A small startup, Clearview AI, has devised a groundbreaking face recognition app. It matches uploaded photos of people to images of that person from the internet. Currently the app is mainly being used by law enforcement agencies.
The People Behind the Clearview App
The Clearview face recognition app was developed by a 31-year-old Australian developer and one-time model called Hoan Ton-That. Ton-That grew up in Australia with his Vietnamese family before moving to the US at the age of 19.
There he met Richard Schwartz, now 61, co-founder of Clearview AI. Schwartz was an aide to Rudy Giuliani when he was mayor of New York and thus has many important contacts. The two met in 2016 and soon decided to go into the facial recognition business together. Ton-That developed the App and Schwartz used his contacts to drum up commercial interest.
Until Clearview, Ton-That was best known for developing an app that let people put Donald Trump’s distinctive hairstyle and color on their own photos. He also developed a photo sharing program, which did not take off.
Clearview, on the other hand, has become very popular with US law enforcement agencies. They use the face recognition app to identify and track down suspects. The company has also licensed the face recognition app to several companies for security purposes. Clearview AI says however, that it is not looking at making the app available to the general public because “There’s always going to be a community of bad people who will misuse it”.
How Does Clearview Work?
Clearview is a simple to use and relatively cheap face recognition app. Its backbone is a database containing more than three billion images and videos. Clearview AI says it scraped the images and videos from social media sites such as Facebook, Venmo and YouTube, and employment sites.
Scraping images is the process of downloading large number of images from websites. Representatives of companies from which Clearview AI scraped the images and video said that their policies prohibit such scraping.
According to Ton-That, however, Clearview AI only use publicly available images. If users restrict their privacy settings on their social media accounts, then these photos will not be scraped. For example, if users change their privacy settings in Facebook so that search engines can’t link to their profiles, then these photos will not be included in the database.
Unfortunately, if photos have already been scraped, then it is too late. Clearview AI keeps all images it has scraped even if these are later removed from a user’s profile. However, Ton-That stated that they are looking at providing a means for people to request their photos be removed from Clearview’s database.
Clearview’s Face Recognition System
Clearview AI perfected the facial recognition AI algorithm by relying on information derived from academic papers.
Clearview AI converts scraped images into mathematical formulas, or vectors, based on facial geometry, such as how far apart a person’s eyes are. Then, the system clusters all the photos with similar vectors into “neighborhoods”.
When a user uploads a photo for matching, Clearview searches the database and retrieves public photos of the matched person, along with links from where the photos were derived. According to the New York Times, “The system … goes far beyond anything ever constructed by the U.S. government or Silicon Valley giants.”
Clearview Renews Privacy Concerns Surrounding Face Recognition Software
There has always been concerns regarding using technology that identifies people based on faces due to loss of privacy and the technology’s possible misuse. As stated by Eric Golman, co-director of the High Tech Law Institute at Santa Clara University: “The weaponization possibilities are endless. Imagine a rogue law enforcement officer who wants to stalk potential romantic partners, or a foreign government using this to dig up secrets about people to blackmail them or throw them in jail”.
Furthermore, such systems raise concerns because they are not 100% accurate. Ton-That admits that Clearview is only about 75% accurate. According to Ton-That, the app does not always work because the photos in the database are taken at eye level. However, the material that police upload is from surveillance cameras usually mounted on ceiling or high up on walls.
Nonetheless, more than 600 law enforcement agencies have started using the Clearview Face Recognition app over the past year. This was before the public became aware of the app’s existence and started to object on privacy grounds.
US law enforcement officers using the app admitted they had only limited knowledge of how Clearview works. Nevertheless, they have successfully used the app to solve shoplifting, identity theft, credit card fraud, murder and child sexual exploitation cases.