What is Deepfake

How deepfakes appeared

Scientists have been developing AI content technologies since the late 1990s. In 1997, Video Rewrite introduced a video creation program that synchronized a person's articulation with a generated audio track. That is, the facial movements completely coincided with the sound created by the computer. The result was believable conversational facial expressions.

The term “deepfake” appeared in 2017, when an anonymous user with the nickname “deepfakes” posted AI pornographic videos on the Reddit forum: he “glued” the faces of celebrities to the bodies of actors. Anonymous did this using Face Swap technology. At the same time, deepfakes of Nicolas Cage appeared on the Internet—the actor’s face was added to films in which he did not star.

Since then, deepfakes have begun to gain popularity and penetrate everyday life. Nowadays, AI technologies are actively developing: new opportunities are emerging for generating images, voice processing, and combining computer graphics with real videos. Development companies offer professional software and amateur applications. AI is becoming accessible not only to a narrow circle of people, but also to ordinary users.

The scope of application of AI content is also growing. It is used in:

• advertising, marketing;
• entertainment videos;
• computer games;
• science, education;
• journalism;
• painting, cinema;
• fashion industry;
• design;
• architecture.

How deepfakes are used in marketing

Samuel Stefenson from Deep Nudes says that - "AI content helps solve the same business problems as classical methods of promotion: presenting goods and services, attracting customers, improving recognition, increasing sales and profits."

In addition, the tool allows you to:

Reduce advertising costs. To create AI videos and photos, you don't need to hire people to film and pay them a fee. Thanks to deepfakes, the company saves its advertising budget and gets the opportunity to recreate the image of any person, including a celebrity.

Adapt advertising faster. AI makes it easier and faster to create content for different platforms: technologies help quickly create videos for social networks, dialogues for radio advertising, images for street billboards.

Thanks to deepfakes, brands quickly adapt advertising to different conditions: for example, they create a video in several languages ​​at once or change the location (“draw” another city or country in the background). This saves time and effort.

Experiment. Thanks to AI, brands create creative advertising, stand out from competitors and surprise the audience. For example, they offer clients to dress in an online fitting room or bring fictional characters into the real world.

How deepfakes are created

In 2014, American researcher Ian Goodfellow developed generative adversarial networks (GANs) - two machine algorithms that jointly generate AI content. This technology is most often used to create deepfakes.

One neural network (generator) studies real objects in detail and creates a new image, and the other (discriminator) evaluates its realism. 

For example, the first algorithm processes hundreds of thousands of pictures of a person from different angles (facial expressions, characteristic features) and generates a fake photo, and another network checks how suitable this picture is for the task and whether it is unique. Audio and video are created in the same way.

These two neural networks seem to compete with each other: the generator tries to deceive the discriminator, and the latter tries to detect a fake. This way, the algorithms teach each other and improve, resulting in the most natural content possible.

What are the dangers of deepfakes and how to recognize them

Deepfakes can defame a person’s honor and dignity (for example, when publishing fake pornographic content), and also influence the politics of a country. 

In 2018, a falsified video appeared on the Internet where Barack Obama criticizes Donald Trump. It was as realistic as possible and quickly went viral. In this way, the content authors wanted to demonstrate the danger of deepfakes and draw attention to the problem. 

In the United States, deepfakes are considered a national security threat. In California, it is illegal to publish AI content featuring politicians and post fake pornography without the consent of the person whose image is based on it.

To protect yourself from scammers, you need to follow the following security rules:

• trust only official sources of information;
• be wary of suspicious calls asking to transfer money (ideally, call back the person on whose behalf the call was made)
• inform loved ones about different methods of fraud;
• regularly create backups on devices, install a high-quality antivirus, set up two-factor authentication in social networks and government services (to combat hackers, protect personal content and personal data).