What is a deepfake?

This function is a development of artificial intelligence that has acquired new forms of use. Artificial intelligence (AI) is constantly evolving and creating new variables for its use. Among the best known is the so-called generative artificial intelligence, which, according to the United Nations (UN), is a type in which a program performs things (such as texts) as if they were done by a human being.

Among them is also the deepfake, which uses an AI technique for its creations and is available to users. Currently, this modality has become popular with different uses, including disinformation. But... what is it and how does it work?

Artificial intelligence: what is the deepfake

The so-called deepfake consists of images or videos that are generated by means of an artificial intelligence technique. It is a “machine learning” called deep learning in English (in Spanish: deep learning), explains the University of Virginia (United States) on its website.

As the Encyclopedia Britannica points out, deepfake is made up of two English terms: deep, which refers to artificial intelligence, a machine learning that in turn is made up of many levels of processing; and fake, which refers to the falseness of the material that is obtained as a result.

These images or videos that are created by this AI capture something that, at first glance, seems true and realistic. However, the content is false and is based on people who exist or existed, mentions the world organization.

How is deepfake generated?

This type of material is generated by means of two different types of AI algorithms. The first is responsible for duplicating a real image or video as faithfully as possible. While the job of the second is to detect if it is false to inform the first of the different things that exist between that replica and the original.

In this way, one feeds back to the other and the procedure is done as many times as necessary until the second algorithm does not find false data.

When it comes to deepfake videos, a specific character's voice can be duplicated by feeding the AI ​​model with real audio data of the person. This way, the artificial intelligence is trained to imitate a subject's tone as closely as possible.