Last week, La Quadrature du Net submitted its position on so-called “algorithmic” or “automated” video surveillance (also known as “Artificial intelligence-assisted video surveillance” or “AI video surveillance”) to the CNIL [the French Data protection authority]. We supported this position with over 170 contributions from people who mandated us in a popular rogue-consultation.We will delve into the specific arguments that lead us to fight any legitimization of these devices in a future article. For now, it is essential to re-examine the meaning of this term and the exact nature of the technologies that our country has been deploying for several years.
So, what exactly are we talking about?
Definitions
According to the CNIL, “augmented video refers to video devices associated with algorithmic processing implemented by software, which enables automatic, real time and continuous analysis of images captured by a camera.”
Institutions and companies like to call it “Augmented” or “intelligent” video surveillance/videoprotection, we prefer “Algorithmic” or “Automated Video Surveillance” (AVS). All these terms cover the same reality, with more or less fuzzy characteristics and more or less fulfilled ambitions. They refer to an algorithmic layer added to the so-called “classic” video surveillance cameras. The aim is to do an automatic analysis of images captured by cameras, in contrast to the analysis that was up to now carried out by humans, i.e video operators within Urban Supervision Centers (USC). So why these different choices of words?
Because words carry a lot of weight, we prefer “automation” – this term erases the notion of intelligence that technology would supposedly bring to the process. Automation is not a neutral process outside the social world, but one that carries1Florent Castagnino, Rendre « intelligentes » les caméras : déplacement du travail des opérateurs de vidéosurveillance et redéfinition du soupçon, Sciences Po, 2019. with it representations and norms from our social world. And we use “algorithmic” to highlight the addition of software, which is manufactured by start-ups and multinational corporations we know very little about.
This algorithmic layer aims to do video analysis, whether in real time or after the fact, and to find… whatever the police wants to find. This ranges from “detecting suspicious behavior”, to “loitering” (being stationary in a public area), “crossing a line”, tracking a person, or detecting an abandoned object, a fight, a theft, etc.
The deployment of AVS in cities
Algorithmic video surveillance is deployed almost everywhere in France, whether in Toulouse in 2016 with IBM (a project discontinued in 2019), in Nîmes since 2015 with BriefCam [private company], in Marseille since 2018 with the SNEF [French private company], in Paris with the RATP [public transport company] – which authorizes companies to test their algorithms on the metropolitan users,– or in the municipality of Suresnes which allows its population to serve as guinea pigs for the Parisian start-up XXII. Although it is difficult to quantify how many cities use AVS because of their lack of transparency, it is possible to identify at least fifty of them in France alone–the real number unfortunately probably exceeds a hundred.
For several years, we have been following this deployment, which is often obfuscated–not all municipalities are as forthcoming as Estrosi’s [Mayor of Nice]. Up until recently, AVS was primarily implemented through rogue experiments. In January 2022, the CNIL explicitly asked security companies to provide feedback on the use of these technologies “in order to assist their deployment”. By doing so, the CNIL clearly picked the side of the security industry. The AVS technology seems to now be on its way to flooding the digital urban security market.
And the AVS market is taking shape: while players such as IBM in Toulouse have failed to make an effective product and seem to have withdrawn from that market, the Israeli company BriefCam (part of the Canon group) claims to dominate the market in France. In the meantime, cities are signing partnerships with French start-ups or companies, supported by political decision-makers, in order to make French industries competitive in the international market of digital urban security.
Two examples: BriefCam and Two-I
BriefCam is a company that produces algorithmic video surveillance software and is very well established in France, the United States and in some 40 other countries. In 2020, more than 35 French cities were already using its software. What follows is a rather revealing demonstration.
In this video, the company claims to be able to condense hours of videos into a few minutes, to track people based on varying attributes (woman, man, bag, hats, colors of clothes …). We also know that BriefCam’s software has a facial recognition option, and security officials are very eager to use it.
Two-I is a French start-up based in Metz. Having first tried to establish themselves in emotion detection (notably the aborted experiment in Nice with emotion detection in streetcars), its founders finally launched their algorithmic video surveillance activity [“Video Content Analytics”] and the design of so-called “hypervision platform”. These platforms map and translate the large amount of data collected by cameras and algorithms, with the aim of “optimizing city management”. In short, these platforms make algorithmic video surveillance usable, by mapping the data and alerts captured by cameras and algorithms.
These examples of BriefCam or Two-I software are only two examples among a dozen of other security companies of the same type. They clearly show not only the reality of this new security market, but also a shift in the concept of security itself. From now on, the private companies that design algorithms sold to local authorities get to decide the meaning “abnormal or suspicious behavior”. Through automation, companies acquire a police power and dictate norms and behaviors in public space, perfectly in line with expansion of security policies.
