How Security Agencies Use AI to Stop Crime
Modelingposted by Caspar Wylie, ODSC October 10, 2018 Caspar Wylie, ODSC
Crime is a social problem. It is therefore not the kind you might expect machine learning to solve, or even improve. Unlike the many problems that machine learning has already cracked, social problems tend to include more anomalies, inconsistencies, and unexplained results. However, most crime — from the perpetrator’s perspective — is planned and evaluated to some extent. That planning contains the logic which gives rise to the patterns that inform machine learning models. We’ll look at how agencies use AI to stop crime like this.
For instance, a criminal may choose an optimal time and location for a specific crime. His or her own demographics including age, sex, substance abuse habits, education, and criminal record will also come into play. When observed on a larger scale, crime has an amazing amount of consistency.
The CIA works closely with companies like Cylance and Palantir Technologies, both of which are based entirely on understanding and predicting crime through machine learning. Many people have circulated claims that the CIA uses these methods to predict terror threats, big time cyber attacks, and more. But the truth is — not surprisingly — no one really knows the specifics or extent of machine learning in the agency.
On a more localized level, more than 50 police departments in the United States use PredPol, a company famous for predicting crime. Using three data points — crime type, location, and date/time — PredPol’s software provides law enforcement agencies with customized predictions of where and when crimes are likely to occur, narrowed down to 500-square-foot areas. The program automatically displays these locations on maps generated for each working shift of each day, which helps inform each officer’s operations.
Unfortunately, the reality of these automated tools is slightly different than how they are portrayed in movies. There are some key limitations, which are understandably difficult to surpass.
First, PredPol’s predictions do not include who will commit a crime, only where it is likely to occur. Second, if the crime in question is more serious, 500 square feet may not be precise enough to save a victim. But it is definitely better than nothing, and the results are impressive.
PredPol is meeting its big picture goal: Not only does it predict 30 percent more accurately than specialists whose jobs that used to be, but the software has actually made a difference in decreasing crime. Certain departments using the tool have seen a decrease in actual crime to the tune of 32 percent, with an average figure of about 14 percent. This is in part because police have been more successful in finding and catching criminals.
The quality of the data required to learn and predict crime is vital. It is generally difficult to define what data is relevant to a specific social problem, but it is crucial to supply as much information as possible without overloading the system with completely irrelevant data.
These days, police collect so much data in so many different places and forms that people question their privacy. Over the next fifteen years, it is said that this data will all form part of the prediction of crime. Although more powerful and complex machine learning methods have not yet been properly implemented within the industry, there is no doubt that they will be part of its future. Crime will become increasingly predictable as a result.
Ready to learn more data science skills and techniques in-person? Register for ODSC West this October 31 – November 3 now and hear from world-renowned names in data science and artificial intelligence!