New beat: LA police are issued predictive policing maps Photo: Getty Images
The morning watch at the LosAngeles Police Department's Foothill division begins at 6am, when the shift's officers gather for roll-call and a briefing.
The dozen or so officers file into the roll-call room and find space on the benches. They are briefed on incidents from the previous shift, and on any offenders or vehicles they might be looking for. Then, before they hit the streets, each team is a handed a simple A4 map, warm from an office printer.
It doesn't look like much, but the maps are the result of more than half a decade's work by anthropologists and pure and applied mathematicians from two leading American universities, funded by supporters including the US Army and Navy.
LA crime hotspots.
Marked on the maps are a series of little red boxes outlining 150-metre-long squares superimposed over the streets they will patrol through the shift. The boxes indicate, with spooky accuracy, where the division's most common crimes are most likely to be committed during the 12-hour shift.
The Predictive Policing project, or Pred Pol, as it has become known, has been so successful in cutting crime rates since it was tested in Foothill in 2011 that it has spread to seven other LA police divisions, with more to come. Another 20 or so police departments across the US have signed on, as have departments in Britain.
It is understood at least one Australian force is in talks to sign up, though its creators decline to discuss potential customers.
One of the driving forces behind Pred Pol is Dr Jeff Brantingham, an anthropologist at the University of California, Los Angeles (UCLA), who struggled bravely to find simple-enough language to describe the science behind the technology.
Crime prediction is not a new field. All street cops know the crime hot spots on their beat, and larger police forces have been employing crime analysts and software to assist them for years. It has been successful, too. The standard measure of accuracy for crime prediction is multiplications of random. Current techniques are usually effective about three times random. Pred Pol is twice as effective, accurate to six times random. And it is likely to improve.
Brantingham, a former archaeologist who never lost the interest in crime ingrained in him by his criminologist parents, says the key insight that informed early work on Pred Pol was to tackle crime prediction via mathematics, rather than the social sciences.
Usually, crime analysts look at crime patterns combined with social and geographical factors. They would look at a particular area's crime history, demographic make-up, the location of transport and late-opening bars that might attract potential crooks – mostly young men – and victims.
Pred Pol strips all that away. It simply feeds three different factors – crime type, time and location – through the dizzyingly sophisticated equations and algorithms the team has created.
''The problem with considering [things such as] how many open bars there are is that it misses the point,'' Brantingham says. ''The key information is the number of assaults or robberies, and you already know that.''
The team began in 2005 by considering crime through a spatial rather than social lens. If you consider a potential criminal and apotential target to be, for example, two atoms floating in a body of fluid, what factors would cause them to meet – and how and when? And, later, what other factors might interrupt that meeting?
Working with Brantingham, UCLA's applied mathematics department began investigating models that described events such as chemical reactions and the way substances diffuse in fluid. Even fish migration predictions came up in thediscussion.
Soon the LAPD had become involved. Brantingham had appealed to the department for access to crime data, and had feared he might be met with old-guard suspicion. Instead, the department threw itself into the project.
At the time, the LAPD suffered from an awkward problem: crime was falling relentlessly across the city and had been for a few years. The city that had registered 1092 murders in a single year two decades ago lost just 298 of its citizens to violence in 2012, when it became the safest big city in the US.
This was good news for citizens, but since the only real measure of a police force's success is lower year-on-year crime rates, it meant innovation was needed to keep up the momentum. As one officer puts it, all the low-hanging fruit had gone.
The then LAPD chief, William Bratton, not only supplied the UCLA team with any data they asked for, but also assigned one of his captains, Sean Malinowski, to work with them. He even kicked in extra funding.
''Whenever anyone starts talking about data, I tend to get the mission,'' Malinowski says. This is probably because the 48-year-old cop has a PhD in public policy. Data and statistics don't scare him and he can speak basic academic.
