NtechLab, an AI startup out of Russia, is building an AI system which plans to use Machine Learning algorithms to detect and label the ethnicity of people that are run through its facial recognition software. According to a secondary source, the company had the ability to classify people with terms like “European,” “African,” and “Arabic,” as well as with other ethnicities. Apparently, due to some public outcry, any mention of this has been taken off of the internet. The first thing that naturally comes to mind is the large, negative implication which can be drawn from this that such a system would merely automate racial profiling. If it fell into the wrong hands, then nefarious groups could use it to identify whether people belonged to certain ethnicities that they wanted to discriminate against.
Related to this, a highly similar negative implication also exists with in connection with voting. If political groups got a hold of this sort of software, then they could possibly use it to target campaign resources against certain groups. The precise product or platform, from which all of these implications have been drawn is called FindFace. According to NtechLab’s main website, it’s already operational, even at the enterprise level.
It’s not just your face and your ethnic group
According to a piece by Mashable, FindFace uses the same Deep Learning algorithms to identify people’s age, gender and yes, emotions.
How does this work?
It all starts with FindFace taking in a photo, which needs to be of someone’s face, and then generates a very specific set of 160 numbers from it. These numbers are supposed to serve as the data that the system uses to describe the face. Therefore, they should include traditional parameters, meaning anything that serves to provide a basic list of outward features like hair color, eye color and whether the person has facial hair or not. Beyond this, this list seems to also include apparent age, gender and finally, the emotions that the person is exhibiting at that time.
Benefits?
NTechLab says that the biggest benefit of Find Face is what it can do for law enforcement agencies. With self-stated capabilities like searching through a billion pictures in around half a second in addition to boasting of achieving only 7% error while identifying the faces in 10,000 pictures. This 93% accuracy rate hasn’t been proven. Even so, FindFace has been recorded in a competition with Google, Beijing University and others as having achieved 73% accuracy while identifying a database of 1 million pictures. Supposedly, other companies have not yet hit this level of precision with similar systems.
As far as the unique differentiator of being able to classify pictures based on the emotions in them, NtechLab has clarified the suggestion that this feature is fully operational by saying that the system is still being trained to do it accurately. In other words, the skill is operational and being used live, but the FindFace team is consistently using training data based on the system’s past performance in order to improve its accuracy.
Partnerships
FindFace exists as a website and an enterprise application, at the same time. Their website is directly integrated with the Russian version of Facebook, which is called Vkontakte.The issue is that their primary product has already been criticized, due to its apparent skill at revealing the true identities behind anonymized social media profiles. It goes without saying that the privacy implications related to this are relatively endless. Already, more precisely, they’ve come under fire for enabling the exposure of sex workers to public ridicule on Vkontakte. One can only imagine what could happen if the same were done to journalists or political dissidents in highly volatile countries.
Beyond Vkontakte, NtechLab seems to prefer keeping most of its customers secret, though it is public knowledge that the total number of them sits somewhere around 2,000. Included in this number is the connection between them and FIFA’s Confederations Cup, which is happened this in month, in Russia.
Some have extrapolated that this as well as the fact that FindFace exists on a public website, means that the Russian government has no need for an enterprise version. Despite this, there has been speculation that FindFace did make its way to the 2018 World Cup. As of now, however, this is simply speculation.
Looking ahead
As of now, controversy still remains around FindFace and NtechLab. Their given explanation for including an ethnicity identification feature appears to be that it could help law enforcement agencies fully identify suspects in crimes. To this, they’ve added the idea that the full platform could even be useful in the job interview process, in the sense that it could provide the employer with a full picture of each candidate, including his or her emotional responses in certain situations. On the other side of things, lies the fact that the FindFace website which was supposed to be consumer facing, does not seem to work. In truth, it seems to have been shut down as when it is accessed with the link below, one is met with a “404 Forbidden,” error.
It is also possibly important to say that the articles on the company blog are written as if they are for people who have no technical experience at all. What this comes down to is, as of yet, fairly unclear. One can only hope that the World Cup or an upcoming similar major partnership did clear up some of this obscurity but at this point, from an ethical standpoint, it might be wise to steer clear of such an endeavor.
References:
Primary Source: https://www.artificialintelligence-news.com/2018/05/25/russian-ethnicity-detecting-ai/
Startup Website:
FindFace Website (supposedly):
FindFace Business Platform:
FindFace Company Blog: