How Artificial Intelligence Is Used by Prosecutors to Track Criminal Networks

a Artificial intelligence (AI) Prosecutor’s Office databases and social networks can be scrutinized to track down potential suspects in many types of crimes. This is a project that a team of scientists has been working on for at least two years, Chilean Public Prosecutor’s Office Initially, we investigated crimes against property, but soon hope to apply it to violent crimes and drug trafficking.

It doesn’t have an official name yet, but it’s already implemented and prosecutors are training to use it.say so Dr. Richard Weber, the German engineer and researcher responsible for the team that develops this system. He shares his team of three other researchers, most of whom are members of the Institute of Complex Engineering Systems (ISCI), with his 12 dissertation students from the University of Chile, the University of Bio-Bio, and the University of the Andes. I am in charge of the team.

“What we are developing on that platform is a model that uses artificial intelligence and social network analysis to detect organized crime structures.” Weber explains to BioBioChile. “For example, we can help prosecutors working on cases find similar cases to enhance case investigations and provide reliable information to prosecutors.”

So far, the platform has worked with data from public ministries and companies, he details. Sosafe , owner of the homonymous application where victims and witnesses report crimes and incidents.It is also possible to use twitter downloads messages from social networks and scrutinizes them for matches.

The country’s pioneering initiative originated as a project of the Fund for the Advancement of Science and Technology (Fondef), funded by the National Institute for Research and Development (ANID). “We already said at that moment, before working with the Public Prosecutor’s Office, that we could do a little more, but we need funding for that,” he recalls. Memory with a rudimentary version of the system thus became the promise of an important criminal tool.

Already tested and successful.

“This idea isn’t that revolutionary. I’m sure many of you have thought about it and tried it out in college. Installing it from there is another challenge,” assesses Weber. He personally doesn’t know of any other similar systems around the world, but says it’s entirely possible they already exist.

To date, he says: In the next phase, we plan to extend the system to run on a server or in the cloud, making it accessible to all prosecutors in the country. ”

This stage is already more delicate as digital security emerges to protect such sensitive tools.Weber recalls, for example, what recently happened Chat GPT personal information of payment service users was accidentally leaked. “These are problems that must be avoided, but they are not typical of artificial intelligence. They are problems of system implementation,” he qualifies.

As for the system itself, it has already been proven to work.

Even four years before full-scale work on the current project began, the model was already being developed and tested with information from the Biobio Regional Prosecutor’s Office.

“They told us about the crimes committed by 12 criminals. The prosecution already knew about them, but only two of the 12 they gave us. We ran the model in various configurations and found 3 to 8 out of 10 criminals that didn’t exist at the time,” recalls the engineer. “andvery good result This is because it supports the prosecutor in charge of the investigation.”

That is, in doing so, the system searched suspects “after the fact” for evidence, but indicated that the search could have been easier if it had been used earlier.

criminals profiled

“We have two models that we use in the prosecutor’s office,” Weber elaborates. One is a natural language-based artificial intelligence model like GPT and the other analyzes networks and connections. The system then analyzes potential suspects based on a variety of criteria, including criminal prowess, background, proximity to previously identified individuals, and whether they have participated in past crimes together.

“So we network of suspects, We run an optimization algorithm to provide associations that satisfy the group’s advanced criminal prowess. Ask the algorithm, “Give her a group of eight candidates out of five who are also highly criminal and have good relationships.” Those “”, he exemplifies.

Investigators, on the other hand, apply artificial intelligence such as GPT to interpret language and “analyze the victim’s story” embodied in the affidavit. “For example, ‘Last night at 10pm he came into my house with five people. They beat us up.'” Also, there are misspellings, but humans know how to interpret them and extract important information. modus operandiHow many people were there, when did they enter, and where did they come from?

“Classical computing without artificial intelligence has trouble processing this text, but LLM models (Large Language Models) are able to extract important information.You can apply them to determine similarities between stories. ” explanation.

In this way, the prosecutor in charge of the case can quickly see if other previous cases are similar to his own, how long ago they occurred, and how similar they are. increase. modus operandimakes us think about the possibility of crimes by the same person.

“So what we apply here is an intelligent filter that goes through all the reports in the database and delivers only those cases that have a high degree of similarity to the prosecutor in charge. Instead, select 5 or 10 cases that are very similar. modus operandi Thanks to the artificial intelligence that interprets the text. ”

Similarly, both this case and other similar cases show that it is important to consider: Human biases when programming machines so that it does not affect its operation.

What is the future of national criminal investigation?

All of the above are tasks that were done manually until today, and this platform aims to speed things up without losing efficiency.

Weber is State Attorney Angel Valencia but also by his subordinates, when he presented the training this Thursday, only about five or six officials were expected, so about 30 prosecutors arrived.

As he points out, they, along with his team, hope their invention will impact research time and allow them to narrow down a search of thousands of “candidates” to, say, five but it’s hard to estimate how much.

However, Weber is clear that his platform is merely a tool, not a replacement for essential human judgment. “I’m not going to replace the prosecutor in charge.”

“We believe the system should not work alone and make decisions alone. It’s just a filter for the prosecutor in charge and one should always have the last word.” ‘ he emphasizes.

And it’s still in development. “There are several aspects to predicting the future. One is adapting the model to different types of crimes: murder, drugs, arson, which was a very important topic four weeks ago. There is also another aspect of development that has to do with the software itself,” he predicts.

“We need to upload it to the cloud and make it available and ensure the security of the data. Obviously, in that case we are very sensitive. engineering software And they aren’t even artificial intelligence,” he explains.

The third dimension would be “Think of other organizations and scale the system.”I know that Chile has only one public ministry, but for example Carabinieri You can become a user Government offices of neighboring countries.

“And the fourth line is to feed more data into the model. Topics that I think are very important are: The models we use today use very little information. I think it’s positive that we got such a good result.with more information, e.g. military police or of konaf In case of fire and better integration with data Poi and Carabinieri We were able to enrich the model even more and improve its performance,” says Weber.

The possibility of this consolidation has already been discussed, according to the engineer, stressing that it has received significant support from the public prosecutor’s office, which is leading these investigative efforts. “The new State Prosecutor took office in January and we met in February. I presented him with what we had and invited him to speak at an event last week. He accepted and offered us his “home”. We did it in the public prosecutor’s office auditorium,” he says. “There’s interest, but there’s also commitment. It’s very powerful to go out in public and say ‘that’s what we’re going to do’ in front of the authorities in attendance.”

Fundamental Issues: Law and Education

In addition, Emphasizing the need for legal regulation of this and other similar technologies, To educate the population, as big countries such as the European Union are already doing.

“In my opinion, we need to educate artificial intelligence in schools, not so that children have the ability to develop it, but to understand in which contexts it applies and in which not. It’s about being capable,” says the engineer. “It cannot be that it is not known, that one does not know what it is.”

For example, Weber recalls: They used it for a simple calculation: 3 x 4 and the bad calculator returned -7. The kids believed it because the calculator said so, and the calculator wasn’t wrong. It’s clearly happening here as well.”

We already see it with artificial intelligence-generated “pictures” and the difficulty some people have in telling whether they are real or not. “Artificially created images look perfect. Donald Trump, terrible. Not just in the case of images,[people]believe it without developing such critical reasoning. ”

“I also know a letter signed by over 1,100 tech people saying artificial intelligence training must stop. We believe that the path must be different, by educating ourselves to better understand its possibilities and limitations, rather than hindering us,” says the expert.

In addition, there is the problem that the industry is highly competitive, and there is no guarantee that competition will continue if a major company closes its labs. “So I don’t know if it really makes sense. It can’t be implemented.”