If police could predict when and where crimes are likely to occur, they could deploy their resources to maximal efficiency and be in the right place at the right time to stop crime. If they could predict who is likely to commit a crime, they could perhaps intervene with social services to prevent crime.

This is the hope of predictive policing, which seeks to leverage big data and powerful computer algorithms to aid police. This is often compared to the move Minority Report, but that is actually a bad analogy. In the movie the crime unit used three psychics to predict exactly who would commit a specific crime at a specific place and time. That is nothing like predictive policing.

If you insist on using a science fiction analogy, the series Person of Interest is much closer. In that series the government possess a powerful computer program that is fed all of the surveillance data in the country and uses it to predict when a crime is about to happen. It then spits out a social security number of someone who is about to become either a victim or a perpetrator, the “person of interest,” and our heroes have to figure out and stop the crime. It’s a good narrative device, but not very realistic (at least not anytime soon). 

What Is Predictive Policing

Predictive policing takes a few potential forms. The simplest version of this is to use a computer algorithm to analyze past crimes to create a predictive model of when and where crimes are likely to occur in the future. The program essentially identifies hot spots so that police resources can be deployed when and where they are needed.

These models do not predict specific crimes. This is like the difference between climate and weather. The model predicts climate, but no one can predict the weather (not beyond a few days) because there are simply too many variables.

Hot spot policing is already a thing, of course. Police know where the crime is and tend to hang out there. Predictive policing just hopes to make it better. Two questions are immediately raised, however: does it work, and are there any unintended consequences.

Research so far has not been kind to predictive policing. However, existing research is mostly preliminary and of questionable value, but it has not shown big effects from predictive policing. At best it is an incremental advance over traditional methods.

The National Institute of Justice, however, is funding more rigorous research into specific predictive policing programs. This is actually a complex question and will likely take a lot of thoughtful research to sort out.

The incremental advantage is likely to be just using big data to make hot spot mapping more accurate and up to date. For example, such data indicates that when one home is robbed, the other homes in that neighborhood are at higher risk over the next two weeks of also being robbed. A predictive policing program could ensure that this data is utilized to patrol at-risk neighborhoods.

The research, in fact, is most promising for reducing property crime. In some cases, utilizing predictive policing software reduced crime by up to 12% compared to nearby districts that did not use the software.

Figuring out unintended consequences is much trickier, and many concerns have been raised. What if the data being fed into the algorithms is biased? Sending police to alleged hot spots then becomes self-fulfilling, reinforcing the predictions. Police will find crime where they look, and those crimes are fed into the algorithm predicting more crime in that area, which of course is found.

There is a great deal of concern about racial bias in policing, and predictive policing might reinforce those racial biases but give it the respectability of science.

A more subtle unintended consequence is creating the illusion of more predictive power than really occurs. It is always tempting to take the short cut, to consult the oracle, rather than do the hard work of deep thinking about a problem. This is something, for example, that we have to teach medical students – don’t rely on the high-tech tests, you still have to think about patients in a thorough way.

Police have to make sure that predictive policing does not crowd out more traditional methods that rely on human pattern recognition, understanding the neighborhood, and experience. These models should augment other methods, providing another layer of information. More experience cops are likely to see this, but younger cops who have yet to develop a lot of experience might find it tempting to just following the algorithm. If they spend their career with predictive policing they may not develop the same level of experience and instincts.

This dilemma is a generic one that exists in any profession trying to incorporate big data into their practice, such as the example of medicine I already gave. To be clear, I am not championing instinct at the expense of data. I am pointing out the need to thoughtfully incorporate both – using big data to inform experience and thoughtful investigation, and avoiding the trap of substituting a predigested answer for careful thought.

The second type of predictive policing, trying to identify people at risk of either being victims or perpetrators of crime, is basically the same as Person of Interest, just orders of magnitude less detailed. The algorithms are using social media, social networks, and other demographic data to predict that someone is statistically at higher risk of crime. This could then be used to offer social services to that person or family.

For now, this seems like a pipe dream. We just don’t have the data or the algorithms for this to have any significant accuracy. And again, keep in mind that we have to evaluate such programs against other options. Applying predictive policing to individuals also raises serious concerns about privacy.

Conclusion

No one is saying that predictive policing should replace other policing programs. Police still need to form relationships with communities, address problems of implicit bias, and use other methods to determine how to best deploy their resources.

Right now predictive policing is in it’s infancy. At best it provides an incremental advantage in predicting future crime risk. If utilized properly, at least in some cases, it can reduce crime, although certainly more research is needed.

It comes with pitfalls, however, such as seeming like an easy fix or giving the illusion that it is more powerful than it is. Predictive policing may also be self-fulfilling and reinforce existing biases.

Still, it may evolve into a useful tool. It will never be a panacea, and as long as it is thoughtfully incorporated as one of many tools, with its own strengths and weaknesses, it can be a net positive.

It does also provoke speculation about how our world is changing. Person of Interest may not be that far off, if we continue to increase state surveillance of private citizens.