The term “deepfake” is a combination of the words “deep learning” and “fake.” Deepfake technology effectively allows anyone to create hyper realistic pictures, videos, and even audio clips that make people appear to do or say things that they did not actually do or say.
In the year 2020, deepfakes of popular celebrities, politicians, and other icons took off, including convincing deepfakes of former President Barack Obama and Facebook co-founder Mark Zuckerberg.
These deepfake videos are often created through a process called generative adversarial networking (GAN). The GAN procedure uses two separate neural networks with one model working to build the deepfake and the other model working to detect the counterfeit media. These models work iteratively against each other until the second model is no longer able to distinguish the deepfake media from the “real” media. At that point, it can be assumed that the deepfake will also be able to fool any human who might see it.
The GAN technique works especially well on aforementioned famous personalities since they can provide the most data for the neural networks to analyze. With so many recordings and footage on these people, GANs can create more realistic and accurate deepfakes.
Another method of media manipulation takes the form of “shallowfakes.” Unlike deepfakes however, shallowfakes refrain from using any artificial intelligence. Instead, shallowfakes use basic video or photo editing software to manually alter pre existing media. However, this process takes significantly longer to produce realistic forgeries when compared to using deepfake methods.
This means that deepfakes have an arguably greater potential to quickly produce manipulated media. In the fields of cinema or gaming, deepfakes make it possible to use the faces and expressions of real actors without having to film the actor on set for extended periods of time. This system can also be applied to audio recordings, with an artificial voice actor to say anything at any time without needing to rely on pre-made scripts.
Unfortunately, deepfakes also have adverse implications. Deepfakes created with malicious intent have the ability to corrupt a person’s reputation and spread disinformation. This can be particularly dangerous in the context of politics. In fact, in California, sharing falsified media about a politician within 60 days of an election is a cybercrime.
Still, even with laws and policies against harmful deepfakes, it is difficult to track down forgeries and enforce these kinds of rules. As deepfakes become more advanced, experts believe that “perfect” digitally manipulated videos may be less than a year away.
So, how can you spot a deepfake? Previously, it was easier to spot deepfakes when programs were still evolving their animation capabilities. The result was that counterfeit subjects either blinked too frequently or not frequently enough. Since then, a published study about the flawed blinking from the University of Albany encouraged the emergence of new deepfakes that no longer had this issue.
Now, the main features you should look for are limited to lighting or focusing issues in regards to the particular subject. Deepfake algorithms reuse the same lighting from the clips provided in the GAN modeling which does not always match the lighting in the target video. This can cause the lighting to look warped or unnatural. In the same way, if the subject’s skin appears less focused than the environment that they are in, it may be a hint that the video is a forgery.
New deepfake detection software has also been important to combat the wave of hostile deepfakes. New technologies such as Operation Minerva, work to detect deepfakes by checking to see if a video has been blended with other videos. Many tech companies have also hosted deepfake detection challenges with incentivizing prizes to look for more innovative solutions.
Overall, deepfakes have the potential to create utter chaos if detection software cannot keep up with advancing deepfakes. The way we approach the deepfake dilemma will ultimately decide our ability to secure an anti-misinformation world.
“Deepfake Technology: What Is It and How Does It Work?” Techslang, Techslang, 25 Jan. 2021, www.techslang.com/what-is-deepfake-technology/.
Johnson, Dave. “What Is a Deepfake? Everything You Need to Know about the AI-Powered Fake Media.” Business Insider, Business Insider, 22 Jan. 2021, www.businessinsider.com/what-is-deepfake.
Toews, Rob. “Deepfakes Are Going To Wreak Havoc On Society. We Are Not Prepared.” Forbes, Forbes Magazine, 26 May 2020, www.forbes.com/sites/robtoews/2020/05/25/deepfakes-are-going-to-wreak-havoc-on-society-we-are-not-prepared/?sh=1f1dd8b27494.
“What Is The Difference Between A Deepfake And Shallowfake?” Deepfake Now, 21 Apr. 2020, deepfakenow.com/what-is-the-difference-between-a-deepfake-and-shallowfake/.