How effective is predictive policing? A web conference organised by Efus’ Security & Innovation working group

Also in... (Français)

Paris, France, July 2020 – In the framework of its working group on Security & Innovation, Efus launched a new series of web conferences on new technologies in urban security to promote the exchange of experiences amongst peers. Held on 25 June, the web conference on predictive policing offered participants the opportunity to hear from experts in the field and discuss the effectiveness and impacts of this relatively novel policing approach in EU countries.

Predictive policing: forecasting the probability of crime

Policing has been intelligence-led before the advent of artificial intelligence and big data. However, developments in computing power and the digitisation of information have greatly facilitated access to large amounts of data. This enables the police to speed up the process of collecting and analysing data and develop predictive policing methods. Using data and statistical methods allows to “forecast” the probability of crime in two target areas: potential crime locations and individuals that might become involved in criminal activities, either as victims or perpetrators. 

In the framework of the Cutting Crime Impact (CCI) European project, researcher Maximilian Querbach from the State Office for Criminal Investigation of Lower Saxony (Germany), reviewed the state of the art of predictive policing in EU countries. The Netherlands rolled out their Crime Anticipation System (CAS), which identifies hotspots within a territory, in 2017. The data are sourced from multiple databases, such as the Dutch central crime database, and include information on previous crimes and known offenders. The state of Lower Saxony opted for a predictive policing approach in response to a spike in domestic burglaries. The PreMAP in-house developed software uses a “near-repeat” approach to identify predictive factors such as the type of stolen goods and the modus operandi of the burglaries. Other German states use the PreCops software which is also based on the near-repeat approach. Additionally to historical crime data this software draws from findings in the domains of criminology and psychology. 

The ethical, legal and social aspects of predictive policing

At a time when a series of events all over the world have triggered an international debate about policing methods, in particular concerning alleged racism and the inappropriate use of force, it is important to understand how ethical concerns are addressed. Dr Oskar Gstrein from the University of Groningen (Netherlands), who is a contributor to the CCI project,  analysed the ethical, legal and social aspects of predictive policing and highlighted issues related to data selection and machine bias, transparency and accountability and the stigmatisation of certain communities and neighbourhoods. Involving these groups in the process and systematically tackling questions related to data selection processes and the fairness of impact assessments could mitigate stigmatisation and increase police accountability.

Günter Okon and Kira Langanki from the German Institute for Pattern-Based Forecasting (IfmPt), which develops PreCOPS, shared some lessons learned. They highlighted that police officers need repeated training to understand how the technology works and properly evaluate and respond to the findings. Officers sometimes doubt the efficiency of the approach and don’t necessarily have the resources or time to develop innovative responses to the forecasts. It is important to understand the expectations of a police service in order to develop a software that meets such expectations. This was echoed by Dr Gstein who highlighted that police must also understand how a system works in terms of data protection. The speakers agreed that the predictive policing approach needs to be implemented within existing systems in order to supplement existing techniques. While the evaluation of predictive policing methods has shown limited effectiveness, it can be particularly interesting as a tool to improve internal management and communication within police departments. 

Some takeaways:

  • Developments in the digitisation of information make it easier for  law enforcement agencies to collect and analyse data and thus “forecast” the probability of crime. 
  • Despite the potential of predictive policing, in some cases police officers question its effectiveness for crime prevention purposes. 
  • Applying predictive policing methods requires that officers be properly trained in operating the relevant softwares and understanding their purpose and benefits. Ultimately, officers are the ones who evaluate this approach and choose whether and how to respond to the findings.  
  • Understanding the technology behind predictive policing and discussing its effectiveness allows us to go beyond easy narratives – either exceedingly optimistic or pessimistic – and enter fruitful conversations about ethical, legal and social concerns.
  • The technology can also present an opportunity to confront bias in historical police data and question how these were collected and selected. If used correctly and in cooperation with civil society, it can increase police accountability.