Deepfake: A technological marvel and a danger to us all
Deepfake, dear reader, that Queen Elizabeth dances during her annual Christmas engagement, yes it is Queen Elizabeth as we know her and in her royal office, we are used to seeing her with her. Wouldn't that be strange?
Imagine waking up and finding a video of the President of America declaring war on China or Russia from his office in the White House in the voice we all know, what in the world would happen?
Or leave all that, how would you feel if we could bring a dead person back to life and you watched videos of them smiling just like they did when they were alive, expressing their opinions on current topics, or talking to you randomly.
I know you are now very surprised and ask yourself what are these fantasies, which will never happen, but unfortunately or fortunately this has already happened, this is a video of one of the characters of the school of surrealist art, the famous artist Salvador Dali who died many years ago.
There are also dozens of videos of former US President Obama and former President Trump saying strange things that do not come from either of them, but there are specialized programs from our technology that we will talk about today, the focus of her work is to make videos like this for these two characters in particular.
Deepfake or deepfake is this strange, amazing and scary technology that we are going to talk about today, the technology that makes it possible to do things that we previously considered impossible.
What is deepfake?
Simply put, deepfakes is a technology whose primary goal is to create videos - fake of course - through deep learning and artificial intelligence using pre-existing images, videos, and sounds.
You can say it builds new videos from the pre-existing, with it you can add words or delete others, and you can make the character or characters using this technique say things they wouldn't have said before.
You might think that this technology is science fiction, but on the contrary, it is very realistic, and it has been used a lot in recent years, both in cinema and in real life.
The last scenes of Paul Walker after his death in the series Fast & Furious were using this technique through previously filmed videos of him, and I bet you, dear reader, that if you watched the movie, it would not cross your mind. For a moment these videos are not real and Paul Walker came back from his grave to film them.
Deepfake technology is very popular in the cinema of course because we have many photos and videos of the actors, and there are famous incidents of using this technology to make fake videos of them, such as the collection of fake videos by artist Tom Cruise that flooded the Internet a while ago, and this is one of them:
Although deepfake technology is still relatively new, it is indeed powerful and dangerous; Anyone who knows how to make it can make videos where anyone says anything the video maker wants, or worse yet puts their face as a replacement for the face of a criminal in a crime video.
The danger of this technology lies in the fact that you can implement it through materials such as personal photos and videos on a person's accounts on social media, and for this reason, it is considered a great danger to humanity despite its great power.
One of the dangers of deepfake technology is that some bad guys use it to create videos or images that blackmail others, which is why social media networks try to confront and find out when they post it.
How does deepfake work?
Deepfake technology is based on artificial intelligence, specifically on Artificial Neural Networks (ANN), and more specifically on Artificial Neural Networks called GANs for short for Generative Adversarial Networks or Reproductive Adversarial Networks.
These artificial networks work like cat and mouse, and they are made up of two parts, the first is the generator that prepares the clip and the fake sounds, and the other part is the dummy detector “The Discriminator” that determines whether the clip is fake. or not.
The generator works on making the clips so that those clips pass over the error detector, it detects the errors in them and then returns them to the generator to improve them, and thus this process is repeated so much that the dummy detector cannot determine any error in the clip and then the clip production is completed.
This technique is dangerous and smart because it can teach itself, and here the dummy detector identifies errors and transmits them to the generator to reduce their repetition and thus the generator becomes better and better in this matter, making it able to generate very accurate real clips.
This is the process in a nutshell, but in the beginning, the fake video maker collects as much data as the person who wants to fake the clip for him, and this data is videos and images from all angles in addition to his audio clips.
The fake segment may be inaccurate at first or the fake detector used may be a bit weak which affects the final output of the process, then the process is called Shallow Fake.
We note here that the machine learning and deep learning used in this technology makes it very fast and able to teach itself, and with the development of these methods, we find that deepfakes technology has also become a more accurate technology and requires much less time, resources and data.
The GANs algorithm treats the fake clips they make like snapshots, as we all know videos are consecutive images that give us a sense of movement, so the algorithm separates them and processes them one by one. This is true manually, but only in theory, to fake, a five-minute video one needs to sit for weeks in front of photoshop in order to make a shoddy fake.
I should also point out, dear reader, that although the most commonly used algorithms what so far are the GANs we've talked about; However, new technologies and algorithms appear every day that are more accurate, powerful and require less data, but we focused on GANs because they are the most widely used.
Deep Fake isn't just one thing, it's a set of tools that allow exchanging voice, animate faces for still images, lip animate, body swipe, and text-to-speech, so there's no limit to what this technology can do.
But in my opinion, the most dangerous of these techniques is the technology that allows moving faces or swapping faces, because then you can target a person and make them do or say anything in a fake video that is difficult to detect, and this can destroy that person's life.
