Author: Preslav Baldzhiev
The hardware and software development of technology provides more and more opportunities to use the capacity of a computer system to its full potential. Whether using this capacity to play a video game at the highest video settings, watch 4K movies, or to create different types of content, the principle is always the same – higher performance leads to a better end result.
In the context of creating different types of content, a serious problem arises – the digitalization of the world has made the Internet the main source of information, and this global network has a growing market share compared to more traditional communication channels such as television, newspapers, and radio programs. This creates an enabling environment for easy distribution of content that is not always authentic.
In recent decades, special attention has been paid to so-called fake news – articles or reports generated by paid sites that not only tendentiously cover an event, but also generate false information to spread – as the light of the problem on the purposes of this fake news. The goals of fake news can be presented as an attempt to compromise credible and reliable sources of information, to such an extent that the relevant audience begins to doubt what they have heard or read.
Such an emphasis on this serious media problem has managed to limit the impact of misinformation that has flooded more and more countries in recent years, and it is at the heart of the criticism with which the average Internet user approaches any headline in a news program.
Nowadays, consumer attention should be drawn to a new threat to public awareness – the so-called Deepfake content.
What is Deepfake content?
The term “Deepfake” has become a denomination for this type of content through the social platform “Reddit”. A Reddit user with the nickname “Deepfakes” has become extremely popular all over the Internet, uploading fake inappropriate content with the faces of various Hollywood actresses.
Thus, the username registered on “Reddit” has become the main name of an entire “genre” in video processing – the use of authentic video and photo materials to create fake content.
The name itself is not accidental – the word “deep” comes from the phrase “deep learning”, which is machine self-learning, in which computers learn to perform actions similar to human activity. The word “fake” comes to show how “authentic” such materials are.
Simply put, the process of creating Deepfake content involves taking certain features from one video (face, voice, hairstyle, etc.) that are appropriately edited in another video, to maximize the credibility of the newly created video. content – it must be so well made that it is indistinguishable from the originals. That is, the whole process is like a digital version of the plot of the movie “Face/Off” with Nicholas Cage and John Travolta.
How is Deepfake content created?
The creation of Deepfake content is based on a method that relies entirely on artificial intelligence, machine learning, and probability theory. The method involves the use of two independent but interconnected artificial intelligence – in other words, the creation of Deepfake content is based on machine self-learning called Generative Adversarial Network (GAN).
The first artificial intelligence was engaged in analyzing two types of files – this artificial intelligence will be called the “Creator”.
The first type of file is the one that will be used as a basis for creating fake content (the video on which certain features of the other video will be edited – we will conditionally call it “background”). The second type of file is entirely related to the characteristics that will be edited on the first file – that is, if a face will be edited, the artificial intelligence analyses a large array of data (images or videos that show different facial expressions) of the person.
After the analysis, the Creator starts pixel by pixel to transfer the facial expression from the second type of file to the background. Once this transfer is complete, the First Artificial Intelligence submits the final result to the Second Artificial Intelligence (we will tentatively call it the “Controller”).
The controller analyzes the video created by the First Artificial Intelligence and evaluates the authenticity of the new content by comparing it with the two types of files mentioned above. In case the Controller deems that the final product can be determined as non-authentic, the Controller submits this information to the Creator and the whole process starts again.
The cycle is repeated until the Controller cannot distinguish between the generated fake content and the analyzed authentic video materials.
How do we know the technology for creating Deepfake content?
Deepfake content is not recent. The prototype of such an algorithm can be found in the applications “Facebook Messenger”, “Snapchat”, “Instagram” and others, where the user can take a picture using various visual effects. Around these effects, there is a public rumor that the use of these visual effects is dictated by the need for content through which to improve facial recognition.
The facial recognition technology itself is built into every smartphone that allows unlocking in this way. However, for such facial recognition to be applicable for unlocking a mobile device, the face is required to be static and still – this is not typical of Deepfake content, where dynamism and facial expressions are essential.
How to Deepfake Content Recognized
It is becoming more and more difficult for a user to distinguish which video is authentic and which is the result of Deepfake technology. In earlier years, Deepfake content could be detected with the naked eye – often the areas where the pixels are edited remain blurred. Another telltale sign is the unnatural lighting in the area of editing compared to the other images in the video.
Now that technology has evolved, Deepfake content recognition is possible thanks to specially designed software that also uses artificial intelligence and machine learning.
Such software is currently being developed and implemented by Facebook to limit counterfeit content that is distributed on social media.
the pros of Deepfake content
- Film industry – The principle of creating Deepfake content is strongly represented in the film industry. The most popular example of generating such content is the seventh part of the movie “Fast and Furious” – before the shooting of the last scene, actor Paul Walker died tragically in an accident. However, he “managed” to get involved in the ending of the movie, as his face was recreated by software from footage already shot with Paul Walker himself, as well as video and photo content of his brothers Cody and Caleb Walker. Similar is the case with Carrie Fisher, better known as Princess Leia in the Star Wars series. She also passed away during the filming of the part entitled “The Rise of Skywalker”, forcing the producers to resort to the principle of Deepfake content to successfully complete the movie.
