The American start-up Clearview AI has protected one of its technologies by filing a patent with the United States Patent and Trademark Office (USPTO). This is its method of collecting and preparing training data for its facial recognition algorithms.
Publicly available data
Hoan Ton-That, CEO of Clearview AI, welcomed this patent claiming that his technology has the ability to “create accurate and unbiased facial recognition algorithms from publicly available information“. However, it is precisely because the data is collected in this way that the company is criticized by several regulators.
Clearview AI markets itself as a provider of facial recognition technology for law enforcement agencies, like law enforcement, allowing them to easily track down a suspect. This is a kind of search engine to identify a person from whom they have an image from, for example, a CCTV recording.
It works entirely thanks to the data available on social networks, professional sites containing photographs of their employees as well as blogs. Images are also taken from videos available online, for example on YouTube. This collection concerns images of adults and minors, no filter being applied in this regard. Only hundreds of URLs, associated with “adult” sites with the largest audiences, are blocked and excluded from collection.
Being able to identify “almost everyone”
In February 2022, the start-up said it was on track to achieve “the 100 billion” of facial photographs in its database within a year with a clearly stated and assumed goal: to be able to identify “almost everybody“.
However, for several regulators, this collection is illegal since the people appearing in the photographs and videos have not consented to this reuse. Recently, it was the Greek equivalent of the National Commission for Computing and Liberties (Cnil) which imposed a fine of 20 million euros on the company. In addition to this sanction, she ordered him to immediately stop sucking up the personal data of Greeks and to delete those collected so far. A similar procedure is still pending before the French Cnil, which had given it formal notice in November 2021 to cease the collection and use of public photos and videos sucked from the Internet.