Impact of deepfake technology on society
Deep Fake technology is modern and advanced technology, and it can help us a lot in improving the reality of our lives and doing many impossible things. It is possible to bring the dear deceased person back to life and restore his smiles and dialogues, and you can meet the great dead characters, such as Einstein, Newton and others.
But that is only one side of the story, you can fake videos that condemn innocent people, this is really happening especially with politicians, artists, journalists and media people who have a lot of graphic material on the internet that easily allow fake videos to be spread. making them.
With the development of this technology more and more, professionals can create fake videos using a small number of photos, videos and videos, which makes the average person vulnerable through photos and videos on their social media accounts.
Many organizations and private companies today are trying to devise powerful technologies capable of identifying deep fakes, no matter how accurate and detailed they are, but we still have a lot to do...especially since the technologies they create are also developing very quickly.
Developed countries have also noticed this and are trying to put an end to it by raising awareness and imposing sanctions. Awareness here means educating citizens about this technology and its dangers, and not believing anything they see or investigating the source published through it.
It is also trying to legalize matters and put them within a specific legislative framework that allows it to preserve the security and safety of its citizens from any assault or technical assault that might harm them, which is what happened to many people at the beginning of the period. The rise of technology when they self-isolated out of fear of people's reactions to private videos. falsify them.
Social media is also trying to counter this, as these videos are usually posted through, so they have launched a filter that allows the processing of fake clips created with deepfake technology, and although this is still in its early stages, it is an important step for protection.
Facebook has hired a number of experts in deepfake detection from reputable institutions such as the universities of Berkeley, Oxford, and others, with the aim of building a detection system built into Facebook that allows it to identify these videos and prevent them from spreading.
Twitter has also made major tweaks and changes to its rules in order to identify fake videos, and this is what YouTube has done specifically to block any fake videos related to the 2020 US presidential election.
How do we identify deepfake videos?
There are many problems caused by deepfakes technology, there are many companies that have gone bankrupt due to fake clips or even fake audio recordings, and there is a famous story about a company that is causing a fake audio clip of its CEO to go bankrupt.
This was by a hacker sending him to the CFO, urging him to make an urgent money transfer of millions of dollars, and of course, none of this was true, the company lost all that money and then went bankrupt.
What is more dangerous than losing money is the outbreak of war between nations. If a video of a head of state inciting violence or attack on another country spreads... Here, that other country may make a decision to initiate the attack, and then war will start between them that will kill hundreds of thousands because of a fake clip.
Dear reader, do not think that this is pure fantasy, it has already happened to many important leaders and politicians, and the most famous of these incidents was in the state of Trump, who published on his account a fake video of Nancy Pelosi, Speaker of the US House of Representatives.
We must address here a very important and complex point, which is that there are - as we have said - two types of deep falsehood, namely the deep professional falsehood or the precise falsehood, and there is the superficial or deep falsehood that has many errors.
The second type (shallow) is easily detectable and sometimes detectable by the human eye due to a number of errors, such as:
- The clip or audio quality is very poor, which suggests that it was fabricated.
- There are errors in the processing of shadows in the clip, so we find that there are shadows of things that are not there or vice versa, there are missing shadows in the clip.
- Asymmetry in the movement of the body through the sense that something is wrong with the video or in the movement, and when compared to a real video of the person, this can be easily detected.
- The lip movements are weak and it is easy for you to spot errors in the lip movements while speaking and pronouncing the fake words.
- The style of the fake person, this is not related to the technique itself, but it is possible to happen in sloppy counterfeiting operations, which is the lack of professionalism in preparing the text and the appropriate movements with the person.
- Detecting the perfect fake videos is more difficult
Many of them are semi-realistic, and there is little chance of their falsity being discovered by very small, unnoticeable errors, or by more complex techniques devised specifically for this purpose.
Today, there are many algorithms that have been made to beat this technique, and they usually rely on checking every pixel in every picture or frame of fake clips just to determine the fake of these videos, but unfortunately, they are still in their early stages and they take a lot of time and power computing.
That is why it is a bit ineffective, because after you analyze the video frame by frame and pixel by pixel, the life of the video owner may already be ruined, and by that time it will be too late.
But it is effective in identifying some fake photos which are not so complicated. By searching pixel by pixel, you can identify defects in an image that technology cannot fake if deep fake technology is natural rather than super.
Deep Trace - a term that includes all forms of deep fake detection - is as much an arms race between nations as it is in weapons.
Many countries today are putting in place laws to criminalize the use of such technologies, and among the first of these are the United States of America, China and South Korea with Britain, which is in the process of preparing a law to legalize such dangerous technologies.