What (the bleep) is predictive policing, anyway?
At its root, predictive policing is a mega-extension of the intuition that all good cops develop over time
Making the papers for some time now is a technology called “predictive policing.” At its root, it appears to be a way to forecast what and where crimes are likely to occur so that personnel can be re-allocated or to deploy other preventative methods. It’s controversial, though.
Critics fear a "Minority Report" scenario in which people are arrested for “pre-crimes” — some extension of racial, ethnic, or other profiling techniques.
So, what is predictive policing?
Good Old Officer Intuition
At its root, predictive policing is a mega-extension of the intuition that all good cops develop over time. As a police officer learns their job and the community where they perform it, they come to know when certain events are likely to happen.
Around high school graduation dates, there will be teen parties with heavy use of alcohol — in some cases hosted by well-intentioned parents unfamiliar with the concept of “contributing to the delinquency of a minor.”
If a city is home to a major professional or college sports team, a win or a loss at the national championship game will likely result in a street riot led by rabid fans either celebrating or protesting the outcome.
These are single-event “if X, then Y” equations.
Because few cops have an overview of all events sufficient to make other cause-and-effect outcomes, it’s harder to come up with more complex predictors. Bad weather cuts down on most crimes requiring the bad guy to be outside, because even criminals dislike being wet, cold and/or launched airborne by hurricane-force winds.
Good weather has the opposite effect, but at some point hot weather and possibly high humidity make tempers short and influence the potential to act violently.
Where are the tipping points for weather trends, so that they deter one kind of conduct but promote another?
Some major crimes are predictable from the taking place of more minor acts. Car thieves tend to favor one or a few makes and models, as they know how to bypass their locks and starter mechanisms.
Suppose the car thieves associated with a particular street gang favor Toyota Camry automobiles for their drive-by shootings. The theft of a Camry in one part of the city conveniently located to the gang may portend a drive-by in the territory of a rival gang, especially if the rival has recently done some actual or perceived wrong to the competition.
The gang investigator may not be plugged into auto theft report activity, and will miss this indicator unless he knows to look for it.
Really, Really Big Numbers
There can be a staggering number of telltale indicator factors in play. Public utility service starts and stops, a change in emergency room admissions for overdoses, public bus service ridership, or an increase in the number of people lining up at the homeless shelter can all be telltales of some potential disaster on the horizon.
Human monitors first have to know what to look for, then connect the many dots to make an accurate prediction of some event of interest to the police.
Rarely will someone see an emergent pattern and call it — this job is much-better suited for a computer.
Say there are 25 trend indicators being tracked: air temperature, humidity, traffic on a particular street, school attendance, etc.
Comparing these two at a time results in 600 (25 x 24) possible outcomes.
Comparing all possible combinations with all other possible combinations results in 15511210043330985984000000, or about 1.5 x 1025 outcomes to consider.
That will cause you to miss lunch, and those outcomes might not mean anything at all.
Of course, there will be far more than 25 factors, and some change the formula entirely. Say that your booking and probation records contain the cell phone numbers of your local gangbangers, and an unusually detailed FI card turned in by a patrol grunt indicates that cell numbers from members of two or more rival gangs turned up in the same car on a traffic stop.
Interesting intel, to be sure, but:
1.) Would your gang investigator even be tipped to this situation, and
2.) What does it even indicate?
By comparing past events with preceding circumstances that may or may not have had an influence, predictive policing systems can alert the police of the probability that a similar event is about to take place.
Take a look at today’s related column, where I talk about some real-world applications of predictive policing and how it’s being used to make police resources more effective.