Algorithms are our next weapon in the fight against crime
We all tend to follow habits at work and in other aspects of our lives. Our habits usually reflect preferences, learning, or a combination of both.
Research shows that criminals are not that different from law-abiding citizens when it comes to sticking to a routine. And the police services are taking over, with the help of an increasingly widespread tool: algorithms. Because the algorithms use patterns in the data to predict future behavior, they can for example predict books on Amazon for consumers. But they can also help law enforcement in the fight against crime.
Investments in predictive analytics cause reallocation of resources and thus change the likelihood of individuals being arrested or detained. For this reason, it is important to understand whether these algorithmic tools reduce crime and whether they do so without generating bias against certain groups.
The most elaborate and well-known predictive policing software evolved from hotspot maps. These programs operate on the principle that areas recently prone to high crime are more likely to have high crime rates in the short term. Thus, law enforcement should focus on these areas to deter the greatest number of criminals.
While researchers have shown that these statistical algorithms have greater predictive power than simple averages, proving that they actually reduce crime is considerably more difficult. Police departments, however, tend to adopt predictive policing when crime is high, and subsequent reductions could reflect a natural decline that has nothing to do with this decision. Policing targeted to one area can also simply move crime elsewhere. A good evaluation therefore requires a best counterfactual scenario: what would have happened to crime without the use of predictive policing?
When it comes to biases, it’s not inconceivable that predictive policing could skew law enforcement results. More deprived areas may have higher crime rates and will be more intensively patrolled once predictive policing is introduced. If police resources remain fixed, criminals in disadvantaged areas will be more likely to encounter a police patrol than criminals in more affluent neighborhoods.
But while this is a fair outcome for the serial offenders who contributed to the spike in crime that led to additional patrols, today’s first-time offenders were not previously increasing the number of crimes. Since most predictive policing algorithms group criminal incidents without separating habitual criminals from first offenders, they may be biased against the latter in disadvantaged areas.
To help answer questions of effectiveness and bias, I evaluated predictive policing software used in Milan, Italy. This allowed me to establish a good counterfactual: for historical reasons, Milan has two police departments that share the same goals, but only one of them uses predictive policing.
KeyCrime, the predictive software developed by Mario Venturi and used in Milan, differs from common predictive policing tools because it focuses on apprehending perpetrators rather than deterring them and distinguishes first-time offenders from repeat offenders. The software uses information gathered from victim reports and CCTV cameras to link criminals to commercial robberies, then predicts when and where a particular individual or group will strike next. KeyCrime generates individual predictions, which reduces the risk of bias.
The results indicate that analyzing the habits of repeat criminals more than doubles the likelihood of arresting them. Thieves tend to act the same way over time, targeting a specific neighborhood and type of business, as well as at a certain time of day. So, for example, someone who has already robbed a jewelry store at 9 a.m. is likely to do it again in the same neighborhood, at around the same time, and against another jeweler. Since there are only a limited number of matches that match the predictions, the software highlights potential future targets and the police department organizes patrols to catch the thief.
Micro-predictions based on the behavior of individual criminal groups have proven effective in combating thefts and are now being extended to other types of serial offenders, such as sex offenders and terrorists. It remains to be seen whether predictive policing will also be successful in bringing these criminals to justice, as reduced interaction with victims and the availability of CCTV footage could make it more difficult to link offenders to incidents.
More widespread use of predictive policing may cause criminals to change their ways and become less predictable. But the development of more powerful data collection algorithms and processes is causing optimism among police departments.©2022 Syndicate Project