The strategy, known as predictive policing, combines elements of traditional policing, like increased attention to crime “hot spots” and close monitoring of recent parolees. But it often also uses other data, including information about friendships, social media activity and drug use, to identify “hot people” and aid the authorities in forecasting crime.

Source: Police Program Aims to Pinpoint Those Most Likely to Commit Crimes

This is very worrying. Reading the article it seems they are handling it well – they are inviting potential purpetrators in and explaining what’s going on, hoping to shock them. If a crime is committed, everyone in the predictive chain is picked up and they sling the book at them for everything they can find. Fair enough, they shouldn’t have been breaking the law anyway and if they get picked up for it because they were in an associative chain is just as good as if they get picked up due to any other reason.

However, if you are friends with a criminal, you may get invited to the courts again and again and again, even if you did nothing wrong yourself – the same problem no-fly lists have: false positives. Another thing is that you need to troll through huge amounts of personal data in order to get these predictive models to work. This means that people and organisations could (in practice shows they do!) misuse their access to your personal data.

The article has some figures on how well this does compared to traditional policing and other predictive models, but the jury is still out on that really. It needs longer and more testing.