Many of us are probably familiar with the saying “A camera never lies” or “A picture tells more than a 1000 words”. These sayings, as true as they once were, might not be as accurate in the world we live in today. Technologies such as Adobe Photoshop enable everyone with a computer and some picture editing skills to significantly change the meaning or appearance of any picture. By now it has become widely accepted by society that the big majority of pictures we see on a daily basis were edited to a certain extent, meaning that a picture has lost a lot of its trust during the last years. Until now video recordings have remained untouched, making them a trustworthy source of proof for anyone, even judges. However, with the rise of a new technology called “Deepfakes”, this is going to change dramatically.
What are Deepfakes?
Deepfakes allow the user to swap one face for another one, disregarding if this is in the same or different image or video. I am sure that most of you are already familiar with the basic technology behind the face-swap-technology, and it might be that you have used it today already. I am talking about Snapchat and the application that allows you to replace your own face with the one of a dog, but also with another friend. This technology has also been used for many years in the movie industry, however, until now this required highly skilled people, expensive materials like cameras, and a lot of time. With the rise of artificial intelligence, or more specifically, deep leaning, this has changed forever.
Already today there are hundreds of apps available allowing anyone to create incredibly realistic fake-videos, for free, and within a few hours. This technology, certainly, comes with strong advantages and disadvantages, but first, let`s discuss what this technology actually is all about.
Let`s say you want to create a fake-video in which you face swap two faces within a particular scene, and you use the program called “fake app” to do so. This, essentially, can be done in four relatively easy steps.
Step number one is to simply select a scene you want to edit. Let’ s say you want to face swap the face of Amy Adams in the last Superman movie with the face of Nicolas Cage. The second step is to find a large range of images of the two actors, in this case of Any Adams and Nicolas Cage. Once you have done that, the program will automatically download and use them as its learning data. The third step is to get rid of all the things in the images we do not want, such as buildings, cars, other people, or whatever other backgrounds you can think of. Lastly, we let the AI software do its job. In this step, a neural network model starts to learn how to recreate the required face based on the test data we provided it with, which is in our case the pictures of both actors. The program will continue running until the error in the created image is below a threshold, after that you have successfully created a high-quality fake-video. With this link, you can find the link to the video.
Obviously, such powerful software comes with advantages and disadvantages.
The biggest advantage of this technology certainly is a significant decrease in time and money for any kind of video or movie maker. Another advantage is also that knowing the possibilities of this technology might make us as a society more aware and sceptical to daily news, which is something, I think, has not been put enough attention to during the last years.
Regarding the disadvantages, the most obvious one is that it becomes even more difficult to distinguish real from fake. Particularly in the field of politics, this becomes very dangerous, especially if it is combined with another program, called “Adobe Voco”. This allows users to change the entire speech of a person, by simply analyzing a previously held speech of this particular person. With the help of learning algorithms, the user can then make the person say anything, with as little as 40 minutes of content required as learning data. With the use of these two programs, almost everyone can make every politician say anything, making it extremely difficult to determine what is true and what is fake.
If you are asking yourself now what the best way is to determine the originality of a video, paradoxically the answer is again Artificial Intelligence. Simply because if we have the ability to create fake videos with the help of AI, then we are almost certainly also able to create an AI which is particularly focused on detecting which videos are fake and which are not.
To conclude, I, personally, think that these Deepfakes are neither good nor bad, but just another step in broadening the opportunities this technology provides us with. However, until this step will be overtaken by the following applications, we as a society might want to start asking ourselves twice about the quality of information before we start believing, or even acting on them.