Clearview's Facial Recognition AI One Of The Best In The Business, Says NIST
from the man-but-what-a-business dept
Some good news for Clearview, the bottom dweller of the facial recognition field. The prodigious scraper of web content has finally submitted its algorithm (the one it actually sells to government agencies) to the National Institute of Science and Technology (NIST) and has obtained a score that justifies its frequent blustering about how accurate its AI is. Here's Kashmir Hill with the details for the New York Times.
In results announced on Monday, Clearview, which is based in New York, placed among the top 10 out of nearly 100 facial recognition vendors in a federal test intended to reveal which tools are best at finding the right face while looking through photos of millions of people. Clearview performed less well in another version of the test, which simulates using facial recognition for providing access to buildings, such as verifying that someone is an employee.
This vindicates CEO Hoan Ton-That's heretofore unproven claims that Clearview's AI is one of the best in the business. Prior to this, the only data backing Ton-That's assertions came from an internal test performed by a non-independent entity. Now, Ton-That can add the NIST results to his company's sales pitches, which may make it a bit more palatable to government agencies on the fence about shelling out tax dollars for a sordid little product with an unproven AI. Its previous submission to NIST tested its one-to-one AI (something used to unlock phones, for instance) -- a feature Clearview does not use and does not market to customers.
But this is hardly a vindication of Clearview as a whole, much less its business model. Scraping the open web for images is a really shady way to generate a database -- one that now holds more than 10 billion images. The more hay you put in the haystack, the greater the chances of confusing hay for needles. And these results don't change what's already wrong with all facial recognition tech: the fact that it's always worse when it comes to correctly identifying anyone other than white males.
It also won't make Clearview's web-scraping ways any more acceptable or legal. It has been evicted from two countries (Canada and Australia) for violating national privacy laws. It is being sued in two US states for the same reason. And it has become a pariah in an industry full of morally-suspect tech purveyors -- a company so unseemly even other bottom feeders marketing biased tech to cop shops refuse to have anything to do with it.
And it won't make CEO Hoan Ton-That any less full of shit. This test may have proven Clearview has the AI chops to run with the big boys, but it doesn't change the fact that his claims about helping law enforcement clear cases have been continually rebutted by the law enforcement agencies he says relied on his tech to find dangerous criminals. As long as Clearview remains solvent, it is a threat to people's freedoms and safety. If the NIST test results in more customers for Clearview, it will increase the chance someone will be misidentified and put at the mercy of people with a whole lot of guns and power.
Thank you for reading this Techdirt post. With so many things competing for everyone’s attention these days, we really appreciate you giving us your time. We work hard every day to put quality content out there for our community.
Techdirt is one of the few remaining truly independent media outlets. We do not have a giant corporation behind us, and we rely heavily on our community to support us, in an age when advertisers are increasingly uninterested in sponsoring small, independent sites — especially a site like ours that is unwilling to pull punches in its reporting and analysis.
While other websites have resorted to paywalls, registration requirements, and increasingly annoying/intrusive advertising, we have always kept Techdirt open and available to anyone. But in order to continue doing so, we need your support. We offer a variety of ways for our readers to support us, from direct donations to special subscriptions and cool merchandise — and every little bit helps. Thank you.
–The Techdirt Team
Filed Under: facial recognition, nist
Companies: clearview
Reader Comments
Subscribe: RSS
View by: Time | Thread
" In all cases, each person is enrolled with one image only."
That and I lack the will to keep trying to figure out the racial demographics for the participant images.
Hey when there is only 1 picture of the person in the study they can identify that person really well.
[ link to this | view in chronology ]
Now just waiting
for the AT&T-Murdoch-Clearview-NSO mega-merger.
[ link to this | view in chronology ]
If they're the best...
Then how bad are the rest of them? Never mind, I don't want to know
[ link to this | view in chronology ]
Re: If they're the best...
been trying to do this since the C64.
[ link to this | view in chronology ]
It's all relative...
If I have several steaming piles of crap in front of me to choose from, which is the best?
[ link to this | view in chronology ]
Re: It's all relative...
If you're shopping for manure then you want something that doesn't eat meat... theses fellows eat meat, so their crap won't be great as manure :p
[ link to this | view in chronology ]
why?
[ link to this | view in chronology ]
Re: why?
Am I reminded of the china article.
[ link to this | view in chronology ]
Perhaps it would be productive to examine the test and find out what was actually tested, at what success rate, and with what methodology. How do we know this doesn't just indicate that no algorithm was successful but they failed less badly then the others. Successful ID of 1 out of 100 when all others were 0 of 100 would still be a number 1 rating after all.
[ link to this | view in chronology ]
Re:
And the truth is.
They arent allowed to use your phone data without a warrant and justification.
Where with camera's all over the place. they would be in charge of the camera so would not need one.
This is similar, to the cops having s Lic. Plated reader that reads 1000 per min. And the cops wonder around, and Find the people they are looking for, by the use of Their own cars. they can correlate Where you have been by looking into a data base and the most probable locations you WILL be.
And if you are into Drugs, they can use it to find all your dealers.
Or didnt you know about this? its in use already.
[ link to this | view in chronology ]