The effects of AVS
The addition of algorithms to “classic” video surveillance is not insignificant. It reflects a change of scale in camera surveillance which, until now, as Tanguy Le Goff2Tanguy Le Goff, « Dans les « coulisses » du métier d’opérateur de vidéosurveillance », Criminologie, vol. 46, n°2, 2013, p. 91-108. describes it, was a “surveillance job […] considered boring and monotonous” in which video operators set up strategies to only carry out this job in a limited way.
The automation of surveillance is supposed to increase the vision behind the camera screen by tenfold. The result is, in particular, the criminalization of behaviors that were considered previously almost harmless, such as littering, not wearing a mask or dog poop. Automation allows the police to extend their authority and action to new fields in which they now have repressive power. Thus, the police can dramatically increase their power to normalize public space. If marauding, i.e. remaining stationary for more than 300 seconds, alerts the police, one can fear for the people who cannot see the street as a “simple place of passage” because they live there or have made it a necessary social landmark. In a future article, we will talk about the fact that algorithms on cameras increase police repression on populations already particularly targeted by law enforcement.
Another aspect of AVS is the increasing tendency towards data mining. Beyond intensifying the surveillance of public space and the normalization of behaviors, AVS is also a feast for a whole data market. Under the CNIL’s promise of so-called “supervision”, Technopolice companies can use public areas and the people who use them or live there as “walking data”. And security companies would be able to make money on us, to improve their repression algorithms and then sell them on the international market. This is what the French multinational Idémia is doing, refining its facial recognition devices at French airports with the PARAFE or MONA devices, and then selling facial recognition equipments to China to participate in the mass surveillance and genocide of the Uyghurs, or to win European Union tenders for biometrics at EU borders.
What does the AVS actually represent?
Automated Video Surveillance and the Smart City feed the same fiction: the fiction of a city where sensors collect data, where algorithms sort the data and detect issues and where a central platform would allow the police to manage the city remotely.
Algorithmic video surveillance is a security market that is trying to grow by becoming “digital”, that is, based on AI and algorithms. As Myrtille Picaud3Myrtille Picaud Peur sur la ville. La sécurité numérique pour l’espace urbain en France [Rapport de recherche] 01/2021, Chaire « Villes et numérique », École urbaine de Sciences Po, 2021. shows, French industrial and political decision-makers are pushing to organize a strong security industry in order to compete on an international scale, which represents a huge economic market. Major sporting events such as the Paris Olympic Games in 2024 or the Rugby World Cup in 2023 represent a great opportunity to accelerate the development of security technologies. They offer a showcase for French companies and help to normalise these devices.
For the manufacturers, AVS means the opportunity of justifying deploying hundreds of thousands of cameras in France. This deployment is widely denounced4See these three reports: one by Guillaume Gormand asked by the “CREOGN”, another one about vidéosurveillance in small town, november 2021, and a report about municipal police, october 2020. even by public institutions, but by a magic trick, it would suddenly make sense because of the addition of algorithms – an absurd argument that we will discuss further in a future article. AVS would allow cameras to be used to their full potential and even accelerate their implementation: more cameras will be needed, and the ones that aren’t good enough for algorithms will have to be replaced. Above all, this will ensure further excessive deployment of cameras whilst filling the pockets of private companies.
In addition to justifying the multiplication of cameras, AVS is a politically profitable resource5Laurent Mucchielli, Vous êtes filmés, Armand Colin, 2018, page 80., which explains the immoderate infatuation of local elected officials for video surveillance. AVS is a short-term measure that elected representatives can use to show that they are taking action. Digital security6 Guillaume Faburel, Les métropoles barbares, Paris, Le passager clandestin, 2020 [2018], page 46. is also a source of attractiveness and distinction in territorial competition, in the search for symbolic assets to attract tourists and the upper middle class. Algorithmic video surveillance renews the belief in a “technological prophecy”7Ibid. inherent to the Smart City. This belief makes it possible to continue deploying cameras and, above all, to find more and more economic opportunities.
Conclusion
All in all, Automated Video Surveillance is a technology that is in the process of being widely deployed in France, and may already be so more than we think. It is used to justify the size of the French video surveillance fleet, by attempting to make it more efficient through automated detection. This technology is part of the broader narrative of the Smart City, which bases city management on AI and represents a huge economic market. These automation technologies further reduce our freedom in streets and public squares, and further increase repression of people already disproportionately targeted by the police. In future article, we will further explain why we are against this technology and how it is possible to fight against its deployment.
This article has been translated by our volunteer group. Warm thanks to them all <3
References