Wired Italia: CAUGHT!
On the computer screen the map of Milan is marked by some red dots: a line connects them, node after node, following a precise chronological itinerary. It is the crime series carried out by a robber, target after target.
Their identity is not known but the police are convinced that it is always the same individual. “We can also hypothesize where he lives or is based”, explains Mario Venturi, assistant chief at the Milan police headquarters, pointing with his finger at an area of the digital map. It is the empty area on the map, the one in which the suspect will never work as a robber, because it is too close to him. The goal here is not so much to identify the neighborhood in which the robber goes for a walk near his home, but to predict the day, the street and hopefully the shop he will target next time. An ambition that sounds like it belongs in a Spielberg film, but which many universities and law enforcement departments around the world are thinking about. Except that, apparently, in the Lombardy capital of Milan, they have beaten everyone to the punch, in that typically Italian way that mixes the art of getting by and chronic lack of resources.
The analysis and planning office of the police headquarters is an austere square room, full of monitors – there are at least three for each workstation – as well as a giant screen hanging on the wall. On the opposite wall, on some white boards, dozens and dozens of ID photos stand out, creating an modern day mosaic made out of crime.
The atmosphere of an American TV series, buttoned up and filled with scientists, is however tempered by large letters on the wall that dominate the room and bring everything back to a more human dimension: “From a man’s nails, from the sleeves of his jacket, from his boots, from the knees of his trousers, the calluses on his forefinger and thumb, from his expression, from the cuffs of his shirt, from all these things one understands the occupation of a person it is all that is necessary to enlighten an experienced investigator”. These are the words of Sherlock Holmes, or rather of Sir Arthur Conan Doyle, from way back in 1892. That sentence is in fact reflective of the method and philosophy from which this office was born. This, despite the name and the sobriety, is the heart of Keycrime, the software used for years by the Milanese police headquarters to predict where and when the next robberies will take place.
The four who work in the office, although surrounded by monitors and computers and immersed in a constant stream of data, are not computer science or criminology graduates from Berkeley, but just “cops” who squeeze your hand tightly when they shake it. They are used to being in the field, with years of experience in patrol cars and on the streets. This is their trump card that wins over the more academic experiences of traditional predictive policing.
Keycrime was born from the intuition and stubbornness of Mario Venturi, 48, in the police since 1987, a past spent between patrol cars, Digos, anti-drug division and in his pocket an eighth grade diploma. In 2007, after working, in his spare time and with the help of a couple of friends, on a program to try to prevent robberies, he proposed using in real life to the police headquarters. Who accepted, despite the unorthodox situation, and began an experiment. Thus began, on the one hand, a maniacal collection of data for criminal events that took place in the city of Milan, refined over time, in turn improving the software. On the other hand, the first surprising results started to arrive: in 2006, only 10% of commercial robberies were solved; in 2007 they grew to 20%; in 2011 they reached 56%. In 2009 the program was extended to banks throughout the province and here too the crimes solved increased by a few percentage points. Among the most striking cases, that of a robber stopped before even entering a pharmacy, officers found the toy gun he used to “work” in his pants pocket.
Software for analyzing crime data and to generate risk maps, areas and neighborhoods where multiple crimes occur at certain times, in order to plan deployment of forces and patrols, have been around for some time. The first of its kind, called Comp-Stat, was born in the New York of Rudy Giuliani, the mayor with the bent of a sheriff of the 1990s. With Keycrime, however, we are on another level because the analysis is dynamic. “I am not interested in knowing today that a certain area has been hit. Or rather, I want to know, but to understand which will be the one targeted tomorrow”, Venturi specifies, gesturing toward the monitor. His software only applies to robbers for various reasons: the main one is that, to make prediction possible, the crime must be serial. “The risk areas, the hotspots are good for static crimes. But robberies are itinerant and often repeated by their perpetrators”. At the moment in which we speak, a young colleague of his explains to me, putting himself in front of a computer, in Miano there are currently five active crime series, that is, five groups or subjects that perform repetitive crimes, each with its own patterns and territory.
