Get Ready for the Terrorist Avatars

By George Michael

After the 9/11 attacks, the U.S. government has shown no compunction about eliminating key terrorists. For example, on September 30, 2011, two U.S. Predator drones launched Hellfire missiles at a vehicle as it traveled on a road in Yemen’s al-Jawf province, killing Anwar al-Awlaki and other al Qaeda operatives. Once characterized as the “bin Laden of the internet,” his pronouncements were broadly disseminated on jihadist websites and social media platforms. Fluent in both Arabic and English, he was regarded as a gifted speaker capable of moving men into action. Although he was believed to have had an operational role in planning some terrorist attacks, his main function seemed to have been that of a “virtual recruiter” who exhorted jihadists to launch lone wolf attacks on their own initiative. Just a few years later, a similar CIA drone strike was carried out against Adam Gadahn in January of 2015, which left him dead. A convert to Islam, Gadahn was also a key al Qaeda spokesman whose propagandistic lectures were popular in the global jihadist subculture.

Less severe sanctions have been brought to bear against domestic dissidents as well. For example, not long after the January 6th storming of the U.S. Capitol, Nicholas Fuentes, was placed on a no-fly list. The year before, his YouTube channel was demonetized. Over the past few years, Fuentes has emerged as perhaps the most mediagenic figure in the American alt-right. Only twenty-three years of age, he has established a sizeable online following for his organization, America First.

As these aforementioned examples illustrate, extremist and terrorist figures are at an increasing risk of repression and/or assassination. New technology, however, could lead to the emergence of dox-free avatars who might prove to be beyond the reach of sanctioning. With the right combination of apps, it is now possible to construct totally fabricated head-to-toe avatars that are realistic and convincing. The applications of these new technologies could prove to be far reaching.

What are Deepfakes?

Photo manipulation was first developed all the way back in the nineteenth century, but new technology has the potential to modify images that heretofore was unimaginable. Around the end of 2017, a Reddit user named “deepfakes” began posting videos in which celebrities’ faces had been swapped onto the bodies of actresses in pornographic videos. More videos followed, including some with the actor Nicolas Cage’s face switched into various Hollywood films. Essentially, a deepfake enables a person in an existing image or video to be replaced with someone else’s likeness.

Deepfakes technology uses machine learning and artificial intelligence (AI) to manipulate or generate visual and audio content which can make the new images look convincingly real. In 2014, a noted AI researcher, Ian Goodfellow, invented what came to be known as a generative adversarial network, or GAN, as a way to algorithmically generate new types of data out of existing data sets. As part of the GAN process, two machine learning systems—neural networks—are trained in competition with each other. One network—the generator—creates counterfeit data, for instance, photos, audio recordings, or video footage, which seek to replicate the properties of the original data set. The other network—the discriminator—seeks to identify the counterfeit data. Based on the results of each iteration, the generator adjusts to create increasingly realistic data.

The two networks continue to compete, perhaps for thousands or even millions of iterations, until such time that the generator produces data so well that the discriminator can no longer distinguish between real and counterfeited data. For example, a GAN can look at thousands of photos of Donald Trump and produce a new photo without being an exact copy of any of them.

Deepfakes build upon previous innovations. For example, in 1997 the Video Rewrite program was introduced which enabled the modification of existing video footage of a person to depict that individual mouthing the words contained in a different audio track. It is believed to be the first system to fully automate this kind of facial reanimation. To achieve this end, the program used machine learning techniques to make connections between the sounds produced by the video’s subject and the shape of the subject’s face.

In that same vein, the Face2Face program, which was released in 2016, modifies video footage of a person’s face to portray him mimicking the facial expressions of another person in real time. More recently, the “Synthesizing Obama” program, which was released in 2017, modified the image of the former president to render him mouthing the words contained in a separate audio track. On that note, in April of 2018, the acclaimed film director Jordan Peele and Buzzfeed released a deep fake of Barack Obama calling Donald Trump a “total and complete dipshit” in order to raise awareness about how AI-generated programs might be used to distort and manipulate reality. Subsequent programs improved upon this concept.

A proprietary desktop application called FakeApp was launched in January 2018, which allows users to easily create and share videos with their faces swapped with each other. This app was superseded in 2019 by open source alternatives such as Faceswap, DeepFaceLab, and web-based apps such as DeepfakesWeb.com. Not long thereafter, companies began using deepfakes for corporate training videos with avatars to create personalized videos. Similarly, the mobile app giant Momo created an application called Zao, which allows users to superimpose their faces on television and movie clips using only a single picture.

It is now possible to generate completely artificial visages that are indistinguishable from actual human faces. In 2019, Philip Wang, a software engineer who worked for Uber, created an artificial intelligence based website called “This Person Does Not Exist.” Using a program called StyleGAN, the site produces extremely realistic portraits of completely simulated people. The site soon gained widespread popularity; two weeks after he posted about the site on a Facebook group, about 8 million people had visited it. Reflecting on its implications, Wang conceded that this type of technology could be both revolutionary and dangerous.

 

About the Author

George Michael received his Ph.D. from George Mason University’s School of Public Policy.  He is a professor of criminal justice at Westfield State University in Massachusetts.  Previously, he was an associate professor of nuclear counter-proliferation and deterrence theory at the Air War College in Montgomery, Alabama.  He teaches courses in terrorism, homeland security, and organized crime. He is the author of seven books: Confronting Right-Wing Extremism and Terrorism in the USA (Routledge, 2003), The Enemy of my Enemy: The Alarming Convergence of Militant Islam and the Extreme Right (University Press of Kansas, 2006), Willis Carto and the American Far Right (University Press of Florida, 2008), Theology of Hate: A History of the World Church of the Creator (University Press of Florida, 2009), Lone Wolf Terror and the Rise of Leaderless Resistance (Vanderbilt University Press, 2012), Extremism in America (editor) (University Press of Florida, 2014), and Preparing for Contact: When Humans and Extraterrestrials Finally Meet, (RVP Press, 2014). In addition, his articles have been published in numerous academic journals. He has lectured on C-SPAN2’s BookTV segment on six occasions and once on C-SPAN3’s Lecture in History program.

 

---