Can Computers Predict Crimes?
Columbo would have hated the latest trend in crime-fighting. And it definitely would have made Dirty Harry even more unhinged.
But Sherlock Holmes, now he would have been impressed. The logic, the science, the compilation of data-all the stuff of Holmesian detective work.
I’m talking about something known as predictive policing-gathering loads of data and applying algorithms to deduce where and when crimes are most likely to occur. Late last month, the Los Angeles Police Department announced that it will be expanding its use of software created by a California startup named PredPol.
For the past six months, police in that city’s Foothill precinct have been following the advice of a computer and the result, according the the LAPD, is a 25 percent drop in reported burglaries in the neighborhoods to which they were directed. Now the LAPD has started using algorithm-driven policing in five more precincts covering more than 1 million people.
PredPol’s software, which previously had been tested in Santa Cruz-burglaries there dropped by 19 percent-actually evolved from a program used to predict earthquakes. Now it crunches years of crime data, particularly location and time, and refines it with what’s known about criminal behavior, such as the tendency of burglars to work the neighborhoods they know best.
Before each shift, officers are given maps marked with red boxes of likely hot spots for property crimes, in some cases zeroing in on areas as small as 500 feet wide. They’re told that whenever they’re not on calls, they should spend time in one of the boxes, preferably at least 15 minutes of every two hours. The focus is less on solving crimes, and more on preventing them by establishing a high profile in crime zones the computer has targeted.