Why deepfakes are a troubling new trend
Have you, by any chance, seen this “Back to the Future” video featuring Robert Downey, Jr. and Tom Holland? Or, maybe, you even hoped that there’s really a remake in the works?
Don’t get your hopes high — as amusing as this video is, it is not a real thing, but a fake. A deepfake, to be precise.
We’ll talk about what deepfakes are in this blog post.
What are deepfakes?
Deepfakes are a form of artificial intelligence — images or videos created by using a machine learning algorithm (deep learning) to match someone’s face to another person’s body.
Jon Snow apologizing for the disappointing ending to Game of Thrones, Tom Cruise becoming the Ironman, and Sebastian Stan being “recast” as Luke Skywalker in “The Mandalorian” — all of these videos are deepfakes.
How does it work? A deep learning system creates a fake by studying photos and videos of a person from different angles and then mimics their behavior and speech patterns. After that, once a preliminary fake has been produced, it is then made smoother and more believable by generative adversarial networks (GANs), which detect flaws in the fake, leading to improvements addressing the flaws.
After many rounds of detection and improvement, the deepfake video is completed, and the results are almost flawless, as you can see on the videos above.
This technology was created back in 2014, and by now, it is widely used by everyone from academic and industrial researchers to amateur enthusiasts, visual effects studios, and film producers.
You can even create deepfakes on your own on the Deepfakes Web site.
Deepfakes in media
As the deepfake technology has become more widespread and sophisticated, it found quite a lot of applications among the users of the Internet. People use deepfakes for all sorts of mischief, including creating memes, fan videos, or even whole fan movies.
For example, deepfakes allow fans to imagine what their favorite movies would look like with different actors, such as the example with Robert Downey, Jr. and Tom Holland in the “Back to the Future” video from the beginning of this article.
Deepfakes in art
Not all deepfakes are malicious, or simply created for fun. The technology can be used to improve the dubbing on foreign-language films, create amazing animation effects, enliven galleries and museums, and even “resurrect” dead people.
For example, in May 2019, the Dalí Museum in Florida launched “Dalí Lives“, an “art meets artificial intelligence” project dedicated to the artist’s 115th anniversary. The imagery that brings the famous artist to “life” was created by pulling more than 6,000 frames from old video footage and processing them through 1,000 hours of machine learning. This allowed the creators to overlay the source files onto an actor’s face, and then the text was composed of a mixture of quotes from interviews and letters with invented present-day commentary.
This project, as the creators say, is designed to help visitors empathize with the artist and relate to his work.
The most amazing thing here is that this deepfake is interactive. A total of 45 minutes of footage split over 125 videos allows more than 190,000 possible combinations depending on visitors’ responses — including comments on the weather!
The video finishes with Dalí turning around and taking a selfie with the visitors.
Deepfakes in politics
After the deepfake technology emerged, it was only a matter of time before it would target politicians or tackle political goals.
Nowadays, there are lots of videos featuring well-known politicians saying something they never said, appearing drunk when they weren’t, or meeting people they never met.
Will such deepfakes spark any kind of major international incidents? Unlikely. But will they be used to embarrass politicians? Certainly — especially when it comes to elections.
Are deepfakes dangerous?
Making silly videos with faces of well-known people may sound fun, but there are several concerns with the technology, especially as it becomes more and more sophisticated.
For example, deepfakes could mean trouble for the courts, where faked events could be entered as evidence. Besides, they have a vast potential for scammers of all kinds, since they might trick systems that rely on face or voice recognition.
The main threat of deepfake videos is that they, along with other synthetic media and fake news, may create situations where people no longer can — or want to — tell true from false. And when trust starts to erode, it is easier to raise doubts.
The question is, is there some way to tell a real video from a (deep)fake one?
It gets harder as the technology improves and videos look increasingly realistic. There are, however, some telltale signs. For example, some deepfake faces don’t blink normally, the lip-syncing might be off, or the skin tone too patchy or too smooth. There can be flickering around the edges of replaced faces and fine details. The weird lighting and shadows might also be a giveaway.
Besides, there is AI technology that can help detect fake videos, though the technology works best for celebrities because they can train the AI using hours of freely available footage.
Tech companies are now working on detection systems that could find and flag up fakes whenever they appear — but for now, regular users mostly need to rely on being attentive.
Deepfakes make us think: what happens if we can no longer trust our eyes or our ears? For more than a century, not only have audios and videos recorded our history as it was (or so, at least, it seemed), they have also informed and shaped our perception of reality.
But now, more than ever, we need to be attentive to what we see and hear and apply critical thinking to analyze what we see in order to avoid being deceived. Fraudsters are using more and more sophisticated ways to manipulate identities, such as the shadowy disguises of social engineering.
So, when you watch your next video on YouTube, watch closer — it might very well be a fake!
While it can’t detect deepfake videos, Namecheap’s Identity Verification service offers an easy way for you to verify your customers’ IDs, reducing fraud in e-commerce transactions, online auctions, vacation rentals, and more.