All of the mistaken arrests due to facial recognition by US police are of black people.
If the prospect of the use of facial recognition by the police may already seem dystopian in itself, the latter is moreover anything but infallible. And the statistics really leave no room for interpretation: when it comes to police errors due to facial recognition, the victims are always black.
Facial recognition used daily in several cities across the country
Last February, Porcha Woodruff, a black woman living in Detroit, was preparing her children to take them to school when six police officers showed up at her house, accusing her of stealing a car. Eventually released, she decided to press charges, becoming the sixth person (at least) to have been wrongly accused, based solely on the findings of the facial recognition technology used by police to make her arrest. Like her, the five other people who were victims of this type of error were all black.
Contacted, the Detroit police chief acknowledged that the problem was very serious, and that his department would conduct an internal investigation. The mere fact that the technology is being used raises questions, not just philosophically, but purely from an efficiency point of view. Because in 2020, this same police chief had recognized that facial recognition, when used alone, was wrong 96% of the time. This did not prevent the Detroit police from using the technology 125 times in 2022. But in the field, it is that of Baltimore which leads the race, with no less than 800 searches using facial recognition that same year.
For activists and experts, this is anything but a surprise
But for associations for the defense of public freedoms, just like for AI researchers, this type of drift is anything but a surprise, and they have been warning about the issue for years already. For them, if the technology is not racist in itself, it is on the other hand developed by mainly white engineers, and therefore built with the same biases as the latter. This is manifested in particular by the virtual absence of black people in the databases used to train artificial intelligence, significantly increasing the risk of error when they are the subject of research using these technologies.
But for these researchers, the perception of these technologies as useful by the police, which adds to their own racist biases, further amplifies a deeply unjust system that makes any challenge by the victims of these errors almost impossible.
Remember that facial recognition will be used during the Paris Olympics and at least until 2025 in France.
Source : Insider, Wired
11