Deepfake Professionals

Deepfakes are created by using two machine learning–powered deep neural networks that compete against one another until a high-quality deepfake is created. The first ML algorithm (which is known as an encoder or generator) studies and learns by reviewing many hours of authentic video of a person to see how they look and move from many angles and under different lighting, as well as analyzes how their face moves when they are happy, sad, angry, or experiencing other emotions. Then they use computer technology to superimpose a copy of the person on the body of a different person. The second ML algorithm (known as a decoder or discriminator) attempts to detect the forgeries. The creator spends many hours to fix the problems identified by the decoder until it is extremely difficult to detect the forgery. In the past, generative adversarial networks were the main tool used to make deepfakes, but “in fact, the lion’s share of today’s deepfakes are made using a constellation of AI and non-AI algorithms,” according to IEEE Spectrum.

Deepfake technology can be used in both negative and positive ways. In a negative way, it is used to place famous women (actresses and other public figures) into pornographic movies or other media formats, create an impression that a public figure has done or said something that is offensive or even criminal, and manipulate individuals or organizations to unwittingly share confidential information or even transfer money from a corporate bank account to that of the deepfake creator because they believe that a company executive is telling them to do so in a video or audio message. Deepfake artists or creators who make these harmful deepfakes are simply people with tech skills who want to cause trouble, or they are employed by bad actors (e.g., countries, criminal organizations) that seek to create public distrust, harm the reputation of an individual, steal money or other assets, or meet other criminal goals. Deepfake detection researchers are employed by government agencies, computer security firms, tech companies to develop software tools and other methods to identify deepfakes and/or prevent the use of them in the first place.

Deepfakes are also created for positive reasons. For example, deepfake artists design and sell digital avatars that use the faces of paid actors to make commercials that can be broadcast in many countries. The artist changes the language the message is presented in and mirrors the actor’s facial movements to match that language. Deepfake visionaries hope to eventually use the technology to allow anyone to create Hollywood-quality movies without the use of professional actors, cameras, and lighting. In the video game industry, deepfake artists are creating “voice skins” that gaming companies sell to customers to enhance their gaming experience. In the assistive technology field, deepfake artists are working to create synthetic media to power personalized assistive navigation travel apps for pedestrians who have disabilities, as well as enabling and developing AI-generated synthetic media that allow people with amyotrophic lateral sclerosis (i.e., Lou Gehrig’s disease) and other diseases to communicate in their own voice despite not being able to speak anymore. In the fields of education and art, deepfake technology is used to bring historical or creative figures “to life.” For example, Samsung’s AI Center and the Skolkovo Institute of Science and Technology used deepfake technology to bring the Mona Lisa to life. Additionally, deepfake technology is used in forensic science and in the fashion, advertising, and marketing industries.