a digital twin of urban environments. If you feed it data from [what] happened in the past, it will tell you what’s going to happen in [the] future,” Ishanu Chattopadhyay, a PhD Assistant Professor of Medicine at UChicago and senior author of the new study, which was published in Nature Human Behaviour, says in UChicago’s press release. “Now you can use this as a simulation tool to see what happens if crime goes up in one area of the city, or there is increased enforcement in another area,” Chattopadhyay adds, highlighting the idea that “If you apply all these different variables, you can see how the systems evolves [sic] in response.”

In a more technical post on Medium Chattopadhyay goes into further detail explaining how the program works—using a “prediction framework” the scientists developed “from scratch” based off of a tool the Chicago PD implemented in 2012. The tool, a formula developed by academic researchers (thanks to a $3.8 million federal grant), was halted when it was found to be “unreliable,” overly reliant on arrest records for predictions, and ultimately unable to sustain funding.

Chattopadhyay et al. then applied their new prediction framework using neural networks—algorithms that, in effect, behave like biological neurons (with varying weights, layers, etc.)—that were able to “directly learn non-trivial aspects of stochastic [i.e. randomly determined] processes.” The scientist adds in his Medium post that “Ultimately [the program’s ability to learn led] to highly precise predictions of individual crimes — precise both in space (within 2 city blocks), and time (1–2 days), made sufficiently in future (~1 week) for appropriate actions or interventions to be actualized.”

UChicago reports the scientists also generated a second prediction framework in order to study “the police response to crime” by looking at the number of arrests following incidents and comparing those rates among neighborhoods with different socioeconomic status. Chattopadhyay et al. say that with the framework they observed that crime in wealthier areas resulted in more arrests, while crime in poor neighborhoods did not lead to more arrests.

“What we’re seeing is that when you stress the system, it requires more resources to arrest more people in response to crime in a wealthy area and draws police resources away from lower socioeconomic status areas,” Chattopadhyay adds in UChicago’s press release.

With this study “[we] demonstrate the importance of discovering city-specific patterns for the prediction of reported crime, which generates a fresh view on neighborhoods in the city, allows us to ask novel questions, and lets us evaluate police action in new ways,” sociologist and co-author James Evans, the Max Palevsky Professor at UChicago and the Santa Fe Institute, noted in addition in the release.

When applied to other cities—including Atlanta, Austin, Detroit, Los Angeles, Philadelphia, Portland, and San Francisco—the scientists AI program performed just as well as it did for Chicago.

Feature image: Marco Verch Professional Photographer

(Visited 5 times, 1 visits today)

Accessibility Toolbar