Senior Expert in Emerging Technology
Have you ever seen a photo that seemed impossible, such as a man holding a cell phone in a World War II photo or alien spaceships flying over a city? We have been living in a world for a long time where media can fake stories, including photos of the topics in question. In the early days of Photoshop and other digital photo editing software, it was difficult to tell an edited photo from the real thing, but as the technology advanced, the ability to detect fake photos also advanced. It is possible now for computer algorithms to be used to easily detect photos that have been edited and separate them from authentic photos. There is a new fake in town that is much more convincing than fake photos and it is a technology that has been called Deepfake.
Deepfake is an Artificial Intelligence (AI) based algorithm that can be used to replace faces in videos, making it possible to mimic not only the look of a person’s face, but one trained from “watching” videos of the person being created can also mimic their facial expressions and idiosyncrasies, making it nearly impossible to detect the fake when the AI has been trained effectively. The technology gets its name from the merging of “Deep” in reference to the underlying machine learning algorithm set called deep learning and “fake” because it was first used to create completely fake video personalities for fictional characters. Deepfake is most widely used today to create TikTok videos, either containing completely fictional characters, or likenesses of famous people.
It took me a while to see the advantages of such a technology, besides the obvious and deceptive use cases of faking famous people or even world leaders. The technology has become advanced enough, and our important public figures have become active enough in video feeds that it is no problem to gather enough video content to train the AI, making it possible for anyone with the knowledge of Deepfake algorithms to make fake videos of pretty much anyone else’s face speaking on any topic they desire.
The technology has been in development by shared research across several academic institutions since the early 1990s but has only recently become advanced enough to be a useful technology. One of the earliest uses was to bring historic figures and even artworks to life. Russian researchers used Deepfake technologies to bring the Mona Lisa to life and the Dall Museum in Florida recreated their namesake out of archival video footage to attract visitors. Deepfake makes it easier to use stunt doubles in feature films by giving producers the ability to even produce close-ups using Deepfake without having to worry about small differences between the actor and stunt double. In fact, with Deepfake technology they could use anyone, including actors of the opposite sex, in closeup scenes requiring a stunt double.
However, like all technology, Deepfake has its downsides. This same technology that can be used to bring paintings and historical figures to life in museums can be used to create fake videos of real people. Imagine for a moment that someone uses Deepfake to pretend to be someone important, like the CEO of a major corporation, and uses the Deepfake video live on a video conference call to control the corporation. That is not far from what happened with one of the first Deepfakes-based fraud cases.
A scammer used Deepfake to video conference with the CEO of a UK based energy company, faking the voice and video of his Germany-based supervisor from the parent company, requesting a bank transfer of 220,000 Euro to a third-party bank account. Thankfully the CEO called his boss to confirm the transfer and the scam was a failure, but there are many other illegal uses for the technology.
The technology in and of itself is not a bad thing, but in the case of Deepfake it seems to bring more uses for harm than for good, which is not surprising considering the technology’s first broad use case was to produce fake porn videos of famous actors and actresses that have been widely distributed. I’m fairly certain it was not the primary use case the academic creators of the technology had in mind. However, there is still a struggle in the industry to find good, legitimate use-cases for the technology, and it seems the fakes are hard enough to detect that partnership between Microsoft, Google, Amazon and Facebook have combined efforts and are cofounding a $500,000 prize for a successful algorithm to automatically detect Deepfake videos.
Until next week, stay safe and learn something new.
Scott Hamilton is an Expert in Emerging Technologies at ATOS and can be reached with questions and comments via email to firstname.lastname@example.org or through his website at http://www.techshthepherd.org.