Clearview AI is expanding sales of its facial recognition software to companies from mainly serving the police, it told Reuters, inviting scrutiny on how the startup capitalizes on billions of photos it scrapes from social media profiles.
Sales could be significant for Clearview, a presenter on Wednesday at the Montgomery Summit investor conference in California. It fuels an emerging debate over the ethics of leveraging disputed data to design artificial intelligence systems such as facial recognition.
Clearview’s usage of publicly available photos to train its tool draws it high marks for accuracy. The United Kingdom and Italy fined Clearview for breaking privacy laws by collecting online images without consent, and the company this month settled with US rights activists over similar allegations.
Clearview primarily helps police identify people through social media images, but that business is under threat due to regulatory investigations.
The settlement with the American Civil Liberties Union bans Clearview from providing the social-media capability to corporate clients.
Instead of online photo comparisons, the new private-sector offering matches people to ID photos and other data that clients collect with subjects’ permission. It is meant to verify identities for access to physical or digital spaces.
Vaale, a Colombian app-based lending startup, said it was adopting Clearview to match selfies to user-uploaded ID photos.
Vaale will save about 20 percent in costs and gain in accuracy and speed by replacing Amazon.com Inc’s Rekognition service, said Chief Executive Santiago Tobón.
“We can’t have duplicate accounts and we have to avoid fraud,” he said. “Without facial recognition, we can’t make Vaale work.”
Amazon declined to comment.
Clearview AI CEO Hoan Ton-That said a US company selling visitor management systems to schools had signed up as well.
He said a customer’s photo database is stored as long as they wish and not shared with others, nor used to train Clearview’s AI.
But the face-matching that Clearview is selling to companies was trained on social media photos. It said the diverse collection of public images reduces racial bias and other weaknesses that affect rival systems constrained by smaller datasets.
“Why not have something more accurate that prevents mistakes or any kind of issues?” Ton-That said.
Nathan Freed Wessler, an ACLU attorney involved in the union’s case against Clearview, said using ill-gotten data is an inappropriate way to pursue developing less-biased algorithms.
Regulators and others must have the right to force companies to drop algorithms that benefit from disputed data, he said, noting that the recent settlement did not include such a provision for reasons he could not disclose.
“It’s an important deterrent,” he said. When a company chooses to ignore legal protections to collect data, they should bear the risk that they will be held to account.”