Contents
How do deepfakes work, why are they needed and are they useful? These and other questions were answered by the host of the Industry 4.0 YouTube channel Nikolai Dubinin
Fake elections and “resurrection” from the dead
More recently, at the turn of the XNUMXth and XNUMXst centuries, the Photoshop program was super popular. Now it has been replaced by neural networks, namely Deepfake. This technology is much more dangerous, but also more promising than the previous ability to roughly cut out people’s faces from photographs and paste them on other pictures.
Deepfake technology is able to combine photographs of a person and, for example, make a video out of them. Deepfakes have long penetrated politics. For example, in Mexico in 2017, the drug mafia killed journalist Javier Valdes Cárdenas, who wrote about crime and the drug trade. He was “resurrected” with the help of a deepfake, and he was able to address the president of the country with the following words: “Today, Mr. President, you have the opportunity to stand out from the crowd of your predecessors and finally make real changes – give us justice and fairness. A country without truth is a country without democracy. Despite the fact that they want to shut our mouths, we keep talking!”
Another story happened in the 2020 Indian elections. In Delhi, the leader of one of the opposition parties, Manoja Tiwari, made a statement in Hindi and the Haryani dialect. However, he did not know how to speak this language. Manoja Tiwari decided to deceive most likely out of a desire to win the votes of migrant workers in Delhi. They just speak the Hariani dialect.
And there are already many such cases. What is interesting about the story of the leader of the Indian opposition is that he never admitted to using technology: it is quite difficult to prove this, so it is not necessary to admit it.
Another example is the BBC film “Welcome to Chechnya” about activists who criticize the government. The creators of the picture did not “blur over” the faces of the opposition, but superimposed deepfakes of activists living in New York on them.
Is it possible to detect deepfakes
The difficulty in fighting the dishonest use of technology lies in the fact that fakes will only get better.
Deepfake is a generative adversarial network. It has a generator and a discriminator. The generator creates a fake, and the discriminator criticizes and points out its shortcomings. After these remarks, the generator starts working again. As a result, a very accurate fake appears.
Some countries have decided to regulate deepfakes by law.
- In China in 2020, they introduced criminal liability for misleading people with the help of deepfakes.
- California passed a law banning people from creating deepfakes using artificial intelligence to mislead voters during elections.
- In state Virginia The Revenge Porn Act was passed banning the use of deepfakes in pornography.
- Criminal Code France provides for punishment for publishing a montage made with the words or image of a person, without his consent, if it is not obvious that this is a montage. A person posting a deepfake can be sanctioned unless they can prove that they sincerely believed the material was not a montage.
In 2020, the Universities of California and Stanford were able to increase the accuracy of detecting deepfakes to 97%. Scientists have studied visemes and phonemes.
visemes – a visual expression of how a particular sound is pronounced. Conditionally – we say something and our mouth somehow moves. Philologists are trying to compare visemes in deepfakes and in original videos. Indeed, visemes in deepfakes look a little different. If you train algorithms on such a data field, the accuracy of fake recognition will be quite high.
Another way to calculate deception is to study how noisy a real and fake video is. The noise in deepfakes is a bit different, and you can often hear it.
The other side of deepfakes
With the help of generative neural networks, you can create content using a minimum amount of data. From this point of view, the technology is not at all about the “transplant” of faces for fun, but with a practical purpose. Now it is possible to generate a video from one photo and text. The text is synthesized into audio, photographs are animated from the audio – as a result, we get a speaking speaker. Using this set of algorithms, you can create video content very quickly.
Deepfakes can “resurrect” your favorite actors. For example, Philip Seymour Hoffman and Paul Walker, who died while filming their films (The Hunger Games: Mockingjay and Furious 7, respectively), have been digitally recreated. And Carrie Fisher was “resurrected” for filming in the ninth episode of Star Wars.
Maria Chmir, CEO of Deepcake.io, claims that a tool is coming soon that will completely change interactivity. Today, all viewers are given is to choose the ending of the series. In the future, with the help of deepfakes, we can be inside the MCU with our friends. There will be an opportunity to change boring actors. After all, generative neural networks allow you to edit, modify and transform content in a significant way.
That is why deepfakes are not only manipulation and disinformation. With the help of them, it will be possible to shoot films with special effects, bring characters to life and, in general, add a new experience of content consumption.