Untold amounts of photos are uploaded to Facebook every day. Most of these photos have pictures of you or others in them, and Facebook’s DeepFace AI program is matching these photos with a 97.25 percent accuracy rating. The potentially disturbing fact is that even if you are not tagged in the photos, DeepFace is still taking note of you.
Moral implications aside, DeepFace is an impressive bit of engineering. A technique called Deep Learning is what powers the DeepFace AI tech. While the ability to detect a face in picture is not new technology, the process of identifying it is much more complex. The great majority of human faces include the same things: two eyes, a nose, mouth, and so on. The difficulty comes when our faces change by smiling or contort into a different expression. Using a library of 13,000 photographs taken from the internet called the Labeled Faces in the Wild (LFW) library, engineers have been able to use this celebrity filled database of photos as a benchmark to fine tuning their facial recognition software.
In a feature on Science, a series of photos of lead engineer Erik Learned-Miller is shown that has a variety of lighting situations and poses. While the human eye can tell instantly that it is the same person, a computer has a hard time determining this — and that is where DeepFace comes into play.
“We revisit both the alignment step and the representation step by employing explicit 3D face modeling in order to apply a piecewise affine transformation, and derive a face representation from a nine-layer deep neural network. This deep network involves more than 120 million parameters using several locally connected layers without weight sharing, rather than the standard convolutional layers.”
In layman’s terms, the DeepLearning program creates a 3D model of your face and the more photos that are available of you online, the more accurate the model. Once the model is created, DeepFace can adjust that model to match almost any lighting situation or facial contortion with a 97 percent accuracy rating. The creepy part, however, is that Facebook is also using another library to train its Deep Learning algorithms, the photos you upload to Facebook on an almost daily basis. Buried within the abstract publication detailing the system is a library called SFC or Social Face Classification. In this library are 4.4 million tabled faces from 4030 Facebook users.
The Inquisitr has reported previously how Facebook has been cited as the source for a rise in divorces, as couples are finding incriminating photos. It does not take much imagination to see how a technology like DeepFace AI raises troubling issues for privacy both in the digital and physical worlds.
Science fiction has always been a creepy prophetic vision of the future. Minority Report, 1984, and Terminator showcase what can happen when the presented corruption and hubris of mankind is personified in killer robots or totalitarian regimes. Often times, these instances are shown as comical plot devices that usually go overlooked. Guardians of the Galaxy featured a scene where Rocket is scanning the faces of people in a public square to see who is tagged for a bounty. While this was used to advance the plot, DeepFace is something that operates on the same principal and what can be a reality when that technology is made available to advertisers or government agencies.
The race to increase our technological proficiency is something that appears akin to the race for production during the Industrial Revolution. The results then were work houses, orphans, and high crime throughout Victorian England. Prominent scientists like Hawking and Musk have both been on the record as saying that AI development would do well to take into consideration a quote from Nicholas Meyer in Star Trek VI,“Let us redefine progress to mean that just because we can do a thing, it does not necessarily mean we must do that thing.”
[Image Source: Shutterstock]