Scientific experiments, in order to be considered, successful, must be reproducible. It is the ability to replicate a predicted outcome with an actual outcome that informs a researcher that their hypothesis is potentially correct.

The latest trend in policing and crime control is “predictive policing.” This is a typically based on a series of algorithms that use predictive analytics in combination with large data sets to identify patterns of crime and other behavior which is supposed to provide clues as to where the police should focus their efforts to prevent future crimes.

According to the NYT article, they use data including “previous arrests; unemployment; an unstable home life; friends and relatives who have been killed, are in prison or have gang ties; and problems with drugs or alcohol” which suggests who is likely to be involved in criminal actions.

While some of these programs have been in place for years, it is still difficult to show that the predictive policing is a factor in changes to various crime rates. In Kansas City, Missouri, the program has been used against a backdrop of falling murders, but it is unclear if this is causation or merely correlation.

Like any mathematical model, the outcome is dependent on the assumptions used in its creation. And this leads to the issue of is it really predictive or merely confirming biases built into the queries?

The larger concern for all Americans is that this technology will only grow as larger and larger data sets become available to be mined. As our individual privacy evaporates with our anonymity, we must remain vigilant that our data mines do not turn into a coal mine and we all become the canaries.



Source: nytimes.com, “Police Program Aims to Pinpoint Those Most Likely to Commit Crimes,” John Eligon And Timothy Williams, September 24, 2015