The lightning-fast progress of artificial intelligence and machine learning have gained the attention of governments. The target? Pure dystopia: developing predictive technologies to monitor and possibly prevent crimes and criminal behavior. It must be said: the first attempts of application have been quite opaque, between "digital racism" and prejudices.
The University of Chicago's Department of Social and Data Sciences has created a new algorithm that analyzes time series and geographic patterns in public data on violent and property crime to anticipate crime. The algorithm has proven to be correct about 90% of the time in predicting future criminal activities a week in advance.
Is 90% accuracy enough?
Before answering this question (rhetoric, of which you already know the answer), some further data. In a separate model, the research team also studied police response to crime, analyzing the number of arrests and comparing it in neighborhoods with different socioeconomic status. The researchers observed that crimes in affluent areas led to more arrests, while arrests in deprived neighborhoods decreased. This underlies a lack of police engagement in the poorest areas of the city.
Ishanu Chattopadhyay is assistant professor at the University of Chicago and senior author of the new study, published in the journal Nature Human Behavior (I link it here).
How the “anti-crime” algorithm works
The new tool was tested and proven effective on two types of events reported by the City of Chicago: violent crimes (homicides, assaults and battery) and property crimes (burglaries, thefts and motor vehicle robberies). . This data was used because crimes of this type are more likely to be reported to law enforcement even in neighborhoods where there is mistrust in authorities. These crimes are also less susceptible to police bias, like drug possession, traffic stops and other minor crimes.
Crimes in this new approach are isolated by looking at the spatial and temporal coordinates of each event. On this basis, the city is divided into boxes approximately 300 square meters (1000 feet) wide: therefore, the forecasts do not depend on the type of neighborhood or the political preferences of the various areas. Observe everything “without prejudice”.
And it works: the 90% match rate was obtained with data from 8 US cities: Chicago, Atlanta, Austin, Detroit, Los Angeles, Philadelphia, Portland and San Francisco.
Yes, but is 90% enough?
To do what? To accuse someone in advance? But we're not joking, obviously. This “precrime” is not like the one in Minority Report. You will not see police patrols arresting a criminal “in advance” based on a prediction.
Tools like this are needed ex post, to prepare targeted investments and strengthen the areas that most need intervention.
It is a sort of "digital twin" applied to crimes. You tell him what happened in the past, he tells you what can happen in the future.
“It's not magic, it has limitations,” Chattopadhyay is quick to say, “but it works well. And we can also use it to simulate what happens if crime increases in a certain area. An important factor for evolving our security systems."
I can't wait to make it work in Gotham City. Joke.
