Would you buy a used face-scraped dataset from Hoan Ton-That?
The only telling moment in a recent TV interview of face biometrics innovator Clearview AI CEO Hoan Ton-That is when the interviewer asks him who his major investors are.
“They just don’t want to be named at this point, and we want to protect their privacy. That’s important to us,” Ton-That says through the smirk he has worn throughout the Q&A.
That is from the founder of a company that has collected 3 billion photographs of people, almost all scraped from public internet sites without anyone’s permission.
Ton-That almost gleefully explains that those people have no right to privacy because they posted their images before anyone could dream of having their faces fed into a global police lineup.
Nor do social media platforms — a major if not the major source of Clearview AI’s photos — have a case against the company for face-scraping their sites, he said, despite the fact that Clearview AI broke data-abuse provisions of terms of service contracts that it agreed to when its crawlers entered the posting sites.
The CEO also said that Clearview supports responsible facial recognition use with its policies, and would support regulation in the space. Further, Clearview’s massive training dataset has enabled it to develop face biometrics that work equally for different demographics, according to Ton-That.
But that is not the big moment.
That comes when Ton-That’s nose is rubbed in the hypocrisy of refusing to divulge investors — which is his right to do. He probably signed a contract prohibiting it. But context is key.
The interviewer, Hari Sreenivasan of the show Amanpour & Co., points out the absurdity of Ton-That’s position. Some people, who happen to have millions of dollars to invest, deserve privacy. The people who are the product do not?
Through a grin that hints that something special has just been eaten, Ton-That says: “OK. Sure. I know.”
Playing cute like that plays well on TV, but it raises an enormous question about Clearview AI and every other biometrics company, government agency and entrepreneur developing or deploying facial recognition systems.
Why should anyone take them at their word? In the absence of meaningful regulation or full transparency, what has any of them done to be trusted crossing a street with the traffic light much less to profit from the public’s biometric data?
Sreenivasan asked multiple times why anyone should take him at his word that Clearview AI will never, ever sell services to anyone but law enforcement agencies (excluding those in the surveillance states of China, Russia, Iran and North Korea). Or that its algorithms would be used with augmented reality glasses to identify people in real time.
The slim coder asks that people take him at his word; he and his company will be transparent.
But the company did sell or attempt to sell to companies in the recent past, registering accounts with Macy’s, Walmart and the National Basketball Association.
Sometime after that, Ton-That and his investors realized continuing to do so would, in his words, make a world they would not want to live in, and retreated to public service (for profit). And just last month, a Clearview AI patent application filed last August surfaced at BuzzFeed News that seeks to protect intellectual property for identifying people in real time.
The document mentions scenarios including dating, offering subscribers personal information on strangers like nationality, contact information and hobbies.
That application was made months after Ton-That publicly stated that he would only work with law enforcement, according to BuzzFeed News.
Asked if Clearview AI has a “god view” that the company can use to see what users are doing, he jogs a little. Each police department will have an audit tool to keep watch over “investigators.”
Who watches department supervisors? Ton-That assures Sreenivasan that he feels comfortable police will adhere to relevant policies.
If some find that a surprising assertion today, he falls back on the cliché that facial recognition is finding the most despicable criminals — that is, suspects — every day, and reuniting crowds of runaway children.
All of which is doubtless true to some extent, but again, there is no reason to trust these statements at face value.
Ton-That in his interview was coy with a ton (his favorite metric) of empty answers.
For example, he gets “a ton” of email thanking him for making the world better. Like the email praising Clearview AI after a “highway patrol” used the AI to bust a man picking up fentanyl.
All anyone can say for sure at this point is that more and more photos are being scooped up by algorithms every minute, for someone else’s profit.