Using software to catch criminals in the act, or to prevent them from striking in the first place: it seems like science fiction, but it?s already happening in Amsterdam. Despite significant drawbacks to the current technology, it looks like Predictive Policing will soon be a reality.
Predictive Policing involves gauging the risk of crime with the aid of software, based on massive data points that are linked together. The Amsterdam police have spent the past year testing Predictive Policing with the Crime Anticipation System (CAS). Police in the cities of Groningen, Enschede and Hoorn will join the programme this year. What are the advantages of this system, and what we can expect in the future? We spoke with Arnout de Vries of TNO.
How are the amsterdam police using predictive policing?
?To prevent ?simple?, but common crimes such as burglary and pickpocketing.?
How does cas affect a police officer?s work?
?The officers are still out on their beat, but they take a more targeted approach to their work. The system tells them when and where the crime risk is high, but it does not prescribe how the officers should conduct their work, whether that might be on-site surveillance, a wiretap or CCTV. It?s always up to the officers? superiors to decide where they should be dispatched. That decision is now partly based on a CAS risk assessment.?
How well does the system work?
?Experiences so far in Amsterdam have been overwhelmingly positive, but we do not yet have any scientifically validated results. We don?t know to what degree the CAS system has contributed to a reduction in crime.?
What kind of data do the police use in predictive policing?
?They use their own crime data, combined with other data sources. That could be anything from a calendar of events to a weather forecast. Are there lots of people out and about in the city? Will people tend to leave their windows open? The data includes the addresses of repeat offenders, because chances are high that a new crime will occur in a radius of two kilometres of these residences. The CAS system already calculates risk based on more than 100 types of data.?
The more data used, the better the prediction?
?That?s a good question. You can mine social media, for example, to get the most recent data for your risk assessment model. At TNO, however, we think that more data will eventually only result in marginal accuracy gains. The useful yield will then shrink. Moreover, there is a risk that the police themselves might lose their grip on the model: the more data you use, the more complex the calculations become. We want to avoid a situation where the police fail to understand why the system designates a particular location as high-risk.?
CAS focuses primarily on burglaries and pickpocketing. Can the system be expanded to target other types of crime?
?Such an expansion is certainly possible. But it is a complex operation, requiring a heavy reliance on data science. It is unclear whether the police have the resources and expertise needed to make it worthwhile.?
Is predictive policing more reliable than traditional police intuition?
?The software performs calculations based on the collective memory of the police force, which is far more than any individual is capable of. However, mathematical models also have inherent risks. Tunnel vision is one such risk, because the software relies on data collected in the past. If the system sends officers to a specific district, chances are high that they will find something. If the system then calculates that the risk in that district is higher than elsewhere, it will send the officers there again. This could give rise to the false impression that this particular district is more crime-ridden than neighbouring districts. The model reinforces this loop in calculation upon calculation. This is why it?s important to break the loop by randomly sending officers to other districts from time to time.?
“The software performs calculations based on the collective memory of the police force, which is far more than any individual is capable of”
Critics of predictive policing worry about unjust arrests. Is this fear justified?
?A common example is that someone might be arrested unjustly simply for walking around at night while carrying a screwdriver. The system is also susceptible to racial profiling: mechanisms that result in more frequent arrests of individuals with an ethnic background. Just imagine, for example, that the district I just mentioned is highly multicultural. You have to be aware of the nature of the data being fed into the system, and you have to know what types of data are lacking. In fact, an independent IT ethics committee should review the entire system.?
Prescriptive policing could be the next step. What does that involve?
?Predictive Policing only tells you where you should be and when. Prescriptive Policing predicts the most effective measure in that specific context. Police systems contain a wealth of information that is currently unutilized. You can use smartphones to track officers? locations and their activities on site. Were their activities effective??
Are the police ready for these developments?
?The police brass in the Netherlands are currently very focused on the professional freedom of their officers. They will be very unaccepting of software that limits this freedom, and it will be difficult to change their minds. Moreover, the police force is highly allergic to figures and statistics: the value of a cup of coffee and a chat in a community centre cannot be expressed in figures, even though it can be an immensely valuable aspect of police work. It is crucial that both officers on the beat and their superiors become aware of the added value of such a system. The mindset is far more important than the dataset.?
Is it conceivable to just ignore these technologies?
?Given the operational efficiency gains that can be achieved, it is only logical to make a strong case for predictive and prescriptive policing. The police are currently unaware of the true yield of the various potential interventions at their disposal. Both politicians and society in general expect the police to be accountable for their results and to demonstrate what they have achieved. And for their part, the police are expected to do more and more while facing cutback after cutback. This technology makes it possible to do more with less. Another key factor is that the business community has been keen to accept these new systems. We see a clear trend towards private security in more and more public spaces such as football stadiums, industrial sites and petrol stations. This could result in a situation where the police force is simply made redundant by technologies that are far more effective.?
Ten myths about predictive policing
- Criminal behaviour is unpredictable
Criminals are often creatures of habit, just like law-abiding citizens. After a successful burglary, for example, they are often inclined to try again in a similar property in the same area. Patterns like these make burglaries relatively predictable.
2. Robots will replace the police
Predictive Policing uses algorithms to help officers prevent crime. The officers will not be replaced by machines, although their role may change.
3. Crooks will have no chance with Predictive Policing
The Predictive Policing algorithm points to places on the map: the risk of crime is high there, and here and there. The best course of action is often less clear. Using big data mined from tremendous quantities of cases, the police can gain an insight into the various methods. One thing is certain: crooks are an inventive bunch.
4. Data on everyone is needed for reliable predictions
The police have been analysing official reports for years to gain insight into criminal networks. Predictive Policing does the same, but it can identify far more links between time and place. There is no need to collect data on everyone in order to predict the risk of crime in specific locations.
5. Predictive Policing is a kind of thought police that will arrest you before you?ve done anything wrong
The police hope to use Predictive Policing to prevent crime by taking pre-emptive action. That does not mean that innocent citizens will be arrested before they have done anything wrong. However, it is crucial that Predictive Policing is based on unbiased and verifiable data.
6. Predictive Policing will rid society of crime
Predictive Policing is not a catch-all solution for all types of crime. Figures and statistics cannot fully express and describe risks, safety and the intricacies of police work, which means that computer models will sometimes miss the mark. The cop on the beat will always take the crucial human approach, and crime will always be present, lurking somewhere in the dark recesses of our complex society.
7. The community police officer?s common sense always trumps an algorithm
Common sense tends to be plagued by more prejudices than software that is based on objective, statistical models. Clearly, however, those models may not suffer from unintentional bias. Predictive Policing is a valuable accessory to the officer?s common sense.
8. Predictive Policing is old wine in new bottles
Previously, the police used push pins on a paper map to identify criminal hotspots based on historical crime statistics. Predictive Policing does basically the same thing, but using digital maps to pinpoint potential future crime scenes. Also, Predictive Policing creates links between far more data points. So really, it?s new wine in old bottles.
9. Predictive Policing is plug & play
Predictive Policing seems so simple: you load the crime data into a computer and it produces a map of glowing hotspots.
From an organizational point of view, however, the implementation of Predictive Policing requires a cultural change. The police officers will have to adjust their way of thinking and working.
10. The police won?t take orders from an algorithm
Once police officers experience the added value of Predictive Policing for themselves, they will be more likely to embrace the new technology.
Sources: TNO Time