Last month, a Russian software developer named Andrey Mima announced that he had been able to track down two women he’d taken a photo of in 2010, with nothing more than their faces. Mima posted on Vkontakte, Russia’s equivalent of Facebook, that he’d used a facial recognition app called FindFace. All he had to do was input the photo and FindFace’s face recognition software did the rest, scanning publicly available Vkontakte photos and finding the two women, so he was finally able to send them the photo he’d taken.
Mima called it “Shazam for people.”
At the end of his post, Mima briefly noted that the service’s implications for privacy were a little alarming. Overall, though, he was writing to recommend FindFace and praise its effectiveness.
At that point FaceFind was about a month old, and in the intervening period the service has proved to be what it sounds like: an unmitigated privacy disaster.
Photographer Egor Tsvetkov took photos of strangers on the St. Petersburg subway, then ran those photos through FindFace, and was able to successfully identify 70 percent of the people, according to RuNet Echo. Tsvetkov hedged on whether or not his project was potentially dangerous, saying, “I think we should wait and see how it develops.”
In early April, it got more dangerous. Posters on Dvach, the Russian variation of the anarchic imageboard 4chan, learned about FindFace and used it to out Russian porn actresses against their will.
FindFace is ostensibly meant to help people make friends. The logic is that if you have a photo of someone, you can find their social media profiles, and learn more about them before trying to become friendly with them; it’s a business model based on second-order Facebook stalking, which is definitely not creepy at all.
FindFace’s founder, Maxim Perlin, told the Russian news site TJournal that the company is “making every effort to protect all Vkontakte users from potential malicious acts.” However, he also admitted there’s little it can do to stop the service being used this way. FindFace didn’t respond to multiple requests for comment for this story.
What makes FindFace so alarming is its effectiveness in making matches, which has little to do with the FindFace team. That high success rate in identifying strangers’ faces is due to the face recognition technology that FindFace is using, which was developed by a little known startup from Moscow called NTechLab.
NTechLab was founded in 2015 by Artem Kukharenko and Alexander Kabakov, a computer scientist and digital communications consultant, respectively. A spokesperson for NTechLab, Masha Drovoka, sent me the company’s origin story, though it’s short on details.
“Artem was in the deep-learning space for more than 10 years, working for Samsung and MSU [Moscow State University] Computer Graphics and Multimedia Lab, when he met with Alexander,” Drovoka explained in an email. “Together they found a huge opportunity to develop a technology that will help find people based just on the photos of their faces.”
They appear to have succeeded. Last December the algorithm developed by NTechLab edged out Google and many other entrants in the University of Washington’s MegaFace face recognition challenge. Presented with a large set —500,000 images of more than 20,000 users— the NTechLab program had a success rate of around 73 percent, Google’s FaceNet program had one just above 70 percent.
The professors in charge of the challenge, Ira Kemelmacher-Shlizerman and Professor Steve Seitz, had never heard of NTechLab before the company entered last year’s competition. They were surprised that the startup did so well, but because of the way the competition works, don’t how NTechLab achieved the results it did.
“We don’t have access to their algorithm,” Kemelmacher-Shlizerman explained via email, “just the features (results of their recognizer on all our images), so can’t comment on their method.”
NTechLab’s answer about how their program works is, unsurprisingly, also vague.
“We have found a special type of internal architecture for neural networks that perfectly fits the face recognition tasks,” they wrote, “then found some special tricks which simplified the process of training of that network, and then we trained our algorithm on 20 million photos.”
FindFace which has been downloaded over 400,000 times, according to NTechLab’s spokesperson, got ahold of the technology so early because the FindFace founders are buddies with the NTechLab founders.
“FindFace was made by our friends,” NTechLab’s Drovoka explained. “We have helped them to launch this project, because that actually was the very first project based on our technology. Now we’re deciding what we are working on closer partnership in the future.”
How NTechLab will vet potential clients, or put safety practices into place, is yet another question that still has only abstract answers, phrased mostly in the future tense.
“We’re going to launch the cloud face recognition software platform, which will be available for every business to plug into and use for their own recognition tasks,” NTechLab told me. “But we will thoroughly monitor its usage schemes and ban those organizations and people, who will try to use it inappropriately. To determine these cases we’re now developing algorithms that will help us detect an inappropriate usage.”
They didn’t elaborate on what an inappropriate use might be, which is worrying since they are working with a technology that already has few rules for its use (at least in the U.S.) and which has already been used to carry out harassment. Russia’s own laws dealing with online abuse are sparse, and the government is all-but-officially behind some forms of mass online harassment itself.
In the meantime, though, NTechLab is hoping to grow, in part with U.S. money. When I emailed with questions Drovoka explained that the company (which lists five staff members, all male, on its site) is very busy with investor meetings in California at the moment. It’s unclear which investors those are exactly, though I did confirm that they interviewed for the startup accelerator Y Combinator.
Asked about the abuse FindFace had already enabled using NTechLab’s face recognition software, the spokesperson implied that ill-treatment of anonymous strangers is simply the price of progress.
“Our technology is revolutionary and actually all such technologies and improvements have both positive and negative effects, depending on who uses it,” Drovoka wrote. “And in our cases it wasn’t just people haunting the porn actresses, as there are more and more positive examples of helpful applications of Findface.”
By way of illustration,Drovoka offered what was presented as a comforting anecdote:
Two pyromaniacs in Saint Petersburg arranged the arson of a building, and the camera record was the only clue. One of that building residents uploaded the video to Pikabu, a Russian online community, which is quite similar to reddit, and asked for help. Soon, the commentators suggested to use FindFace to find the criminals by taking frames from the record, and FindFace successfully identified those people and showed their profiles in the largest Russian social network VK. With that information the residents were able to inform the police, so our algorithm has really helped people.
Hopefully the arsonists weren’t misidentified. NTechLab, after all, has a 27 percent failure rate.