Minisite Gear

Experience the pulse of the Americas with Minisite Gear: Your go-to destination for digital news and analysis.

The Spanish Police already uses an automatic facial recognition system in their investigations |  Technology
Science and Technology

The Spanish Police already uses an automatic facial recognition system in their investigations | Technology

The National Police has been using a facial recognition tool for nine months in various parts of Spain, as sources from the Ministry of the Interior confirm to EL PAÍS. The State Security Forces and Corps have worked for at least four years on the project, the details of which this newspaper reported in November 2022. After several delays, the ABIS program (automatic biometric identification system) was launched. finally launched last August. It uses artificial intelligence (AI) to determine in a few seconds whether the face of someone of whom there are records appears in a given image.

Since it became operational, the tool has been used in at least 400 police investigations, according to the same sources. In 40% of these cases, positive results were obtained that allowed the identification of those involved in the crimes. Interior does not have specific figures about how many arrests these actions have resulted in.

At the moment, there are 13 operational ABIS stations spread across the country. The National Police has two in Madrid and one in Barcelona, ​​Granada, Málaga, Seville, Valencia, Valladolid, Las Palmas, Zaragoza and Bilbao, to which another will soon be added in Pamplona. The Civil Guard, for its part, has two facial recognition stations in the capital. “The system is currently in the expansion phase,” they confirm from the Interior. The Mossos d’Esquadra are also working on the adoption of the system.

The project, coordinated by the General Subdirectorate of Information and Communications Systems for Security, represents a revolution for Spanish police practices. Until now there were two ways with expert validity to confirm the identity of an individual: through their fingerprint or through DNA analysis. This third path, that of the face, does not require taking samples of the subject.

Each person has a unique arrangement of facial features, which varies little over the years. In a first phase, automatic facial recognition systems extract the face from the image using a technology called computer vision; They locate where in the photograph there is a face. Then apply an algorithm to that face to obtain a pattern that represents it and distinguishes it from others. Artificial intelligence allows us to search for that pattern, which is unique for each individual, in extensive image banks and offer the results that are most similar.

A facial recognition system shown at the CES Asia fair in Shanghai identifies a group of event attendees in just a few seconds.Zigor Aldama

The European Artificial Intelligence Regulation prohibits the use of real-time biometric identification systems in public spaces. But the ABIS system does not process images in real time (it is not connected to surveillance cameras). As this newspaper has learned, the Spanish Data Protection Agency (AEPD) has not examined the fit of the tool in the legal framework. The Agency has not received any complaints or queries regarding this technology, and therefore has not investigated it.

France, the Netherlands and Germany are some of the EU countries where police forces have experimented with this technology or where it is already fully operational. Facial recognition has been used for a year at the Union’s borders to register non-EU citizens arriving on the continent. In the United Kingdom, some forces, such as the London Metropolitan Police, have gone further and regularly placed vans with cameras equipped with these systems in busy parts of the city.

In the United States, one of the pioneer countries in the police use of facial recognition, this technology has been declining since the Black Lives Matters movement linked it to police segregation. Other powers, such as Russia or China, regularly use these types of tools to locate suspects.

How does it work

The ABIS algorithm, named Cogent, was developed by the French military technology company Thales. The system compares the image entered by the agents, extracted for example from a security camera or a mobile phone camera, with the photographs available in the system to search for matches. The database against which the searches are carried out is made up of 4.4 million reviews of detainees (3.2 million from the National Police and 1.2 million from the Civil Guard).

The facial patterns and DNA of the registered individuals are stored in this repository. “The system is updated biweekly with the incorporation of new images of detained detainees and the elimination of others due to cancellations of police records,” they say from the Interior. The database is not connected, at the moment, with other European ones, although it will be. The EU Prüm project, which already shares fingerprints and genetic analysis, will soon include facial patterns.

When a search is performed, ABIS orders those 4.4 million tokens from most to least similar to the image entered. The experts took the first dozen responses from the list in search of a match. Each verification is done independently by two operators. It is only considered successful if both agents reach the same conclusion.

