On October 19, four men sneaked into the Louvre museum and stole jewels worth an estimated 88 million euros. As reported by the newspaper Libération, the museum’s security organization was largely insufficient. Whether it’s a ridiculously weak password (the very secure “Louvre”) to access security cameras, aging surveillance technologies, roof security “easily accessible in case of work” or even poor management of visitor flows: several security flaws have been blamed.
This context allowed the four men to easily blend into the Parisian crowd. Equipped with high visibility vests, the thieves managed to pose as construction workers and arrived with a freight elevator to reach one of the museum’s windows. However, wouldn’t it be a little too easy to blame security management alone?
If four thieves managed to break into the most visited museum in the world, in broad daylight and in front of everyone’s eyes, what exactly does that say about us? For journalists Vincent Charles and Tatiana Gherman from the media Ars Technica, the thugs’ daring strategy worked because we all have cognitive biases, that is to say we do not see the world objectively.
In his work The staging of daily lifethe sociologist Erving Goffman compared social life to a theater stage where each individual plays a role according to the social context. For him, we adapt our behaviors, our language and our appearance according to the norms of the audience. In other words, we are all actors in our social life. In the case of the Louvre theft, the performance of normality became the burglars’ camouflage. They adapted to the signals that the public expected.
When artificial intelligence reproduces our biases
Before Goffman, a plethora of philosophers had thought about these errors of perception and judgment. Already in the 18th centurye century, Kant theorized the notion of “categories of understanding”mental structures that shape our experience of the world. For the German philosopher, we never perceive reality objectively, but through categories that allow our mind to organize our different perceptions and sensations into intelligible experiences.
If humans are victims of their cognitive biases, it seems that machines are not immune either. Indeed, many artificial intelligence systems used for tasks such as facial recognition or detecting suspicious activity in a public space work in the same way. If humans constantly categorize mentally to give meaning to the people, places or situations they encounter, AI adapts to a strictly human database, indicating what is supposedly normal and what is not. Because AI does not invent its categories, but it learns ours.
Whether it is a mathematical or cultural scheme, the process is ultimately the same: classifying individuals according to seemingly objective indices, but in reality influenced by our biases. And just as the museum guards ignored the thieves because they seemed like they belonged in their environment, the surveillance AI will be able to ignore certain patterns.
This can lead a facial recognition system to disproportionately flag certain racialized or gendered groups as potential threats because they do not match the standards of the data used to feed the algorithm. In the second half of the 20th century, the philosopher Michel Foucault warned about this classification of individuals. Normality, he said, is socially constructed and is akin to a powerful mechanism of power.
After the Louvre theft, Rachida Dati promised to invest in new surveillance cameras. But, no matter how sophisticated these systems are, they will always depend on our categorizations. The thieves mainly succeeded in breaking into the museum because they mastered, knowingly or not, the sociology of appearance, using the categories of normality as a tool. So, before perfecting our machines, we should first learn to question our own outlook.
