Deep fake is a technique whereby using artificial intelligence one can easily manipulate the images, audio, and even videos of a person. The techniques of machine learning and artificial intelligence are being creatively used to manipulate such audios and videos which can easily deceive anyone. Autoencoders and generative adversarial networks are being used to base the deep learning and training of self-generative architectures of neural networks. Indeed it’s a revolutionary creation in sense of improving technology but in concerns of the negative uses of this technology, its use has been restricted by the government.
History of Deep-Fake:
The concept of photo manipulation is deeply rooted and way back in time since the 19th century. It was being used in motion pictures. With the advancement in technology since the 20th century the technology improved with the introduction of video digitalization.
A program was introduced in 1997 which was Video Rewrite that leads to the first start of using the already existing video of a person and manipulating its audio so that the person appears mouthing but speaking different words which were extracted from different audio. This was the first-ever reanimation that makes a connection between the audio and the subject face. This paper was published by Christoph Bregler, Malcolm Slaney, and Michele Covell in which they describe this innovative program which can later be used in movies. The results produced from this program were shockingly amazing but short. This was a breakthrough in the history of Deep fake. Many new programs are just upgraded algorithms from the philosophies of this published paper-like Final Cut and premier pro.
In the 2000s improvements in technologies of facial recognition were being made which now makes the deep fakes so convincible to people. In 2001 Gareth J. Edwards, Christopher J. Taylor, Timothy F. Cootes debuted a published paper related to the models and algorithms of Active appearance models (AAM). A statistical method was being used to track facial movements and match the shape to an image. It was quite famous at that time and proves to be the step forward while creating deep fakes.
Later on, projects like Face2Face by the Technical University of Munich which majorly focus on only replacing the area of the mouth of the target video in which the actor performs, and the Synthesizing Obama project by the University of Washington were remarkable and proves to be consumer-grade level hardware. Both of these projects have improved graphics which made an image very close to reality so that it’s almost undetectable to cross a line between real and fake.
Face2Face doesn’t provide the audio but there are many other ways to get human synthesized audios. While the Synthesizing Obama project is mere the latest version of the old Video Rewrite and is called Video Rewrite 2.0 with the better and latest animations, impressions, and textures. This is the more realistic technique that matches the color, dimples, and even wrinkles. The movement of lips also synchronized with the movements of eyebrows.
Later on, in 2018 the concept of deep faking was broadened to the whole body when the researchers at the University of California, Berkeley in August published a new paper that introduces the Fake dancing app that can establish a notion of skillful dancing using the technique of Artificial intelligence.
How to create a Deep fake?
Well to create a deep fake two main steps are required which include encoder and decoder. With the use of encoder various shots of two people are run using an AI algorithm. The main purpose of this encoder is used to find the similarities and the common features of two people and then compress them together to make a manipulative image. The decoder then runs another AI algorithm to recover these images. Both images of person X and person Y are decoded in this way. In face swap technology one just need to put the encoded image of Person X to the decoder of Person Y. Decoder then just simply reconstruct Person Y face using the expressions and the positions of person X on each frame to make a decisive video.
There’s another method of creating a deep fake is in which Gan pits use two AI algorithms that work against each other. Generator and discriminator work in this process. Random noises are fed countless times into the generator which develops it into synthetic images and then these images are fed into the generator. At last, repeating this process countless times develop manipulative realistic faces of people.
With advancement work, it becomes easy even for beginners to make a manipulative deep fake these apps include Faceapp, Deep face Lab, Chinese app Zao and Face swap app.
GitHub also provides different deep fake software. But most high standard software’s are used for entertainment purposes that’s why it’s almost impossible to restrict the use of this software but it can also be used for negative purposes.
Uses of Deep fake:
The technology of Deep fake was created to benefit mankind and to be more successful in terms of technology and media but people exploit this technology as well. Here are some enlisted beneficial and harmful uses of Deep Fake.
- As an Art :
Deep fake technology has been used by Moscow researchers to give a life-like appearance to Mona Lisa’s portrait in which she can move her eyeballs and feels like looking here and there. A video of David Beckham was released by a UK charity to deliver the anti-malarial message in approximately 9 languages. Advertising companies are also taking benefit from this technology like WPP created a presenter that speaks the language of the recipient and even calls them by name.
Peter Cushing has introduced again in Rogue One which is A Star Wars Story by using artificial intelligence. This technology even allows the creators to change a scene or audio back in time without re-shooting the entire scene.
