Can facial recognition really improve crowd management?

Three days after refusing to maintain order at the Stade de France, Nice Mayor Christian Estrosi called for the use of facial recognition to facilitate the management of large events. But what contribution could algorithms have made to apparently disorganized crowd control? One year before the Rugby World Cup, two years before the Olympic Games in Paris and full work on the European rules for artificial intelligence is not a trivial matter.

Technosolution or technodiversion?

First, it is a matter of organizing the Champions League final. Home Secretary Gérald Darmanin was quick to mention 30,000 to 40,000 British fans with counterfeit tickets, which is so unlikely that Liverpool club president is demanding an apology from the government. It’s the same at Real Madrid: we want an explanation for the chaos on May 28, and quickly. Talking about facial recognition in this context is a distracting strategy, says Arthur Messo of the digital rights association La Quadrature du Net: a human organization that technology would never allow to be repaired. »

The author of a study on the effectiveness of video surveillance systems used by the gendarmerie, Guillaume Gorman, agrees: “The problem is not so much the availability of the tool, but how it is used. With proper operation, surveillance systems around the stadium would potentially suffice. But if the organization is not monitored, then neither the camera nor the algorithm will compensate for the flaws in management.

Video surveillance algorithms

Nevertheless, Christian Estrosi’s proposal fits into a certain context. The first is the bloat of surveillance devices in the public space, which Guillaume Gorman describes as a concomitant private lobby. “After legalization in 1995, we first saw how large municipalities were settling down, then the market froze. It started again when Nicolas Sarkozy ideologically promoted video surveillance during his campaign and later his mandate. “In 2019, the Gazette des communes had 11,470 cameras in the 47 largest cities in France, i.e. 2.4 times more than in 2013. Except that this is just a base, – the researcher continues: – Today we add a lot of tools, counting software, algorithms, to conduct experiments. Unless testing is good, but then you need to check if it works. »

However, the effectiveness of the two technologies – pure video surveillance, firstly, the effectiveness of the algorithms applied to it, secondly – is a matter of debate. In 2018, sociologist Laurent Mucchielli reported that only 1 to 3% of offenses committed on public roads are revealed through camera images. In 2020, the Accounts Chamber indicated that the amounts spent on these devices are constantly increasing, with very few studies proving their effectiveness. In a paper commissioned by the Research Center of the School of Gendarmerie Officers, Guillaume Gorman, for his part, notes that video surveillance does not have a deterrent or preemptive effect on crime. In addition, the collected images were only used in 1.13% of the surveys in his corpus. However, he elaborates, video surveillance “can be useful in managing large events, as it allows the authorities to project themselves onto the ground, to see live what is happening there.

Regulate to avoid abuse?

Therefore, video devices can be useful from this point of view. Although, the researcher further points out, “in sensitive events, actors on the ground turn to what they do best: radio, live broadcasts, information collected on the spot … New technologies come second. And then it does not solve the question of algorithms. Using them to identify individuals “banned from the stadium” who are going to “pick them up early in the morning”, as suggested by Christian Estrosi, fits into the context of more or less regulated technological experiments at sporting events. In January 2019, supporters of the Metz football club, for example, were surprised to learn from the press that they were being tested by Two-I algorithms.

Precisely to avoid such problems, the three senators presented on May 10 a report on biometric recognition in public space. In 30 sentences, they recommend “drawing red lines to avoid surveillance of society and passing an experimental law within three years to test different use cases,” summarizes co-reporter Mark-Philippe Dobress (LR). Among the proposed restrictions: a ban on real-time facial recognition without the ability to obtain people’s consent, “with the exception of access to dangerous facilities, such as nuclear facilities, or to large events with the risk of overcrowding or terrorist attack, such as the Olympic Games.” In fact, such a proposal reopens the discussion about the use of facial recognition in 2024, although in October 2021 this prospect was rejected by an inter-ministerial delegate at the Olympics.

Little Music Surveillance Technology

First of all, the senators are calling for clarity in the debate on the use cases of this type of algorithm. In this, they join researchers from the Department of Legal Implications of Artificial Intelligence at the University of Grenoble-Alpes, who have authored a detailed mapping of the use of facial recognition in Europe. Essentially, these actors are saying that authenticating a user with their consent (for example, comparing their face at the entrance to a stadium with a photo that they allegedly took on their smartphone a few days earlier) has much less impact. crowd, as the London police do.

But for Arthur Messot of La Quadrature du Net, “it is at this time, when there is little ideological debate about the interests of good organization and shortening lines, that the government or the police are most adept at promoting privacy-threatening technologies. . “The introduction of facial recognition to smooth out the flow of supporters seems harmless, relatively neutral, but accustoms the population to invasive technologies. “The fact that Christian Estrosi is offering facial recognition at the Stade de France is actually almost absurd. But by offering this sort of thing every six months, they are accustoming us to little surveillance music. »

In the background regulation of artificial intelligence

If Marc-Philippe Dobress and his co-rapporteurs Arnaud de Belenet (LReM) and Jérôme Durin (Socialist) call for a biometric recognition system that allows certain uses and prohibits others, on the side of La Quadrature du Net, we are therefore strongly opposed. In early June, the association even launched a collective complaint campaign to the CNIL against “Technopolice” – a term by which it groups together many of the technologies promoted by Christian Estrosi – video surveillance, facial recognition algorithms, or even automated systems. behavior detection.

The two perfectly illustrate one of the debates that is likely to rock not only the Olympic Games preparations, but the European Parliament when it comes to regulating algorithmic surveillance technologies: wouldn’t it be better to ban them outright, as Germany in particular is calling for? Or rather to accept certain applications subject to experimentation?