“The system works very well,” says Chief Inspector Sergio Castro, head of ABIS. His team has been responsible for the technical implementation of the tool, which they coordinate from the General Scientific Police Station in Madrid. His seven collaborators have trained operators for the ABIS stations that have already been launched in Spain. “The tool itself is very easy to use. The complicated part is the facial comparison, determining that two given references belong to the same individual,” Castro emphasizes. They have designed a specific course that they complement with a training phase in which they show students pairs of images from a database with reference material from cases already solved, so that they can practice and then know if they have got it right or not. . .

Chief Inspector Sergio Castro, of the National Police Corps, is responsible for the ABIS facial recognition system.Samuel Sanchez

The Police facial recognition service receives two different types of orders. First, they are asked to perform one-on-one checks: certify whether the face of the person in an image matches that of the suspect. For example, see if the images of the bank robbery taken by security cameras match those of a detainee to confirm her involvement in the crime. In this expert task, as in fingerprint or DNA analysis, artificial intelligence does not intervene.

Second, do self-checks against the database, or blind search: try to determine if a particular face matches any of the reviews stored in the system. For example, analyze footage of a bank robbery without any candidates, hoping to find suspects to start investigations on. This totally new task could not be carried out without AI, which is responsible for sifting through the millions of records to find matches with the selected facial pattern.

ABIS is good at offering candidates, but it is not capable of solving cases on its own. “The result of a fingerprint analysis gives two results: identified or not identified. With facial recognition, on the other hand, we offer progressive responses: extremely strong, strong or moderate support. We give potential candidates, a starting point to researchers so that, based on their work, they can determine whether the candidate is valid or not,” explains Castro. In practice, that is enough to close cases.

The use of the tool does not necessarily have to be authorized by a judge. Police investigation groups can also request it, says Chief Inspector Castro: “We find that investigation units do not usually require verification of the candidate because the investigation group continues along its classic routes (monitoring, telephone interventions, etc.). ) and sufficient evidence is gathered in this way to make or rule out an arrest,” he adds.

Interior sources indicate that “the system, as declared in the ABIS Treatment Activities Registry, is used with fines for prevention, investigation and detection of criminal infractions, as well as for protection and prevention against security threats. . public.” That is, its use is not limited to the investigation of serious crimes, as was intended before activating the tool.

The last bastion of privacy

There are personal data, such as name, address or identification document, that can be changed. Biometric data, on the other hand, accompany us for life. They refer to unique characteristics of each person, usually physiological or physical. This data is extremely valuable because it can be encrypted and remain immutable over time. We have the same DNA from the moment we are born until we die. The same thing happens with fingerprints, unless we burn them.

The face evolves over the years (we gain or lose weight, we age, we change our hairstyle, we lose hair, we grow or cut our beard), but there are algorithms capable of establishing unique patterns—for example, measuring the distance between the eyes, or the these regarding the nose and mouth—which allow people to be recognized with a high level of accuracy and sustained over time.

“The face is, essentially, the last bastion of privacy,” journalist Kashmir Hill, an expert in this technology and author of the book, said in an interview with EL PAÍS Your face belongs to us (Your face belongs to us, Random House, 2023). Automatic facial recognition systems can be very good at capturing criminals, but uncontrolled use of this tool can jeopardize the anonymity of citizens, as has been seen in China or Palestine. That is the danger that arises from the use of this technology.

A tool with all the guarantees?

One of the main concerns of experts and activists is knowing how the ABIS algorithm has been trained and what type of impact evaluation it has undergone prior to its activation. “Before its launch, the National Police carried out a series of test exercises for four months with real matters linked to crimes already clarified. In this way, the reliability and robustness of the new tool was tested, with very satisfactory results,” they point out from Interior.

But several specialists draw attention to the lack of specificity and detail in the meaning of those “very satisfactory.” According to Carmela Troncoso, professor at the Federal Polytechnic School of Lausanne (Switzerland), “they did some tests based on something that we don’t know what it is, and we decided it was good.” With irony and concern she addresses the topic Troncoso, author of the secure protocol used in Covid tracking applications.

“I see self-affirmation, because the resolved cases are those in which it has been possible to obtain a frontal photo, either of good quality, or there has been some type of additional information with which to triangulate. It would be important to be able to measure how many people the biometrics tool can identify by itself to assess its real usefulness,” says Lorena Jaume-Palasí, expert in ethics and philosophy of law applied to technology and advisor to the European Parliament on issues related to intelligence. artificial.

You can follow EL PAÍS Technology is Facebook and x or sign up here to receive our weekly newsletter.