O Deepfake is a developed technology that uses artificial intelligence (AI) to create fake videos from images. The technology is not new, and it already inhabited the forums of Reddit since December 2017. 4j1i12
This year, a named Deepfake started posting videos of explicit content of famous people on the forums. For this, he used software deep learning, which applied faces of famous people in unusual scenes. Some cases became famous, such as Gal Gadot and Emma Watson.
After these events in Reedit, the practice of edited videos using machine learning algorithm and other possibilities for IA started to be called Deepfake. And, with that, the technology started to be studied with great caution, since it can, as in its beginning, be used in unorthodox ways.
The Deepfake Problem 1w271l
Algorithms, however well developed, are not perfect. However, this practice can deceive many people, and therein lies the danger. Misuse and bad nature can cause people to be placed in situations they were not.
The warning came when the practice began to involve politicians, with lines that they themselves did not say. And, nowadays, with information running very fast and fake news propagate faster, it becomes very easy to spread false information according to self-interest, ed by alleged video evidence.
As much as many use it as a joke, this practice and the dissemination of this type of content can represent a danger. In addition, these videos could even threaten the credibility of everything that is published.
In the case of fake intimate videos, this raises several ethical and legal problems. Misleading creations can harm a person's life, whether famous or anonymous, and for now, the law provides penalties to the call porn revenge.
Where does Samsung enter this story? 6g59
A Samsung, however, also decided to implement a new AI that will develop the same technique. The downside comes in how the technology is used by s, and how the algorithms will learn the information.
Before going into practice, the algorithm will start with a long meta-learning stage. This means that the algorithm will watch several videos in order to learn how human faces behave.
It then applies what is learned to a single or small batch of photos to produce a reasonably realistic video. Unlike a true deepfake, the results of this AI are intended to leave imperfections in sight.
This will make the image look like it has been altered so it doesn't lose its reliability. For example, in the video of Marilyn Monroe using technology, the mole on her face was removed. This was done to make it clear that it's fake, giving full credibility.
Check out the video below:
Despite all the problems, the Samsung still bets that the algorithm will still be able to faithfully reproduce the images, even if it leaves clear marks that the video was generated by an AI.
Source: Cnet
