NPR recently broadcast a story asking, “Can Software That Predicts Crime Pass Constitutional Muster?” It’s a topic that’s worthy of discussion. But unfortunately, this discussion comes across as an example of what happens when people don’t have the necessary information or insight to tackle a complex issue.
The problem starts with the example featured in the story. It’s a limited application of predictive analytics in policing that only focuses on force deployment. These types of applications are not focused on proactive and preventative prosecution but on smarter deployment of police resources to proactively prevent crimes by being present. The application discussed is based on a very limited set of algorithms that only describe movement of crime waves through the city/territory. This is not different from the process police are currently using with “hot spot analysis/mapping”--where they map crimes and try to derive patterns through visual interpretation of the results. The difference in this application is that it uses the data more consistently and in the same way across all entities, minimizing the effect of prejudices and biases and other human decision making frailties and judgment errors.
IBM offers a broader solution for Crime Prediction and Prevention than the one described in the NPR story, that takes a broader set of data and algorithms to allow for decision making on the multiple levels needed in policing: strategic, tactical and operational. There's no use of personal data of individuals. In our solution, data can, for example, focus on crimes (including type of crime, location, time and modus operandi), crime triggers (such as local events, pay days and holidays), crime enablers (for instance, police presence, weather, status of streetlights and the presence of other protection measures (f.e. security cameras or private security firms)) and geo-spatial information about the location (including f.e distance to ATMs, number of empty buildings, average distance between buildings, and alarm systems).
Data relating to individuals, but not identifying those individuals, can be useful in other areas of predictive policing. For example models that predict repeat victimization that allow the police to better identify, protect and help people who are very vulnerable. This type of data may also play a role in models for predicting repeat offenders. Having better insight into the risks for re-offending allows for better offender management and can help reduce recidivism.
The IBM solution uses advanced algorithms to examine historical data and detect patterns in criminal behavior. And it uses these patterns to help predict future outcomes. In essence, it emulates the work of human experts, but does so across more data and more characteristics. It’s also more targeted than a human could could possibly be. As stated before, it then consistently applies these models to all data to and makes the results available to humans as input for their final decision. The software itself does not automatically make decisions. The use of the information in the decision making process is orchestrated by the police.
Overall this solution allows police decision making more effective and efficient. It also makes it more transparent and it makes police departments more accountable when, like our solution, it incorporates governance capabilities that allow for data security, decision trace-back, review and auditing. From a constitutional and human rights perspective it then makes sense to support the implementation of such a broad solution.
Interactive web portals