PREDICTIVE POL ICING
CAN WE PREDICT CRIMES BEFORE THEY HAPPEN?
It’s 4: 30am on a Friday morning in August and there’s a heavy police presence in a quiet London suburb. It’s a respectable, leafy area and right now, nothing is happening. In fact, it’s been quiet for the past few days. But the officers are on high alert. They’ve been sent at the say-so of a computer that’s calculated, on the basis of the data fed into it, that a wave of break-ins is highly likely within the next 24 hours. In other words, they’re policing crimes that they think will happen, rather than ones that
have happened. This is predictive policing. And it’s about to get much, much more sophisticated.
The idea of predicting where crimes will take place isn’t new. For decades now, police forces in the UK and US have been creating ‘hotspot’ maps that identify the areas where most incidents are taking place, and then sending more police officers to those areas. Predictive policing takes this to the next level, crunching big data using algorithms based on those that help to predict when and where the next earthquake aftershock will be, or how a disease will spread.
These algorithms generate information that police officers can act on, and it seems to work. In tests, their predictive powers appear to outperform the more traditional techniques used by crime analysts. Their successes have led to predictive policing being adopted by several US police departments, such as California and Arizona, as well as Kent Police in the UK.
But not everyone’s convinced about predictive policing – or how it’s implemented at least. Among them is criminologist Prof John Eck at the University of Cincinnati. His problem isn’t so much with the predictive policing software itself, but the idea of sending out large numbers of staff to patrol problems highlighted by the algorithms. “Why would you want to keep sending large amounts of expensive public servants to these locations?” he says. “Instead, we should be asking why this location has a persistent crime problem, and what we can do to keep it from happening.” Eck would prefer it if the police encouraged owners of businesses and other properties highlighted as crime hotspots to step in and make changes, such as shops with high shoplifting rates repositioning displays. Critics have raised other concerns too, such as the possibility of crimes simply shifting to other locations when problem areas are targeted by the police.
But predictive policing is becoming more and more widespread, and it could be about to change radically. Earlier this year, a bunch of mathematicians led by Prof Mark Girolami at Imperial College London were awarded £3m from the government to take predictive policing to the next level. Whereas today’s tools just rely on crime data – such as the locations, dates and times of incidents – Girolami and his team will be working on how to integrate the likes of Twitter feeds, newspaper reports and socioeconomic data to sharpen the predictions. Text documents will be converted, or ‘coded’, into numerical representations, with counts of words and phrases – such as descriptions of assaults or break-ins – to highlight geographical areas of concern. “All of these streams of information will be coded and integrated using our ‘secret sauce’,” says Girolami, referring to the complex maths that will draw all of this disparate data together.
What’s more, this new predictive tool aims to work out the extent to which crime will be displaced to a neighbouring area when the number of police in the original area suddenly shoots up. “Our models will be able to propagate what would happen,” says Girolami.