A hologram of holocaust survivors can be made by syncing their audios with their deep fake avatars which makes a person an impression of directly talking to them in reality. A 3D face model image of a person can be created without even #D-scanning him.
- Research Purposes :
Fake brain scans are being created to use algorithms and train them in the detection of tumors based on real patient data. Generative Technology is playing a major role in the field of medicine using Deep fake.
This technology can also be used to create the private avatars of people which can be used in apps and allow people to choose different clothes, styles, and hairs to know which suits them better.
- For Training of professionals :
Deep fake technology can be used to make the avatars of real people train the professionals in their jobs within the era of COVID-19 and social distancing where it is quite difficult to do training in person.
- Fewer investments :
As cheap creation of ads, games, and videos are possible by using these generative technologies it allows companies to low their financials in these areas and earn more profit.
- Help People with Disabilities :
Deep fake technology and Artificial intelligence are together put in work by Microsoft through which they develop “Seeing AI” app which assists blind people and people with low vision to narrate the world in terms of describing locations, recognizing people and even scanning documents, etc.
The idea of creating fake images and videos creates a whole new era of propagating fake news. Like Mark Zuckerberg’s deep fake video in which he was showing creating a secretive agency for social network success.
Many actors’ deep fakes were created to form pornographic videos like Scarlett Johansson etc. Deep nudes’ site was being proposed but due to ethical reasons, it got canceled. Videos being shared on Reddit related to deep fake pornography. This technology has been widely used for embarrassing people in terms of making funny videos which later got criticized by the audience on social media. TikTok’s got famous for Tom Cruise’s deep fake videos.
The copying of the same audio and even video money fraud cases has been increased because of Deep fakes. Scammers copy audios of company CEOs and convince people to transfer huge amounts of money to their accounts. They got successful in defrauding $243,000 from a British company this way last year. Employees have been tricked by using the same method in sense of giving sensitive information and even passwords to the scammers.FBA and other theft control agencies are working to identify these scammers and to decode such deep fakes but scammers got an upper hand as they have looted a lot of people since the advancement in this technology. People lose faith in news and other informative videos because of this.
- Political Controversies:
Many political controversies are being created by the Deep Fake technology like the replacement of Angela Merkel’s face with Trump’s and Mauricio Macri’s face with Hitler’s face. Even Trump’s deep fake video in which he seems taunting his skin color and appearance during the address of the Oval office was being aired by Fox and later results in the firing of the responsible employee.
How one can detect Deep fakes?
Even though this technology has gone too far in terms of advancement there are still some glitches like the faces in these videos are being created using a 2-D face so the image feels a bit distorted in the 3-D video. DARPA has been working to identify such fake videos. If one takes a keen look at these videos he can notice certain points and easily detect whether it’s a fake video or not like:
- Deep fakes have some kind of jerky movements in them.
- Face could be seen blur than the whole environment.
- The shifts in skin tone and lighting are not accurate.
- The subject could be appearing blinking weirdly or even no blinking at all.
- The syncing of lips could be matched imperfectly with the audio.
In concerns with the harmful aspects of deep fakes, various countries introduce laws and acts to control the deep fake use.
Canada Communication security establishment released a report with concerns about the use of deep fakes to discredit politicians and voters, such deep fakes could interfere with the country’s politics. Citizens are provided with many ways to deal with such deep fakes if they got targeted by them. In UK deep fakes could be prosecuted in terms of harassment. China Cyberspace administration stated deep fake as a clear crime on their site.
The United States introduced Assembly bills on October 3, 2019, as Bill No.602 for the deep fakes of sexually targeted people without their consent that allows them to take legal action against the content creators and no. 730 prohibit the spread of a targeted running candidate malicious audios and videos in a public office within 60 days of the election. An Act against Malicious Deep Fake prohibition was being introduced in US Senate in 2018 and the Deep Fake Accountability Act was introduced in 2019 in the U.S House of Representatives. The U.S States like Texas, New York, Virginia, and California also introduced the legislations separately.
As the world is becoming more synthetic within the advancements and novel achievements in technology there are possibilities of even way better future work in Artificial intelligence. It is almost impossible to limit the human brains in a bubble. There is a possibility that one day it becomes impossible to differentiate between what’s real and what’s fake. A slight estimation of total deep fake videos is almost 15,000 but who knows what the future holds and how rapidly such videos will increase in number with the adaptation and evolution of this technology. So it is better to have a skeptical approach towards the latest technologies and to dually check the credibility of the authentic source of information without spreading it to others because it is an individual responsibility to protect others from such malicious software.
Article author: Maryam00000 (Fiverr.com)