This algorithm can predict crimes before they happen

Prof  Ishanu Chattopadhyay at the University of Chicago claims he has developed an AI that can predict a crime before it happens, reported Jason Goodyer from BBC Science Focus.

Ishanu said that his team looked at a total of eight cities that have now started to put out crime event logs for public use.  In Chicago, he said, these logs are updated weekly.

The team used this data and digitised the city into small areas. He said that they would look at the time series of different events like homicides and property crimes. 

"This results in tens of thousands of time series that are coevolving."

The algorithm sees how time series are dependent on each other and how they shape each other. This ends up in a very complex model.

Ishanu claimed that the algorithm, based on this model, can predict a crime a week prior to it.

"Using this AI, you can say that next Wednesday, at the intersection of 37th Street and Southwestern Avenue, there would be murder."

Some people are concerned that the AI will be used to misjudge people and put them behind bars before they commit crimes.

However, that is not the case.

The algorithm simply predicts a crime at a location. It cannot inform the authorities who is going to do the crime. 

Ishanu said that if we looked closely,  murders are not very convoluted events in Chicago.  Most people, he said, die due to gang violence and it will not be rocket science to predict such an event.

The inventors of the technology do not want to use it for prediction too. They want policy optimisation. 

Previously, such algorithms have been criticised for being racist as they derive data from a judicial system that may already be biased.

Ishanu's technique however is quite straightforward and simple. They are using event logs only and not criminal profiles.