Deepfake: Innovative or malicious?

Samuel Dorven

20 septembre, 2023

Deepfake: Innovative or malicious?

Deepfakes have entered our daily lives, arousing both fascination and concern. These faked videos and images, generated by artificial intelligence, raise essential questions about their usefulness and potential risks. In this article, we dive into the world of deepfakes to better understand what they are, how they are created, why they are used, how to detect them, and what the associated risks are.

What is a deepfake ?

Before we get started, a little etymology. The word "Deepfake" comes from the contraction of the English terms "Deep learning" and "fake", which can be translated as "hypertrucage".

And that's the art of deepfake: it works by using real images and videos as a database to create, for example, "face swaps". This practice involves taking one person's face and "pasting" it onto another's, giving the illusion of cloning. In 2018, a video made the rounds featuring a Barack Obama deepfake by American actor Jordan Peele (which you can see just below). The video proved just how powerful this new technology can be.

There are other ways of creating deepfakes, such as creating a person who never existed. The AI creates a new face that can be customized as desired (eye color, etc.).

How they are created ?

Deepfakes are created through the use of artificial neural networks called GANs, which consist of two parts: a generator that creates falsified content, and a discriminator that tries to differentiate between falsified and authentic content. Over time, the generator improves in its ability to produce higher-quality content, while the discriminator becomes more effective at spotting it, resulting in the creation of increasingly convincing deepfakes.

How to use them ?

Initially intended for creative purposes such as voice dubbing for the cinema or the creation of more realistic video game characters, deepfakes now have a number of practical alternatives. It can be used, for example, to generate voice-overs for podcasts and tutorial videos, create faces for promotional videos, or dub into different languages. Here are just a few examples of tools using this technology:

AI voice-over generation tools :
  1. An AI text-to-speech tool that can clone existing voices.
  2. Murf: Use realistic AI-generated voices or clone your own.
  3. WellSaid: Generate realistic voice narrations thanks to AI.
  4. Generate voice-overs in any language.

Quels sont les risques de deepfake ?

Deepfakes pose several major risks. Firstly, they can be used to spread false information, undermining public confidence in the media and politics. In addition, they can cause personal damage, by defaming individuals. Finally, deepfakes raise complex ethical and legal issues concerning privacy and consent, as they enable compromising videos to be created without the consent of the person depicted.

This is clearly illustrated in the "Defend Intelligence" video, where the youtuber gets President Emmanuel Macron to say false statements in order to spread false information. It's important to note that this video is not intended to undermine the president's integrity, but rather to make viewers aware of the need for caution when confronted with such videos. Its main message is to remind people not to take everything they see as absolute truth.