“The software examines three aspects: mathematical analysis, classical investigative activity and behavioral psychology”, Venturi resumes explaining, and there is no doubt that he, with the airs and build of the solid Friulian that he is, embodies all these dimensions. The challenge was to bring them over into a computer program.
So the team led by Venturi has drawn up a specific protocol, with detailed forms that are distributed to agents in the area that are to be filled in by all witnesses of a robbery. Not only that: twenty-four hours after the event, just to take into account that mental process that blurs memories immediately after a trauma, Keycrime analysts call the victims for a second interview, proceeding with a data entry guided by the same methodology. The software varies the questions depending on the previous answers. Everything has a weight: the logo on a jacket, the sticker on the helmet, the words used, the color of the sweater, the way he walked, how he pointed the gun. Above all, their attitude is analyzed. “Crime is terribly revealing,” wrote Agatha Christie. And behavior is even more so. “Especially in times of stress, you tend to do the same things over and over again,” comments Venturi.
Once the information on a new criminal event is entered, Keycrime begins to process it and compare it with the database of 7,000 robberies. The software is elastic: it doesn’t go haywire just because a parameter doesn’t match at a certain point.
After all, a witness may mistake a gray car for a white one. After a few seconds it brings out other events attributable to a series, which the operator checks one by one to verify a possible match.
Only at this point does the second phase begin: examining series patterns to identify the area and targets that could be targeted in subsequent crimes. It is one of the crucial aspects of KeyCrime and Venturi does not give away too much: “Once we have a series, we monitor it over time on a map. Sequential analysis allows us to understand the logic behind their actions to the point of predicting their next move”.
Keycrime is so effective that it has even aroused the interest of the University of Southern California, which asked Venturi to participate in an American symposium on crime-predicting policies. The robber catching software was validated by the study of an economics professor at the University of Essex, Giovanni Mastrobuoni: data in hand, he defined it as a formidable tool, more advanced than those in use in the United States. However, something deeply handcrafted remains in the program. Contrary to the philosophy of Big Data and the undifferentiated collection of elements, here the procedure is very targeted and, although it makes use of thousands of data points, it aims at the identification of a criminal’s individual characteristics. In short, we are in the area of Criminal Minds, to mention a TV series again, even bordering on that of the precogs in Minority Report. Which from a certain point of view makes it a little more reassuring than that threateningly predictive software that examines wide-ranging information and raises a lot of privacy questions.
“We investigate every criminal event as if it were a murder”, Venturi points out, “and we link it, if this exists, to a series. In addition to preventing subsequent crimes in the series, all this allows, once a suspect has been arrested, to produce evidence also on the facts and when suspects are questions based on specific and detailed elements from previous crimes, many end up admitting the old robberies”. An interesting portrait of the Milanese robber also emerges from this picture, as it was sketched in Mastrobuoni’s analysis: he is mainly Italian, has an average age of 31 and makes an average haul of 2000 euros. In one in four cases he is armed, and in 10% of the crimes he uses a knife.
However, there is one aspect where Keycrime’s “manual” approach is no longer a positive. Keycrime is used only by the Milan police to which Venturi, who is the owner, has allowed them to use it at no cost for the state coffers. Despite the results obtained, judged “extraordinary” by the commissioner Luigi Savina himself, he continues to find no investments: yet, to be extended to other realities and to get out of the status of a promising prototype, he would need resources, work and a better database. “For me the software is like a son, I worked on it for ten years”, says the assistant chief, with a certain weariness. “But I think the time has come for us to make a qualitative leap.” Venturi once again starts his program on the computer and looks at it with increasing satisfaction: he knows he has a key in his pocket that can open some doors, but the way to get there seems blocked. And his fear is that in the end it will turn out to be a dead end.
Source: Wired Italia