The next breakthrough came when Brantingham's collaborator, Dr George Mohler, now an assistant professor of mathematics and computer science at Santa Clara University, discovered that equations used in mapping the aftershocks of earthquakes could be manipulated to map crime. Mohler's discovery had the dramatic effect of allowing the team to quickly apply their theory and create a working model of Pred Pol. With hindsight, Brantingham says, the link between seismology and crime makes sense. Research already showed how criminal events caused their own aftershocks, such as when burglars returned to homes or streets where they had already had success.
Before each 12-hour shift, or watch, begins in Foothill, a senior officer logs in to Pred Pol with a password and calls up the maps for the coming hours on a desktop computer.
Software runs up-to-date and historical crime data through Pred Pol's algorithms and spits out that shift's crime ''boxes'', which are all printed neatly on a Google map.
The operator can also pull up historical crime records going back hours or days, weeks or months.
So far the system in use predicts and maps only car theft, burglary from a car, and burglary, because these constitute more than 60 per cent of crime.
The team is working to expand into gun crime and homicide.
In the long run, it is feasible that the technology could allow police to enter, say, the time and location suspects are seen on the street and instantly be fed background material, such as what their gang affiliation is likely to be, and how that gang fits into the broader criminal ecology of the area.
The technology is already so sophisticated that police on patrol could be given their own logins and pull up even more detailed maps via smartphones or tablets. The boxes could be ranked for risk and police could even watch them move slowly across the map during the watch. As police attention focused on particular boxes, they could change colour as the risk of crime abated with police presence.
So far, the developers believe that could lead to information overload. Police are basically hunters, Brantingham says. Should the boxes be ranked, they would focus on the top tiers.
Besides, says Andrea Bertozzi, UCLA's head of applied maths and the lead mathematician on Pred Pol, the folded piece of paper is good technology for its purpose.
When police on patrol have time to spare, they are directed to spend time in their allocated boxes. The beauty is the simplicity. They are not distracted by interacting with new technology or equipment.
Police report that once in their boxes they find themselves ''hyper-alert'', Malinowski says. They take note of who is around, they talk to shopkeepers and home owners, they check on suspicious activity. Their goal in the boxes is not so much to catch crooks – though that has happened – but to suppress crime.
Malinowski dismisses early criticism that the program might simply diffuse crime, saying new research showed that the type of crooks they were after, such as car thieves, were opportunistic, and often give up in the face of risk. ''We are not talking about committed criminals like a serial rapist,'' he says.
As Brantingham puts it: ''It is not too hard to convince them back home to their Xboxes.''
The American Criminal Law Review has raised the concern that the program could warp crime statistics, either by increasing the arrest rate in the boxes through extra policing, or falsely reducing it through diffusion.
Civil-rights groups are taking the former concern seriously because designating an area a crime hot spot can be used as a factor in formulating ''reasonable suspicion'' for stopping a suspect.
Malinowski says so far there is no evidence of an increase of civil-rights complaints, and the statistics clearly show crime is decreasing.
In a six-month test in Foothill when the division used Pred Pol strictly – that is, Pred Pol alone was used for crime prediction, rather than in concert with the department's analysts – crime rates dropped by 12 per cent overall and 25per cent for car theft. In a later six-month period, when other intelligence types were used, crime rates rose, he says.
The results are so dramatic that the force's chief, Charlie Beck, is backing further research, not only to refine the technology but to push it into new crime areas. In effect, Malinowski says, Beck has so much confidence in the model, he is allowing the researchers to use parts of the city as a laboratory to study crime prediction. And his confidence is mirrored around the country and the world as other police forces adopt the technology.
Because the system uses such basic data, the technology is easily transferable and blind to cultural difference, Brantingham says. Once the team has access to a police department's computer system, they can be online within a couple of weeks, he says.
The research has clearly made a splash at UCLA, which is enjoying not only new sources of research funding but also the flow of Pred Pol subscription fees.
In February, Bertozzi was awarded a new named chair, one of the university's highest honours.
UCLA cited her work on the unique mathematics behind Pred Pol as one of her the department's key successes.