Facebook Disabled Network Of Accounts Using 'Inauthentic Behavior' And AI-Generated Faces

22/09/2020

Artificial Intelligence has become so smart that it can imagine things that don't exist.

From fursonas that don't exist, the nudes that don't exist and a lot more, people have found that computers can indeed have some wild imaginations.

And that includes imagining faces of people that don't exist.

While it can be fun to see faces of people that were never been born, on the web, fake faces can mean a lot of trouble.

One of which, is the ability for malicious actors to impersonate non-existing people to spread hatred or propaganda, and get away with it.

Fake people's faces
Nine profile pictures, some of which were cropped or had stickers added (left). All nine images rendered opaque and superimposed to show the alignment of the eyes, an indicator of synthetic image generation (right). (Credit: Graphika)

This is what Facebook, the social media giant of the web, is trying to take care of.

The company has taken down a China-based network of fake accounts that used AI-generated faces to spread government propaganda across the platform, according to research by analytics firm Graphika.

On its news page, the social network announced that it had removed 155 accounts, 11 Pages, 9 Groups, and 6 Instagram accounts for "coordinated inauthentic behavior on behalf of a foreign or government entity."

Most of the fake accounts in question were based on Southeast Asia, where malicious actors were posting events that were based from Beijing’s interests in the South China Sea and developments in Hong Kong.

Facebook also found that a smaller number of fake accounts were posing as Americans citizens posting content that both supported and criticized then-presidential candidates Pete Buttigieg, Joe Biden, as well as Donald Trump.

Before the takedown, Facebook asked Graphika to analyze the data.

After doing reverse image searches, the company found that the accounts were using stolen profile pictures of real people, as well as fake faces created by AIs.

Fake accounts Facebook
One of the posts Facebook suspected as part of a coordinated inauthentic behavior in the China and Southeast Asia regions. (Credit: Graphika)

According to Graphika on its report:

"One way that the operation tried to get around that risk was by using profile pictures that appeared to have been generated by the form of artificial intelligence known as Generative Adversarial Networks (GAN). This form of AI is readily available online, and its use (or abuse) by covert operations has exploded in the last year."

The company found 12 profile pictures that it suspects were AI-generated, due to their distorted backgrounds and asymmetrical peripheral features such as ears, glasses, and hair.

Nine of the fake profile pictures had a perfect alignment of the eyes, which is “an indicator of synthetic image generation,” according to Graphika.

Graphika spotted some of the AI-generated fake faces by rendering the images opaque, to then superimposing them on top of each other to expose the alignment of the features.

The firm suspects that the malicious actors behind the scheme cropped the photos and have included some stickers to disguise the fact that they are fake.