At the end of last year, San Francisco facial recognition start-up Everalbum has won a $ 2 million contract with the Air Force to provide “AI-based access control.” On Monday, another branch of the US government inflicted a setback on the company.
The Federal Trade Commission said Everalbum had agreed to pay the fee whereby it applied facial recognition technology to images uploaded to a photo app without users’ permission and kept them after telling users they would be deleted. The startup has used millions of photos to develop technology that is offered to government agencies and other clients under the brand. Paravision.
Paravision, as the company is now known, has agreed to delete data inappropriately collected. But he also accepted a newer remedy: purge everything algorithms developed with these photos.
The regulation casts a shadow over Paravision’s reputation, but chief product officer Joey Pritikin says the company can still fulfill its Air force contract and obligations to other clients. The startup shut down the consumer app in August, the same month it learned of a possible FTC complaint, and it launched facial recognition technology developed without app data in September. Pritikin says these changes were in motion before the FTC hit, in part due to “changing public sentiment” about facial recognition.
FTC Commissioner Rohit Chopra, a Democrat, issued a Monday statement praising the rigor of the commission with Paravision, claiming that she was rightly forced to “renounce the fruits of her deception”.
He opposed the settlement to a 2019 deal in which Google paid $ 170 million for illegal data collection children without parental authorization. The company was under no obligation to delete anything derived from this data. “Commissioners previously voted to allow data protection law violators to retain algorithms and technologies that derive much of their value from poorly obtained data,” he wrote. “It’s a significant course correction.”
Ryan Calo, a law professor at the University of Washington, says demanding that Paravision do away with facial recognition algorithms formed with allegedly poorly obtained images shows the FTC recognizes just how much the rise of machine learning closely related data sets and potentially dangerous software products.
Tech companies used to make software just by paying people to tap the right keys in the right order. But for many products like facial recognition models or video filtering software, one of the most crucial ingredients now is a carefully curated collection of sample data to feed into machine learning algorithms. “This idea, you have to remove the model and the data recognizes that these things are closely related,” Calo says. Facial recognition systems deserve special consideration because their creation requires very personal images. “They’re like Soylent Green – made of people.
David Vladeck, former director of the FTC’s Office of Consumer Protection and Georgetown law professor, said Monday’s regulation was in line with precedents that required the deletion of data. In 2013, software publisher DesignerWare and seven rental-purchase retailers have agreed to delete geolocation data collected without the consent of spyware installed on laptops.
Monday’s more expansive removal requirement with Paravision was approved unanimously, 5-0, by the FTC, which is still controlled by a Republican majority. Following President-elect Joe Biden’s inauguration this month, the committee could become a majority Democrat and potentially even more keen to control tech companies. He could get new support and resources from Democrat-controlled Congress.
Calo hopes the agency will gain more technical resources and expertise to help it scrutinize the tech industry on a level playing field. One use for more tech know-how might be to devise ways to check whether a company has really cleaned up not only poorly gotten data, but also the benefits or technologies that come with it. This might be difficult to do for systems involving complex machine learning models built from multiple data sources.