Although deep fakes have been around in one form or another for a long time, they have been getting a lot of attention lately. According to MIT Technology Review, Facebook has even been creating some deep fakes of its own.
For those who might not be familiar with deep fakes, it is essentially a technology that allows fake videos to be created. Typically, these videos feature someone saying or doing something that they would never do in real life. For example, a deep fake video might show Buzz Aldrin (the second person to walk on the moon) telling a crowd of people that the moon landings were fake and that the Earth is flat.
Deep fakes and ‘South Park’
The adult entertainment industry is often credited with inventing or at least perfecting deep fakes. Personally, though, I suspect that the TV show South Park might have pioneered deep fakes without even realizing it.
One of the things that South Park was best known for during its first few seasons was its crude animation techniques. I won’t pretend to know the processes that were actually used to animate the show, but I would speculate that animators created the show’s characters using construction paper. From there, they may have taken a series of photographs, and then animated the characters by displaying the photographs in rapid succession.
Although South Park features its own regular cast of characters, the show sometimes features celebrities or people who are in the news. Rather than trying to draw the person in the usual South Park style, it was once common practice for the animators to paste a photograph of a celebrity’s face onto a cartoon character’s body. If you want to see a particularly amusing example of this, just do a Google search on the phrase “South Park Saddam Hussein.”
It probably goes without saying, but the celebrity appearances in South Park aren’t going to fool anybody. Not only is Saddam’s role in the show completely implausible for example, but the quality of the animation was never meant to be convincing. Besides, the show’s disclaimer even states that all celebrity voices are impersonated — poorly. Even so, South Park’s technique of pasting a photograph of someone’s face over the top of someone else’s body really isn’t all that different (at least at a high level) from the way that modern deep fakes work.
As previously noted, modern deep fakes are probably best known for the ways that they have been used in porn. It has become a somewhat common practice to superimpose a famous actress’ face over the top of an adult film star. Rather than the static, lifeless photographs that South Park uses however, the porn industry allegedly uses artificial intelligence to perfectly produce lifelike facial expressions. Admittedly though, I have never actually seen an adult film that uses deep fakes and therefore lack any firsthand knowledge of how convincing the effect actually is.
Deep fakes are no longer unconvincing
Like South Park’s paper cutouts, early deep fakes were less than convincing. As the technology has matured, however, deep fakes have become much more realistic. Additionally, deep fake use is no longer limited to porn. The Internet is filled with examples of deep fakes that have nothing to do with porn. While some of these deep fakes are more convincing than others, many of the examples are unsettling, to say the least.
At least some of the deep fake examples shown on the page that I linked to above seem to have been created just for fun, or perhaps as a way of demonstrating what the technology is really capable of. At the same time, though, there have already been deep fakes that were created with malicious intent.
The malicious use of deep fakes has the potential to become a huge problem. After all, people tend to think of video “evidence” as being trustworthy. If you show someone a picture of a famous person doing something that seems to be out of character for them, most people will probably assume (or at least wonder if) the picture has been photoshopped. However, if you show that same person a video of the same event, they are more likely to believe that it is authentic because historically it has been much more difficult to create video forgeries.
According to the MIT Technology Review article I linked to above, Facebook is anticipating an eventual onslaught of deep fake videos. In an effort to prevent fake videos from gaining traction and being seen as credible, the company has begun creating its own deep fakes. The idea behind this effort is that once the company is in possession of good quality videos that are 100 percent known to be fake, Facebook can begin developing techniques and algorithms for distinguishing between legitimate videos and those that are fake.
This, of course, raises the question of just how realistic a deep fake video can really be. Are deep fakes really so convincing that they can fool all but the most keenly observant among us?
As it stands today, the average person probably isn’t capable of producing a deep fake that would hold up to any kind of serious scrutiny. Having said that though, the technology for creating a really good deep fake does exist — and has existed for a long time.
Not a new technology
Consider for a moment that the movie “Forrest Gump” was released in 1994. The movie featured extremely realistic depictions of historical figures such as President John Kennedy. (You can see a clip of that scene below.) The footage looked so good, in fact, that it would be easy to assume that the film was pulled directly from the presidential archives.
To put Hollywood’s ability to create fake videos into perspective, consider something that I mentioned a moment ago. “Forrest Gump” was released in 1994. In other words, the movie proved that Hollywood had the technology to create realistic fakes a quarter of a century ago. Granted, “Forrest Gump” had a production budget of $55 million. Even so, the movie conclusively demonstrates what can be achieved if sufficient funding is available.
So, with that in mind, consider that it has become common for U.S. presidential candidates (those who have received their party’s nomination) to spend hundreds of millions of dollars on their campaign. An unscrupulous candidate with that kind of funding available to them could conceivably create a plethora of deep fake videos for use in attack ads. This is the sort of thing that Facebook seems to be trying to prevent with its deep fake detection algorithms.
Although I am often skeptical when it comes to some of the things that Facebook is involved in, my hope is that it will be hugely successful in creating a tool that can differentiate between real and fake videos with 100 percent accuracy. While I’m not really all that concerned about deep fakes being used in political ads (most people seem to ignore the ads anyway), there are simply too many other ways in which deep fake technology can potentially be abused.
It’s an absolute given that technology improves over time. As technology continues to progress, it will inevitably only be a matter of time before the average home user will have the ability to produce deep fakes rivaling those found in Hollywood films. When that happens, it will be all too easy for cybercriminals to create fake videos for use in blackmail schemes. Deep fake technology could even be used to frame an innocent person for a crime.
Featured image: Shutterstock