Patent Awarded To Clearview AI For Its Innovative, Web-Scraping Fuckery
from the here's-hoping-actual-bankruptcy-will-swiftly-follow-Clearview's-moral-ba dept
Clearview AI has begun travelling the long road back to semi-respectability.
Over the past 18 months, Clearview has gone from unknown to pariah, rejected even by its in-group. It turned web scraping into a surveillance tool, providing everyone from billionaires to repressive regimes with a tool that leverages 10 billion images scraped from the web to create a comprehensive facial recognition system.
Clearview has attempted to reform since gaining international infamy. It claims to have ended all sales to private entities, including retailers and potential VC investors. (Caveat: It has provided no documentation to back this claim up.) It has finally submitted its AI to examination by the National Institute of Science and Technology and received one of NIST’s highest accuracy ratings. (Caveat: facial recognition tech is inherently flawed, no matter how accurate it is.) And it has pulled out of a handful of countries where its scraping of the web was found illegal. (Caveat: It has done so only after being placed under investigation and continues to dispute these findings.)
What’s next for Clearview and its slow escalation towards begrudging acceptance by governments that haven’t banned it yet? Why, it’s some IP protection for its AI and its process for matching faces found in Clearview’s web-scraped haystack. Alexandra Levine of Politico has the story on Clearview’s acquisition of at least one form of government approval.
Clearview AI has gotten the green light on a federal patent for its facial recognition technology — an award that the company says is the first to cover a so-called “search engine for faces” that crawls the internet to find matches.
[…]
The U.S. Patent and Trademark Office sent Clearview a “notice of allowance” on Wednesday, meaning the patent will be approved once the company pays certain administrative fees. The patent covers Clearview’s “methods of providing information about a person based on facial recognition,” including its “automated web crawler” that scans social networking sites and the internet and its algorithms that analyze and match facial images obtained online.
Congratulations, I guess. There’s plenty of prior art out there, but Clearview does indeed have a unique product. It’s not that it created facial recognition AI that performs in a novel way. It’s that it decided it was OK to use the entire web as a source for images and personal information, bundled this all up, and sold it to whoever wanted it. That’s the “creative” aspect of it. This is a patent acquired by being a shit-heel — doing something no other facial recognition company ever considered doing.
The unique aspect — the one deserving of a patent — is the most problematic aspect of the company’s AI. And with this in its hip pocket, Clearview has now obtained federal government protection for a product that violates laws both domestic and foreign.
For that reason, the only congratulations Clearview will receive are ones it bestows on itself. And it already has. The head of the company somehow views this as a vindication of his godawful invention — one that relies on the world’s population to fill its database. Here’s the CEO patting himself on the back for coming up with something universally abhorred.
“There are other facial recognition patents out there — that are methods of doing it — but this is the first one around the use of large-scale internet data,” Clearview CEO and co-founder Hoan Ton-That told POLITICO in an exclusive interview. The product uses a database of more than 10 billion photos, Ton-That said, and he has emphasized that “as a person of mixed race, having non-biased technology is important to me.”
To the first half of that statement: kindly fuck off. That’s not an accomplishment. That’s a symptom of sociopathy.
And to the second half, may I ask what even the fuck are you talking about? The AI will be biased. There hasn’t been one created that isn’t. All facial recognition tech tends to do better accurately identifying white male faces. And operator bias will take care of the rest.
Cops will inject their own biases when performing searches and odds are most faces searched by government agencies will be of minorities. Investigators will be free to impose their biases on the search results, discarding those that don’t point toward their favored suspects and accepting those that align with the narrative they’ve concocted. The tools may change but policing doesn’t. The only thing Clearview adds is the possibility of cop shops avoiding local restrictions on social media monitoring software use and the ability to engage in their biases at unprecedented scale.
Hats off, Clearview. Enjoy your 20 years of IP protection. Chances are, you won’t even need it. Nobody else has tried to emulate your web scraping ways. Maybe that will change over the next couple of decades, but at the moment, you stand alone in your unwavering shamelessness.
Filed Under: facial recognition, patents
Companies: clearview
Comments on “Patent Awarded To Clearview AI For Its Innovative, Web-Scraping Fuckery”
I am at a loss to describe the PTO's new example of idiocy
So does this mean I can patent fishing in the whole ocean and selling my fish to whoever will buy them?
Not only is there prior art, anything they do is also obvious to anyone practicing said art. Mostly obvious that we don’t do that kind of thing, at least not yet.
Welp. Maybe if they go bankrupt, they can become and NPE and sue everyone else.
Can you explain like I'm 5?
The AI will be biased. There hasn’t been one created that isn’t. All facial recognition tech tends to do better accurately identifying white male faces.
This is a serious question, I’m not trolling anybody. I’m not tech savvy enough to know how these programs work, other than what I read that they really don’t. So are you saying that an unbiased AI is objectively impossible to create? Or that it just hasn’t been done yet?
Re: Can you explain like I'm 5?
I’m wondering about this too. If Clearview’s AI was trained using the faces of the entire world it’s probably better in China than in the US/Europe.
Also, thinking of China buying tech from Clearview just made me gag a little bit.
There’s plenty of prior art Re searching images photos images of people’s faces Maybe the patent office should not be giving patents to company’s whose whole purpose is to spy on people who have not given them permission to acess personal data and which has broken privacy laws in many western country’s these programs are more likely to be used by governments who wish to target minoritys protestors or human rights activists
At least with Facebook or instagram you have given them permission to acess your data and store your personal photos and there’s privacy options you can use to make posts or images private to friends or family
and you can close down your account anytime you want to
Hopefully the EU will make it clear it will not give patents to company’s whose programs or ai process data and violate gdpr regulations and collect personal data without asking permission from anyone
My Scraped Picture
I use a picture of a Kangaroo 100% of the time on any web site that I use. Clearview can hop right up and scrap my photo anytime. What goes on in the pouch, stays in the pouch 🙂
Not entirely accurate about white male faces
It isn’t intrinsic to the design like the infamous hand dryers which could not see people with darker skin because it relied upon a type of light reflection found mostly in lighter skin. It is about the training set. White male faces may be the predominant western facial recognition bias but that is not the case with Chinese facial recognition products. Unsurprisingly it is biased towards Chinese faces and performs worse on its relative rarity of caucasians.
It is a matter of how it is taught. The tech winds up resembling human brains in that when exposed to something less familiar they really do look all alike to them until they receive a greater level of exposure to notice the differences.
Accuracy rates are heavily affected by frequency of occurrence. Say you were to give me, a complete novice stacks of X-Rays from a large hospital, and ask me to circle all of the cancerous tumors in the slides. Then doing nothing would give me the best results because I don’t know how to identify specifically cancerous tumors but I do know most people do not have tumors. This sort of principle also explains why it winds up biased – it gets the most overall accuracy that way because minorities do not count as much in terms of overall accuracy they get deprioritized.