- Art installations – In 2019, the Museum of Art in St. Petersburg, California presented the exhibition “Dali Lives”. The concept of the organizers is extremely atypical – using the technology of Deepfake content, the organizers create videos of the popular artist Salvador Dali, through which Dali himself presents his works. Given the artist’s eccentricity, the organizers go beyond the typical exhibition standards by creating content in which Salvador Dali holds a smartphone and takes selfies with visitors, and the photo itself is presented to visitors. At the moment, the exhibition is still active and visitors from all over the world can “meet” the famous artist.
- Police investigations – One of the purposes of the investigation of a crime is to find the perpetrator and bring him to justice. Finding the perpetrator is not always easy, taking into account the conditions under which the crime was committed, the stress the victim experiences, etc. Here comes the aid of photo robots, which with the help of software generate the image of the “alleged criminal”. The technology used to create Deepfake content provides different visual representations of this image (with different hairstyles, different makeup, etc.).
The Cons of Deepfake content
Each coin has two sides. Despite the benefits, creating Deepfake content poses serious dangers and can create unpredictable harm.
- Privacy – Whichever way we look at the problem, creating Deepfake content invades a person’s privacy, and given that this content is not real, the problem becomes even more serious. As we clarified at the beginning of the article, Deepfake content gained popularity through the publication of inappropriate content, which “involved” world-famous actresses. According to a study by the NGO “Deeptrace”, 96% of the nearly 15,000 Deepfake videos found have such context.
- Use of biometrics – Unregulated access to biometric data such as facial features and voice is a serious problem that leads to immense abuse of identity. This will directly affect the work of border services, as the possibility of using biometrics would allow the creation of false identity documents, and thus facilitate the trafficking of people.
- Economic aspect – In early 2020, an employee of a Hong Kong-registered company received a call on his office phone – the voice from the other side was that of the company’s manager, who authorized the employee to transfer $ 35 million, under the pretext that this money will be used to acquire assets on behalf of the company. It doesn’t take much time, and it turns out that the company’s management didn’t have such a conversation at all, and the voice heard by the employee, which resembles the voice of the manager, is nothing more than Deepfake content generated by artificial intelligence.
How to protect yourself from Deepfake
One way for an individual to protect their photos and videos from being used in Deepfake content is to use a watermark. The watermark is a translucent element (it can be a figure, it can be an inscription) that is placed on the image or video file. According to cybersecurity experts, this small detail still confuses the artificial intelligence that creates Deepfake content, because the pixels in the watermark intertwine with the pixels in the image or video itself.
Another option to protect against Deepfake technology is to capture and possibly publish files in low resolution – for the same reason stated in the previous paragraph.
The most basic way to protect yourself from Deepfake content is to make the account on a website or social network private (only certain people can see the post), reduce the posts posted, or stop posting content.
The legal status of Deepfake content
There is currently no legislation that explicitly regulates the creation and use of Deepfake content. Refracting from the dangers of the Deepfake materials, the creators and publishers will be responsible for violating the inherent human right guaranteed in all democratic constitutions, namely the right to privacy.
On the other hand, creators of Deepfake inappropriate content may be subject to the crime of aggravated, namely the insult is disseminated through the media.
Deepfake content also directly affects the copyright of the materials used to create such materials. Deepfake content is a reworked art, but the only difference is that the author’s permission of the primary works for such reworking is almost always lacking.
Conclusion
Deepfake content can be defined as a time bomb. So far, it is widely used for the lesser evil – violation of the rights of individual public figures (but the fact that it is the lesser evil does not in any way justify the use of Deepfake content).
The lack of a legal framework to regulate the use of Deepfake content is a significant problem that ordinary citizens and states will face. Certainly, the use of such technology should not be completely banned, but limited to strictly defined cases (as in the case of the Art Installation in Florida). And until such laws are enacted, everyone has no choice but to hope not to fall victim to Deepfake content.
This material prepared by Preslav Baldzhiev aims to provide more information about deepfake content on the Internet. It does not constitute a legal opinion and cannot be interpreted as individual consultation on any concrete facts or circumstances. The advice of a specialist should be obtained for specific questions and situations. For more information on the above-mentioned issues and individual consultations, please contact the team of the law firm of Krasimira Kadieva at 00359 882 308 670 or make an inquiry using the contact form of the website. Since 2017 Preslav Baldzhiev is a law student at Sofia University “St. Kliment Ohridski “, having previously graduated from the High School of mathematics and natural science “Acad. Nicola Obreshkov” in Burgas. In February 2020 he took a course for industrial property representatives at the Patent Office of the Republic of Bulgaria in the field of trademarks, geographical indications, and industrial designs. He is interested in intellectual property, personal data protection, commercial and law on obligations and contracts and also regularly attends conferences, practical courses, seminars, and webinars.