Los Angeles, Sheriffs and Artificial Intelligence: "More Sim City than GTA"

Los Angeles, Sheriffs and Artificial Intelligence: "More Sim City than GTA"

Designed by an Italian engineer, artificial intelligence follows game theory with the aim of defeating the enemy: criminals and terrorists

Browse gallery

Modernize who used to work from experience and convinced that this solves any problem or allows to take at best any decision a company. It looks like an Italian story, but not so – or almost – that of Francesco Delle Fave, a graduate in Computer Engineering (at La Sapienza University) and researcher in the field of artificial intelligence at the USC (University of Southern California). Francesco for several months has represented science and modernization in the midst of the Los Angeles sheriffs.

The task assigned to his department and to the team he coordinated (including 5 researchers between doctoral and postdoc, a professor and two managers, a researcher in psychology and social sciences, a lieutenant and two sergeants in the Los Angeles Sheriff Department) was to study the Sheriffs patrol system and elaborate how an artificial intelligence could organize them better, both in daily work and especially in case of emergency terrorism. The result TRUSTS (Tactical randomization for urban security in transit systems) a app for smartphones supplied to every patrol that monitors the movements through the GPS, which assigns them the daily tasks and with which they can communicate with the central.

The app installed on an ordinance smartphoneThe app installed on an ordering smartphone

Not only are you not a sheriff policeman, but you have no patrol background, and in addition you are not even familiar with the urban fabric of Los Angeles, how much did you have to study them before you started working on the technology?"I patrolled with the agents for a few months both before and after the development part, first to understand and then to test. But it was also fundamental to get information from them. Although the project was approved, the agents were not very inclined to cooperate, he didn't like a machine telling him what to do and replacing the experience gained. Initially, in short, they were wary and reticent about innovation but after a few months on patrol with them they had widely accepted me, they called me Bro. Maximum loneliness ".

Have you studied what they do normally, but your goal was to improve it?"Otherwise. First I had to understand what their job (they do not just patrol, they must also make metro ticket checkers for example) and how they react in the event of terrorist emergencies or even just criminals. For each operation, I studied which types of agents are used: motorbike agents, Federal Air Marshals (those in black), armed sheriffs and regular controllers. I also studied the different LA metro lines, from the most used (the red line) to the most dangerous (the blue line). Based on all this information, together with some members of my group we have developed an automatic resource allocation system (ie agents) ".

Francesco Delle Fave explains the functioning of the technology during a testFrancesco Delle Fave explains the functioning of the technology during a test

So you were forced to follow them during emergencies?"S. Obviously nothing risky, I was just a burden to them. However in some cases they sent me away because the situation was heating up ”.

How does the TRUSTS IIA work? I guess you don't just place them randomly to cover all areas“No, in fact, also because the stations or even the trains themselves have different importance, depending on the area where they are located and the number of passengers. We have modeled the allocation of resources as a competition between sheriffs and "adversaries", that is terrorists, criminals and passengers without a ticket. The competition modeled through game theory to be unpredictable. Not being able to predict the movements of sheriffs, planning an attack will be very difficult for any opponent. The system takes into account all these factors and produces, on a daily basis, a plan for each agent (based on the number of working hours, and other clearly available agents) that is displayed via a app ".

If you tell me artificial intelligence, game theory, LA sheriffs and attack, I think of gamification and GTA."Don't exaggerate. My job is to use game theory to model strategic resource allocation and if you really need to make a comparison I'd do it with video games with a strong strategic component: Sim City, Civilization, Star Craft … If I game as a security agent who must protect a metro line (ie win the game), I have a number of resources of different types available (soldiers, flying …), how I use them to maximize the chances of winning, taking into account that I have against an opponent , a terrorist group, of criminals (the other player) who does the same to beat me? In this sense we are close to strategy video games ".

I wasn't too far away …"Basically it is the foundations of artificial intelligence: the strategy is a form of intelligence, so to automate the strategy one of the most obvious forms of Artificial Intelligence. Not for nothing many of the most famous strategic games can easily be modeled and solved with game theory: poker, chess, checkers … And the various strategy video games are nothing but evolutions of these games ”.

You did it, finished it, tested it and delivered it. Will you be applied?“The prototype of the system is ready and fortunately it has been appreciated both by the scientific community and by the Transportation Security Administration (TSA), the American agency that deals with the security of American transport systems and which has funded the project. These days, my group's start-up, ARMOR-WAY, is negotiating with the TSA to get funds to turn TRUSTS into a real system. This morning, on the other hand, I received an email telling me that discussions were started to export the system to the railway lines as well as the underground lines ”.

Did you understand where the interest came from?"From what I could understand in these two years, after September 11th the US government decided to invest heavily in the safety of its transport systems. I am constantly at risk of a terrorist attack. As for the subway lines, my feeling is that after the ISIS announcement of terrorist attacks on the American metropolitan systems, the TSA has decided to invest even more in the safety of its infrastructure ".

A sheriff during an application testA sheriff during an application test

You were the project coordinator and you probably know better than anyone the intelligence you have planned. If you were to fight against it, that is, if you were to attack Los Angeles, would you know how?“A very good question, I would say a research problem on which we are beginning to think. According to my boss the only way not to commit crimes and, however simple, the right answer: TRUSTS, thanks to game theory, built to be unpredictable. So even I have little chance of predicting how to allocate resources while knowing the algorithm perfectly. The only possible flaw is the data. Like most modern artificial intelligences (see the algorithms of Amazon, Facebook or Google) TRUSTS uses data in addition to game theory. The problem then becomes: how much can we trust the data we use? If they are wrong or, worse still, manipulated, the system could allocate resources in an inefficient and therefore dangerous way. It is not a problem of the system, but of the inputs that are given to it. If these are incorrect or manipulated, the quality suffers. For this reason we have begun to work on "robust" algorithms, which are able to generate efficient allocations even when the data are uncertain ".


You may also be interested