Imagine determining a criminal’s profile or that of a potential victim before a crime even occurs, or deploying law enforcement precisely where offenders will be in a few hours… It brings to mind Minority Report. In this science fiction film directed by Steven Spielberg, police officers arrest those whose future crimes have been predicted. This vision of a perfect police world, where order and security are achieved at minimal cost, is moving from fiction to reality thanks to the promises of big data.
Major IT companies like IBM and SAP, along with specialized companies such as PredPol, are offering a new service: predictive policing. This service, in the form of software with powerful algorithms, leverages vast “data wells” — an ever-growing pool of digitized information. What’s groundbreaking here is not the method but the scale: the unprecedented volume, variety, and speed of data analysis.
Documented incidents, perpetrators, victims, circumstances, environment, public lighting quality, proximity to public transport or businesses, etc., are all consolidated into a mapped summary. The system’s strength lies in its ability to detect complex causal links, where an investigator might only see clusters of incidents — just pins on a map.
Three main types of analysis guide predictions towards the kind of crime likely to occur, potential perpetrators, and probable victims. Results are projected using two primary methods. The first, Crime Hotspots, enhances understanding of crime causes by focusing on repeat incidents in the same location, facilitating targeted responses to future risks. The second model, Risk Terrain Modeling, goes further by identifying risk circumstances that may appear in places beyond known hotspots. This tool is taken seriously enough that police forces in cities like New York, Los Angeles, and Chicago have adopted it, with federal funds providing $800,000 to support some experiments. In Europe, cities like London and Munich are following suit, and the Paris police department is actively considering it, having conducted initial trials on snatch-and-grab thefts.
Despite the frequent references to big data, this objective of staying a step ahead of crime did not emerge solely from recent innovations. Predictive analysis has been widely used since 2008 by William Bratton, often considered America’s top cop. As former head of the New York and Los Angeles police and now again leading the NYPD, he was even approached in 2011 by David Cameron to head London’s Metropolitan Police. Bratton, an advocate for performance-driven policing through data, saw mass statistics as a force multiplier when combined with conventional methods, especially community policing. This point is worth reiterating, as predictive policing increasingly takes on an aura of magic, where the concept alone is seen as a guarantee of results. Beyond the fascination with technology, turnkey software providers also argue that better forecasting allows for better force allocation, potentially reducing personnel costs. When several U.S. municipalities faced budget collapses in 2011, some local officials took this to heart. In Camden, a struggling county east of Philadelphia, the police force was dissolved entirely and then partially rehired under reduced conditions. Besides a territorial reorganization, this drastic approach heavily relied on the anticipated gains of predictive policing. However, the lack of perspective means it’s too early to assess the results.
In 2013, the RAND Corporation think tank sought to clarify the debate with a publication on the role of crime forecasting in policing, dispelling some myths along the way. They emphasized that the accuracy of predictions depends more on the quality of input data than on the power of the software itself. No matter how comprehensive the synthesis, police officers still need to perform essential analysis work. Predictions remain merely forecasts, and reducing crime requires continuous field action.
Faced with these practical limitations, proponents of predictive policing might be tempted to pursue ever-more advanced technologies, such as tracking mobile connection data in real time to monitor site or area occupancy while preserving user anonymity. However, the line between public interest and infringement on freedoms may blur, especially if other aggregated data, like age or gender, are included. As the rule of law is safeguarded by making as few exceptions as possible, the accumulation of information to predictively distinguish between “normal” and “abnormal” situations demands particular caution. Predicting an offender’s profile or using “intelligent” cameras in public spaces that automatically flag actions deemed suspicious raises the implicit question of preemptive punishment. Where fiction may soon meet reality.
[1]This is precisely the central argument of a study published in 2010 by Todd Goglesong and George Bascom of the Harvard Kennedy School of Government, Making Policing More Affordable.[2]The role of crime forecasting in law enforcement operations, RAND Corporation, 2013.
