New AI predicts crimes a week ahead

Spread the love

New AI predicts crimes a week ahead

AFP – JEAN-PHILIPPE KSIAZEK History repeats itself…

American researchers have developed a new, more sophisticated way of predicting crimes. Based on the identification of areas at risk and on the historical data of the places, the model is more “complex” than the previous ones. If it's not yet “Minority Report”, the technology boasts a 90% accuracy rate.

History repeats itself

For the first time, an artificial intelligence algorithm was able to predict crimes a week in advance, with 90% accuracy. Researchers at the University of Chicago created the model using historical crime data. The experiment was applied in eight major American cities, including Chicago, Los Angeles and Philadelphia. By integrating past crime history data into their “digital twins”(a technology usually applied to modeling smart cities), the algorithms were able to “guess” what was going to happen in the future.

A “complex environmental analysis” model

According to the researchers and authors of the study, if these predictions are strictly linked to crime patterns, they seek to move away from stereotypes and other common biases. In 2012, a study titled “Event-level prediction of urban crime reveals a signature of enforcement bias in American cities”, published in the scientific journal Nature Human Behavior, revealed that the use of historical data could be biased because of socio-economic data.

To counter these biases, the researchers in the new study, instead of applying a so-called “seismic” approach (detecting crime as a phenomenon emerging from “hot spots” and spreading to surrounding areas), used a prediction model complex social environment. “Spatial models ignore the natural topology of the city,”, explained Max Palevskym, a sociology professor at the University of Chicago involved in the study. This new model, according to the researcher, makes it possible to discover the connections between the communication networks of areas of similar socio-economic origin. “Transportation networks respect streets, sidewalks, train and bus lines. Communication networks respect areas of similar socio-economic background. Our model helps discover these connections”, he continues. Also, the algorithm does not guess the perpetrator of the crime, but the locations, which could help the police avoid making mistakes based on bias.

Similar technology is already in use in Japan, to guide citizen patrol routes in certain municipalities where crimes are statistically more likely to occur.