In the 2002 film Minority Report, police apprehend criminals based on the predictions of three psychics. Although the story is science fiction, the potential for law enforcement to predict and prevent a crime before it takes place is not. Because of technology—including smartphones, surveillance cameras, and biometric sensors that can detect markers like fingerprints—more data about individuals is available now than ever before. And analyzing that data can lead law enforcement to crimes before they occur.
From the pushpins police used in the early 1900s to mark where street crimes occurred, in order to allocate foot patrols, law enforcement has moved to computer programs that analyze data to spot areas where crimes are likely to happen. And although those programs can detect criminal activity, there’s still some way to go, says IEEE Member Marc Goodman, founder of the Future Crimes Institute, a group of technical specialists who consult with law enforcement officials on technology’s role in crime and its prevention.
Goodman, a former police officer, has also served as a senior advisor to Interpol, the international police organization based in Lyon, France, whose 190 member countries work together to fight crime.
“Data often replicates what a police officer already knows—for example, that more crime takes place on Friday nights, when people go out, or in places where illegal drugs are sold,” says Goodman. “Data analysis will be more useful when it can reveal more complex information that police officials might not be able to figure out on their own.”
Big-data analysis programs—which can massage the data gathered from so many places in today’s records- and sensor-filled world—may be the answer. Such programs are already being used to complement law enforcement practices.
One company using data to make predictions is Palantir, in Palo Alto, Calif. It designed a program used by the U.S. intelligence community to prevent acts of terrorism. By examining the vast amount of information already available on terrorism suspects, the program can piece together data to connect the dots and indicate what might happen.
Data can include a suspect’s DNA, facial information gleaned from the surveillance of automated teller machines used to wire money, rental-car license plates monitored at different locations, phone records, and places that the suspect is known to have visited. The program has uncovered terrorist networks planning bomb attacks in several countries and, in one case, found suspects in the murder of a U.S. Customs agent.
Forensic Logic, in Walnut Creek, Calif., is another data-analysis company working to help prevent crime. Goodman notes that many contiguous police precincts do not share information with one another.
In one project, the company combined the databases of some 80 cities and towns within Los Angeles County and analyzed the results. It was able to quickly locate several fugitives, simply because they had moved from one police precinct to the next.
“Projects like these are a powerful tool for law enforcement,” Goodman says.
Social networks have also been useful, providing a vast amount of public information for police to comb through. Software can scan for specific keywords and behaviors that could indicate unlawful activity. Programs have not only uncovered plots to commit crimes, such as robberies and drug deals, but also pinpointed those who might commit them, as well as when and where the crimes might take place.
An impressive example of a data project on crime is at the University of Pennsylvania, in Philadelphia, notes Goodman. A team in its Department of Criminology came up with an algorithm to predict who will be a victim of a homicide based on a variety of data, including reports from local police precincts. Rather than targeting the likely murderer, the researchers partner with police to warn potential victims that they are at risk and advise them on how to protect themselves.
The department previously developed software to help parole boards determine which inmates could be released because they’re unlikely to commit a crime again. These predictions are based on 24 variables, including criminal records and the ages at which crimes were committed. About 80 percent of U.S. parole boards now use similar systems, which have been shown to drop the recidivism rate by 15 percent.
Big-data programs have proven best at predicting street crimes, such as auto thefts and homicides during the commission of a felony, as well as street riots and acts of terrorism, says Goodman. Other offenses, such as cybercrime or so-called crimes of passion, are not likely to be uncovered in advance by a data program. “I don’t see how an algorithm can predict them just yet,” he says.
The biggest obstacle to using big data in predicting criminal activity is that programmers and law enforcement are not joining forces. Another challenge is to determine what to do once the data indicate that someone might be up to no good. Prosecutors could ask a judge to place someone under house arrest or issue a restraining order, if enough evidence is there. But arresting someone based on data analytics could be trickier, Goodman says.
“Data doesn’t always show the whole picture,” he explains. “And software programs are not always neutral. There is concern about how their algorithms are implemented.
“As big-data programs and the technologies that provide data advance, there is no doubt that law enforcement will use these tools to help them do their jobs,” Goodman continues. “But before we can achieve that theoretical Minority Report world, the programs need to improve, and questions about their effects on privacy and the appropriate use of them need to be answered.”