It’s been a little while since I dived into the world of data science. So, doing my standard search, I wanted to see what terms were popping up more often and what trends there were to watch out for. One term I’ve noticed starting to show up more is deepfakes. I have a general idea of what deepfakes are, but there seem to be some disagreements on them. Some people think it could be destructive, others think it could be harmless. A lot think we are still years behind in technology, and so it will not really be seen for many years, and yet there are instances of deepfakes that have been going on for years whether we realize it or not.
The goal of this article will be to learn what a deepfake is and how they’re used today. This will be informative, and although I may share some stories or opinions, I have no intention of trying to persuade you to think one way or the other about the technology. At the end of this article, feel free to share your personal opinions or experiences in the comments. I look forward to reading them.
But without any further delay, let’s get started with what a deepfake is.
What is a deepfake?
Admittedly, deepfakes are a little broader than I originally intended, so a clear definition was somewhat difficult to dig up. In the shorter definition, deepfakes are the instance of videos or photographs being altered to contain the likeness of someone else. This is a technology stemming from AI (Artificial Intelligence).
But I know that sounds vague, so let’s look a little deeper. To my understanding, from how I’ve seen it used anyway (not saying it may be the only examples), deepfakes are when you take a video or sound clip that already exists and swap the face, voice, or both, for someone else. What you would be seeing in the faked version is not real, and can change the context of a clip. This is where it can get complicated. The intention of these videos can be a little unclear, or the overall goal of the poster can change by a case scenario.
When broken down, the fake refers to the clips no longer being real, as they were manipulated from reality. The deep, however, can refer to a short but still listed variety of meetings. By this, I mean that the deepness could refer to the complexity of the change. However, it could also refer to the intensity. Perhaps, it could even mean the difficulty to create the deepfake.
Speaking of the difficulty, what goes into creating a deepfake? Now, for the user, the amount of effort could vary. For example, certain Snapchat filters change a person’s face. I’m not referring to the face tuning the app does, but the certain filters that turn you into a different person, character, or even food item. In this scenario, changing your face is incredibly easy for the user, although it’s not convincing. Other apps allow you to record yourself based on a reference photo of another person, but if you move your head even slightly left or right, the photo just doesn’t stretch right and it’s a very obvious fake. However, some more realistic Deepfakes require a mixture of machine learning and neural networks. From what we know about machine learning and neural networks, that would be a very difficult, let alone time-consuming, task. However, this creates a much more realistic image, especially with more reference photos being used.
Examples of Deepfakes
One example of a more realistic deepfake is on TikTok. On the app, there’s a user who deepfakes as Tom Cruise (who isn’t on TikTok currently). Now, he doesn’t slander or even talk about any issues today. Instead, the videos are more to show off what the technology can do. One of the videos is just about eating, which shows how realistic everyday actions could seem when there are millions of photos for a machine to learn based on. Celebrities are much easier than your average people, as there is a great deal more content to reference.
I also mentioned the less realistic face-swapping apps like Snapchat. But that isn’t the only one. On Facebook, you can add your face to some movie characters so you can see yourself in whichever scene is selected from what they have available. It is brief and does not look very realistic, but your face is blended onto a character that you choose from their list.
But that should bring the question into your mind of the intention of deepfakes. Before you can judge whether something is "immoral", "evil", or maybe even "good" and "innovative", you need to consider the intention of deepfakes. Now, not all people would have the same intentions, of course. That’s just how things are. But maybe we should also consider that before you cast your final judgment on the technology.
Deepfakes, are they evil? Creative? Both?
No easy way to dive in so let’s just start with the creative side, which isn’t bad. In my previous examples, they could all be seen as just being creative or having fun. There’s nothing evil about using a funny filter you find for fun. Catfishing is another story, but we’re assuming if you’re going to do a true deepfake, it’s likely recognizable as a fake. Posting on your feed with your face in your favorite character’s body isn’t "evil", it’s just having a bit of fun. So, in this case, there’s nothing wrong with deepfakes.
Now, there are some more realistic versions, such as the TikTok Tom Cruise we talked about, but the account makes it clear that their intentions are just to show off the technology. Because he’s not using it to influence, it’s more for education and entertainment purposes, so in this case, there are also no immoral uses.
Unfortunately, there are some "evil", or immoral, uses of deepfakes. For example, and without saying anything too crude, some people will deepfake a celebrity’s face to a non-appropriately displayed body, such as creating fake adult content of that celebrity. Especially if that celebrity has chosen to not show too much skin, even though it’s a fake body it is still immoral and usually only done for selfish or inappropriate purposes.
Now, we originally talked about the intention of a video. But what if the purpose is to create false information or opinions? For example, if some extreme political views were being shared, then someone deepfakes a candidate onto that extreme point, it takes only one person to believe it is real to lead to defamation. This is both immoral and a form of slander. Ethically, this would be reason enough to not support deepfakes. However, for now, at least by the general public, making something realistic enough to convince people it is real without question may be out of reach.
Are deepfakes still new to use?
In a few of the articles I read to get my resources, I noticed a common trend that people seem to think technology isn’t quite ready for deepfakes quite yet. What they’re likely referring to is that the common person isn’t able to create a realistic deepfake yet. But for those who think it just isn’t being used, I would argue that simply isn’t true. Just ask the meme community. Deepfakes are used to make memes, posts, or photos we send to friends in everyday settings. They may not look very realistic, but still, they are getting better. Unfortunately, the celebrity face on adult content is something very real that happens today as well, and those photos are often touched up to look more realistic. You can still usually tell they are fake, but it is much more difficult to distinguish. Now, we’re not commonly using it for false information on people or slandering yet, which is probably what they’re referring to, or at least with it looking realistic, but we don’t know that slander will be the primary use. There will always be people who use technology for the wrong reasons, but that doesn’t always speak for the majority of people.
Conclusion
In the end, you should make your conclusion on whether you think deepfakes can be used for entertainment or if they could lead down to more immoral paths. We learned that a deepfake is a form of artificial intelligence that takes an existing clip and swaps out things such as the face. This creates a fake version of a video. In common uses, the fakes do not look very realistic, but this is not always the case. Technology, machine learning, and Neural Networks can be used to create a more realistic version of the fake until you are left with a product that is difficult to tell from reality.
The intention of a deepfake depends on the person. Some use deepfakes to entertain, others to educate, but some may also choose immoral routes for their fakes. In this article, we covered a few examples of what deepfakes are after defining them. The goal was simply to learn about what they are and examples to help your familiarity with recognizing a deepfake, but not to persuade you on them. You may cast your judgment on deepfakes, but with a lot of people saying technology is not quite ready for realistic fakes, perhaps it is still too early yet to know. Feel free to share your thoughts about deepfakes or examples where you’ve seen one in the comments. Until next time, cheers!
Read all my articles for free with my weekly newsletter, thanks!
Want to read all articles on Medium? Become a Medium member today!
Check out some of my recent articles:
Quitting Kubernetes Kubeadm and Switching to MicroK8s
Leaving your job for "selfish" reasons
References:
Words We’re Watching: ‘Deepfake’