Reversed Surveillance
2023

Text by Carolina Semprucci:

​

Reversed surveillance by Marcel Top

Reversed surveillance is a visualisation of invasive surveillance technologies, challenging the prospect of automating crime detection during protests in France. The project reflects on the development and legislation of mass surveillance by offering a case study on how individuals can protect themselves using the very tools with which they are surveilled.

Following the enactment of new legislation allowing the use of AI-enhanced surveillance in France (Law No. 2023-380), Marcel Top uses accessible algorithms and available protest footage to simulate how the use of Algorithmic Video Surveillance (AVS) can impact a protest. Regarding demonstrators, AVS uses a series of algorithms (including, but not limited to, face collection, crowd analysis, emotion recognition, and movement analysis) that can flag abnormal behaviour in real time, assisting the police in identifying and possibly removing suspects from the crowd.

Starting from a three-hour livestream of the Paris protest of March 11th 2023, and using the software Video Content Analysis and an AI trained to recognise the seven facial expressions associated with emotions, Top collected around 30,000 faces and categorised them according to the emotional readings assigned by the algorithm. The process revealed that, perhaps unsurprisingly, the AI’s conclusions regarding people’s emotional state were mostly inaccurate: the result of forced categorisation via metrics that do not consider overall emotions but the mathematic average of facial expressions linked to them. The research shows that, given the same input, AI and humans do not necessarily provide the same emotional reading.

Despite reassurances from the French authorities that AVS is merely a safety measure for crowd control that will not overtake human decision-making, their history of systemic hostility towards protesters and track record of misusing surveillance tools risk placing this technology under the umbrella of repressive policing. The project also shows that, when flagging suspects, the accuracy of the algorithm is irrelevant because, as long as those in charge aim to repress protesters, the findings will be used to serve that purpose.

In a second phase, Top reversed the dynamics of surveillance. He trained an algorithm to recognise different police units during protests in France. Using more protest videos and livestreams, he created a police unit recognition algorithm and the blue print for a facial recognition tool that, with the right database, could theoretically link the police officers’ faces to their RIO (Relevé d'Identité Opérateur), the French police’s unique identification number. 

Relentless advancements in the field of surveillance technologies have led to their legislative implementation growing disproportionately compared to the legal frameworks regulating their use. In a time where machines are tasked to decode human emotions and authorities are allowed to make decisions based on probability rather than actual events, every new law that seeks to oppress protestors in the name of safety confirms that it is too late to undo surveillance.

Top’s project originates from the belief that, as it is unlikely to revert surveillance and its effects on society, society should explore ways to reverse it. In Reversed surveillance, artist Marcel Top demonstrates the potential of reversing surveillance dynamics to protect the right to protest: by gaining knowledge of accessible surveillance tools, it is possible to ensure police accountability.