You’re hanging out minding your own business, checking out the latest gossip online or spinning the reels at Vegas Palms Casino when someone forwards you a video. It’s Tom Cruise, you can see clear as day, but he’s doing something you would never in a million years expect. For the sake of being polite and keeping this PG13, let’s say he’s juggling chickens while hopping on one leg. You’re baffled. Why would Tom Cruise allow such a video to go public? Sure, he jumped on a couch once, but even for him this is a step too far.
If you haven’t already guessed from the title of this article, it’s a deepfake. Or a CGI generated fake in which a celebrity’s face is inserted into a video. Normally these creations are amateur at best, and pornographic in nature. But another more creative fake had Jon Snow apologising for the final season of Game of Thrones. It was so believable it went undetected for, frankly, longer than it should have.
How Does It Work?
So how are these impressive fakes made? The simple answer is, fairly easily. The slightly longer answer is pretty easily, but not as easy as you might think to make something genuinely indistinguishable from the real thing.
The software itself is referred to as generative adversarial networks (GANs,) and is based around two software systems going head to head. One attempts to overlay generated information onto an image, while the other continually attempts to detect the job as fake. These two systems play off one another, until something approaching a realistic looking forgery is created.
A Pro Touch
This all sounds very high-tech and cool, and it is. But there are a few things to keep in mind. The first is that creating a deepfake face isn’t the equivalent of snapping your fingers, even with specialised software. The reasons are twofold.
Firstly, the algorithms are required to have as many examples of the targeted face as possible, from as many angles as possible. So in the case of a celebrity or politician, tough luck. For the average person on the street, the process is a great deal more difficult. So rest easy, unless you have a ton of videos online that show off your beautiful mug from every angle. You’re safe, mostly.
Secondly, the process takes hours. It isn’t a case of cut and paste and viola you’re done. Furthermore, creating truly believable fakes that will survive close scrutinising are a true art form. The quality of the image created must be matched in the target video in many, many subtle ways. Ambient light, believable shadows, synchronised movements and seamless transitions are all part and parcel of creating a real looking fake.
Are They A Problem?
Now you about deepfakes will you ever be able to trust anything you see online again? Or is it time to simply start assuming everything is fake all the time? On second thought, that’s the world we have lived in for years already.
In the age we live in ‘fake news’ gets a stunning amount of traction before being called out. The fake news itself isn’t even especially convincing the vast majority of the time. Yet it goes far, convincing millions, or even tens of millions, without much chance of it being exposed as incorrect or misleading. Add to this scenario a realistic looking video, and we have a recipe for some serious mayhem. Tom Cruise juggling chickens? A joke. How about an incredibly realistic fake of a politician inserted into a racist or sexist video days before election time? No longer a joke.
Republican senator Marco Rubio went on record to say that he felt deepfakes are as dangerous as a nuclear weapon. Driving the point home, he suggested that it used to take 10 aircraft carriers and weapons of mass destruction to attack the United States. Now all you needed was a computer and Internet connection.
Not Changing The Game
A very dramatic sounding example, but are deepfakes really changing the game as it stands as much as the senator is suggesting? Tim Hwang, director of the Ethics and Governance of AI Initiative at MIT, had a more level-headed approach. He agreed that the implications of deepfakes are concerning, but that they hardly change up the world of the Internet and social media from how it already stands.
This, in a nutshell, is true. There were a multitude of chicken juggling images that involved celebrities, long before this new wave of technology. This is simply the first time the images are moving. In fact, the fake news epidemic has clearly caused an astonishing amount of damage and misinformation, prior to the recent deepfakes. So to claim that the problem is only becoming real now seems like somewhat of a delayed reaction.
The truth is that deepfakes might be just what it takes to make people more aware of how rampant fakery is online, and to not blindly believe what they see.