Deepfakes are videos or images made with a technique for human image synthesis based on artificial intelligence. It is used to combine existing and original images and videos.
The term deepfake refers to a technique based on artificial intelligence that is used for the synthesis of human images. This uses machine learning technology to combine and overlay existing images and videos with original videos or images. Deepfake also commonly refers to the product of what is made using this technique.
Have you recently seen a public figure (a politician, an entrepreneur particularly exposed to media attention) say something compromising? It’s possible, if not likely, that what you’ve seen is a deepfake product designed to discredit that person in the public eye.
The term deepfake originates from the technology called "deep learning", a form of artificial intelligence. Thanks to it, it is possible to create false images and false videos, but it can also be used to build and validate fake news and carry out acts of cyber bullying or revenge porn.
Deepfake: how it works
One of the most used methods to create a deepfake is by using an autoencoder, i.e. an artificial intelligence program that studies a video to understand how a person looks from different angles and spatial conditions , and then map it to the individual in the target video by finding common characteristics.
To take advantage of this technique it is first necessary to have a starting video to use as a basis for the deepfake and then a collection of other videos of the person you want to include in the target.
A second method to make deepfakes is through a GAN, (Generative Adversarial Network). This system is able to make a deepfake much more realistic, as - through the constant search for defects in the shots - it makes it more and more believable.
The amount of data needed to generate a deepfake using GAN is very large, so not everyone can make one using this method. For those who have enough to be able to use the GAN method, it is necessary to know that the more data entered into the system, the higher the quality of the generated content will be.
Deepfakes: how they are created
To make these Deepfake videos, many people download apps from Github, an open source portal where you can find some of them. Among those available are Deepface Lab or Face Swap; while online there are websites such as deepfakesweb.com and yepic.ai, which are widely used to create deepfakes.
Currently creating a deepfake is not considered illegal, although a well done one can ruin a person’s reputation. Furthermore, what is worrying is that - probably - with the progressive development of technology, even deepfakes will become increasingly difficult to recognize.
How to recognize a deepfake
There are no unique rules for recognizing a video made using the deepfake technique, especially since their content quality has grown over the years.
In the past, deepfake videos were recognizable because they were of low quality and contained obvious editing errors, traceable for example by the unnatural movement of the eyelids, by the bad quality of the audio, or by the lights, often too high or too low .
Thanks to user feedback, deepfake apps are updated periodically, which makes them more accurate and performing. As a result, distinguishing fake content from genuine content becomes more difficult.
Deepfakes and fake news: what awaits us in the future
The artificial intelligence market is in great evolution, in fact new softwares are being born that allows you to create content based on the knowledge of advanced algorithms.
Among those born recently there is Dall-E 2, a program available on the web that can generate images from scratch or modify them. In particular, with Dall-E 2, it is possible to alter existing images at will, although sometimes the result may not live up to expectations.
This possibility opens up new questions that it is impossible not to ask: tools such as Dall-E 2 are available to anyone with an internet connection, which implies that anyone can use it to create false images which, together with the deepfakes can contribute to ruin the reputation of a person.
Original article published on Money.it Italy 2023-01-04 10:00:00. Original title: Deepfake: cosa sono, come funzionano, come riconoscerli
