A new algorithm can tell quite nicely where and when a crime will take place, and can show how proportionally the police react to crime.

What was once a subject for a science fiction movie, is fast becoming a reality: a computer that predicts where and when a crime will take place. data scientist Ishanu Chattopadhyay from the University of Chicago and colleagues now have a designed algorithm that is pretty good at this†

And you can do more with that than police to an upcoming one crime scene steer, the researchers write. You can also use it to determine whether the police divide their resources fairly between rich and poor neighborhoods.

Hotspots and earthquakes

Now Chattopadhyay and his team’s algorithm is anything but the first attempt at developing a crime prediction program. How does this algorithm differ from its predecessors?

“Existing approaches often assume that crime is driven by so-called hotspots”, explains Chattopadhyay. “You assume that crime spreads over an area like a drop of ink in water. But that has not yielded much so far.”

Other approaches try to find similarities between crime and earthquakes, the data scientist continues, with before and aftershocks. “Yet other researchers have thrown every imaginable machine learning technique on the issue, with factors ranging from the weather to how much graffiti is on the walls.”

Some of those approaches do a better job on their own than purely guessing where and when a crime will take place, Chattopadhyay says. However, according to him, it is often of little use in practice. “Events are not predicted long enough in advance, or the location not precise enough, to act on them.”

Hidden connections

The new algorithm has a completely different approach, according to Chattopadhyay. “We don’t assume that crime spreads like ink in water, but recognize that events in different places in the city can influence and limit each other. We get these hidden connections from reports of past events: what happened where and when? We thus determine patterns from the past that indicate increased crime. This allows us to make accurate predictions for the future.”

How accurate? There’s some variation for the eight US cities the team looked at, but on average, the algorithm can predict with 90 percent certainty where a crime might take place in a week.

‘AI is here to stay’

With such scores, police officers will undoubtedly be eager to get started with the algorithm. But there is another side to the story. In their article, Chattopadhyay and his team write that their model also shows that the police are especially large when crimes are committed in wealthier neighborhoods. This would be at the expense of the resources available to the poorer neighbourhoods. In other words, the program can also reveal what the police themselves could have done better.

Chattopadhyay sees “turning around” the algorithm as a small step towards democratizing artificial intelligence (AI). In other words: see to it that these kinds of techniques are used for instead of against the population. “AI is here to stay and will only have more of an impact on our daily lives,” says the researcher. “We must ensure that this technological revolution does not lead to a society in which no one would want to live.”