Deep Learning is getting so mature that now it can be used to “fake” many things. With it becoming so advanced, now our interpretation of reality could shift.
Last December, a Reddit user posted a porn video supposedly of Wonder Woman actress Gal Gadot.
The actress’s face was superimposed on the body of a porn star using AI, and apart from minor details that give it off, it was convincing.
Gadot’s fake clip helped raise awareness about the worrying rising trend of deepfake videos.
Deepfake: AI-Assisted Fakery
As is often the case with new technologies, some people find uses that we can’t always predict.
If you’re a fan of the Expanse book series, then Soren’s slimy attempt to use a deepfake against the magnanimous Bobbie Draper in Book Two might come to mind.
Combing the words “Deep Learning” and”Fake,” these celebrity fakes have become known as Deepfakes.
The same Reddit user who made Gal Gadot’s video uploaded other celebrity fakes of Taylor Swift and Maisie Williams.
Reddit and other social media platforms quickly banned deepfakes.
There are other non-porn deepfakes as well, like those featuring actor Nicolas Cage played for gags, or others showing politicians saying things they’re not supposed to say.
In recent months, the phenomenon of deepfakes has become widespread. Now, more users are making fake videos of anything they think would get hits.
Celebrity Fakes: How Deep Does the Faking Go?
With AI becoming more democratized and sophisticated, how trivial it would become to make deepfakes potentially harmful?
Another Redditor, clearly inspired by the first one, created a deepfake app to make it easy for anyone, with no background either in AI algorithms or in video editing, to make deepfakes using the footage and target of their choosing.
AI-assisted fakery is getting more and more accessible and convincing.
Of course, you’d expect the adult film industry to jump on the deepfake bandwagon.
Last August, a porn film company, called Naughty America, recently released a paying service that allows users to tweak porn videos to their liking. Customers can, for example, opt to use their own faces to superimpose onto the bodies of porn stars.
As if this wasn’t enough, researchers at Carnegie Mellon University also created their own AI-assisted fake videos.
CMU researchers developed a new AI algorithm, called GANs for Generative Adversarial Networks. GANs make it easy to copy the content of one image or video and apply it to the target video while keeping its original style.
In the video published by CMU below, John Oliver, the host of Last Week Tonight, had his facial expressions seamlessly modeled on those of Stephen Colbert, while retaining his style.
The CMU method could have many different applications. The film industry, autonomous cars, or robotics could all benefit from this technology. However, the developers are aware of the deepfake phenomenon and the dangers of it.
“It was an eye-opener to all of us in the field that such fakes would be created and have such an impact,” said Aayush Bansal, a Ph.D. student in CMU’s Robotics Institute. “Finding ways to detect them will be important in moving forward.”
Comments (0)
Most Recent