Minority Report: Crime predicting AI designed for US police departments, fails miserably

For over a decade, law enforcement agencies in the US and UK have been employing “predictive policing” software, using algorithms and AI similar to those used to forecast earthquake aftershocks, in an attempt to anticipate and prevent crimes before they occur, similar to the cult hit Tom Cruise movie, Minority Report. However, as revealed in a joint investigation conducted by The Markup and Wired, this approach appears to have fallen woefully short of its intended success. Geolitica, a predictive policing software utilized by the police department in Plainfield, New Jersey, which was the sole department among 38 to provide information for the investigation, demonstrated a staggering lack of accuracy in predicting crimes. Its success rate was less than 0.5 per cent. Previously known as PredPol, the machine learning software adopted by Plainfield and other law enforcement agencies held the promise of aiding in crime prevention. However, in practice, these prediction models, widely discussed in numerous reports, have raised ethical concerns stemming from both the inherent biases within artificial intelligence and the history of racial disparities within law enforcement. The investigation’s findings indicate that these predictive policing systems have not only failed to address these ethical issues but have also proven to be astonishingly inefficient in predicting and preventing criminal activity. To reach this conclusion, The Markup and Wired conducted a thorough analysis of 23,631 predictions made by Geolitica during the period from February to December 2018. Shockingly, the investigation revealed that fewer than 100 of these predictions corresponded with actual instances of criminal activity, resulting in a success rate of less than half a per cent. While Geolitica’s algorithm exhibited slightly better predictive performance for some types of crimes, such as correctly identifying 0.6 per cent of robberies or aggravated assaults compared to the 0.1 percent of burglaries it accurately predicted, the overall results paint a bleak picture of the tool’s effectiveness. In fact, Geolitica’s predictive abilities were so lacking that the head of Plainfield’s police department acknowledged that the software saw limited use. Captain David Guarino of the Plainfield Police Department candidly shared his perspective, stating, “Why did we get PredPol? I guess we wanted to be more effective when it came to reducing crime, and having a prediction where we should be would help us to do that. I don’t know that it did that.” He further revealed that the software was rarely utilized, if at all, and this led to the decision to discontinue its use. Guarino also raised concerns about the financial investment in Geolitica, which initially involved a $20,500 annual subscription fee and an additional $15,500 for a second yearlong extension. He suggested that these funds could have been better allocated to support community programs. Furthermore, it was reported by Wired that Geolitica is set to cease operations by the end of the year. However, the personnel from the company have already been absorbed by SoundThinking, formerly known as ShotSpotter, another law enforcement software company. Customers of the soon-to-be-defunct predictive policing platform are expected to transition to SoundThinking for their software needs.