ROME - Steven Spielberg’s Minority Report based on 1956 Philip Dick’s namesake book imagined that in 2054 Washington, DC’s pre-crime police department would predict crimes before they were concretely committed and, accordingly, arrest the criminals. Today this is no longer fiction: with the accelerating uptake of increasingly capable and powerful neural networks for machine learning (ML) and artificial intelligence (AI), so-called ‘pre-crime’ systems are now possible.

A team from the University of Chicago has written an AI model able to predict crimes, at different times and in different parts of the city, one week before they occur, and with 90 percent accuracy. Data analysts and researchers feed the algorithm with two categories of historical data: series of crimes committed in Chicago between 2014 and 2016, and the most serious crimes, such as homicides, assaults, and beatings, and crimes against property, like armed robbery, theft, or burglary. This particular data is used intentionally. According to the researchers, there is also the greatest chance that crimes in the categories described above are reported to the police in parts of the city where residents do not trust or cooperate with law enforcement. 

So far, this model has been tested on eight United States cities to create a sort of digital copy of an urban environment. As explained by Ishanu Chattopadhyay, one of the senior authors of this research, if the AI is fed past data, it is likely to predict what will happen in the future.¹ However, unlike Dick’s short story, the algorithm will not cause a criminal to be arrested even before a crime has been committed, though it can help enforcement agencies increase the level of security in the city.

Until now, when trying to predict crimes researchers used an epidemic model, according to which crimes were supposed to first appear in various hot spots and then spread throughout the community. Thi