"Deepfake", AI technology and its application in various industries
In the midst of the development of artificial intelligence (AI) chatbots that have become better by leaps and bounds after the world enters the era of the Internet full of information. Another potential of AI to watch is the development of video and audio, which has evolved to the point that AI can now create videos that mimic the actions of people who are opposed to reality, or "deepfakes" (also known as Digital Humans, AI Avatars, and AI Clones).
The definition of such technology is quite straightforward, when AI simulates a video of a person to act or say something according to a prompt that the user wants, when in fact the person does not actually have to do such an action.
For example, "Bee the Ska" posted a video of herself blended with the face of "Pimry Pie" to live sell things through Wilson's Deepfake technology. The software has generated a surprising amount of positive feedback about the development of AI.
Or in the business sector. SCBX Public Company Limited Using generative AI to prepare content and videos to be presented to shareholders by the Company. It uses a large language model (LLM) to train historical meeting sequences, and then instructs the AI to write the script for the last meeting.
In addition, the system uses generative video to teach AI to understand the video presentation format of the previous shareholders' meeting, and then the system creates a video of the latest business results statement.
Steps to create a Deepfake
However, There are many ways to create a deepfake, but the most common is to use deep neural networks that use the face-swapping technique, or simply put, the user must have a "target video" as the basis for the deepfake, and then compile a video clip of the person you want to insert into the target video.
While the target video could be a clip from a Hollywood movie or a video of the person you want to insert your face into the movie, or it could be a random clip downloaded from any video platform.
Later, the AI in each program will predict what a person looks like from many angles and conditions that are all available. The system will then match the person with another person in the target video by finding the same features and creating the deepfake video that the user wants.
In addition, developers have tried to add another type of machine learning*, Generative Adversarial Networks (GAN)*, which carefully detects and improves flaws that occur in deepfake videos, making it more difficult for video spoofing detectors to decipher.
Although the whole creation process sounds complicated, the current software is quite easy to use and many current applications can create deepfakes with just a few clicks, such as applications from China, Zao, DeepFace Lab, FakeApp, Face Swap, Heygen, D-ID, StableDiffusion, etc. Deepfake-as-a-Service ของ Tencent Many more deepfake software can be found on GitHub, an open-source developer community.
Benefits of Deepfake Technology
1. Dubbing and Translation
Deepfake technology can also be used to benefit many other industries, for example, users use Deepfake technology to create videos by David Beckham. Speak out for a campaign to end malaria in โครงการ “Malaria Must Die In the video, Beckham speaks the campaign script in nine languages with a mouth shape that moves perfectly in harmony with each language spoken.
Beckham's video demonstrates the potential of using Deepfake technology for movie dubbing. An actor's facial movement manipulation can make it seem like they're saying a sentence they didn't say in real life.
More importantly, this concept can be extended by using AI to control the voice of an actor in multiple languages instead of the usual voice-over, which will make it more interesting because the voice of the person is original.
Importantly, there is now technology to manage voice, for example, Lyrebird AI that uses AI to clone human speech. Paired with Deepfake's technology, the result is a "voice-over tool" with enormous potential depending on the user's preferred language.
2. Film industry
Traditional video production often involves the busy work of creating real scenes, which costs money for materials. But by using Deepfake technology for visual effects, filmmakers can easily create virtual scenes and backgrounds. The technology will help filmmakers change backgrounds and adapt to different screenplays without transportation challenges. Labor or additional costs
In addition, Deepfake technology allows filmmakers to imagine new situations with actors who are no longer able to return to the movies. For example, a surprising appearance. of Luke Skywalker In the finale of The Mandalorian season two, which caused the fan base to flood one after another.
Importantly, artificial intelligence technology can now create A clone of a famous Hollywood actor. Or even sketch a digital clone of a person neatly through the original storage of the person in four steps: audio recording, audio recording, and digital cloning. With a digital scanning tool like "Dorothy", the originals can be combined with the target video and created into various characters.
More importantly, this 360-degree camera captures images of people's faces in almost every posture, whether it's smiling, opening their mouths, sticking their tongues, or rolling their eyes back and forth to make the deepfake video as virtual as possible, and every word the clone says moves with the correct mouth shape.
3. Thought Leaders (KOLs)
The thought leaders of Deepfake technology are very useful, especially in terms of direct communication according to the scripts designed by the developers, as well as reducing personnel costs, and they can work 24 hours a day.
One interesting example is Deepfake's technology. Xiaoice These deepfake KOLs have all contributed to making the business sector more efficient through reducing the company's costs and creating brand recognition through thought leaders who do not have the opportunity to change jobs or be bought like real people.
4. Education
In 1963, President John F. Kennedy was on his way to give a speech in Dallas, United States. After all, Kennedy did not deliver the speech he had prepared. However, with modern technology and the development of AI, we can hear Kennedy's statement today. The CereProc team analyzed 831 recordings of his speeches to "create a Kennedy voice."
The sound is separated into 116,177 units of sound, which is quite challenging because the recorded sounds are created on different types of devices at different times. But in the end, CereProc was able to use AI to completely create Kennedy's realistic voice.
For example, developers can create new videos of historical figures to tell their own stories, for example, last year, the Illinois State Museum and Center for the Study of the Holocaust exhibited a rotating hologram of 15 Holocaust survivors. Visitors have the opportunity to ask questions to the holograms of survivors. From the original video from a spherical camera that recorded interviews with survivors, each of which took 5 days to film.
5. Business
The business sector is one of the biggest beneficiaries of Deepfake technology, a notable example of which is the introduction of generative AI in SCBX's most recent shareholder meeting.
However, there are still many examples of the use of Deepfake technology in the business sector. Heygen, an AI avatar creation platform For example, using AI avatars to report business results, using avatar technology to train new employees, or creating videos to sell jobs to customers, etc. The advantage of such technology is that hundreds or thousands of videos are created for customers at a time with just a few clicks.
One example of the use of AI in the business sector today is in the midst of an increasingly isolated society. Therefore, in recent years, dating applications have become more and more popular. When the concept of a dating application was combined with Deepfake technology, it was born. Candy.ai A platform that simulates an avatar for users to talk to AI as if they were friends or lovers. If technology develops more in the future, users may be able to make calls using voice or through video calls with AI avatars that users want and are more virtual.
In addition, businesses that earn money through teaching or training can take deepfake videos and enter commands and create instructional videos without the need to record the videos themselves in the studio. "Chance the Rapper" who came to teach the audience about the work of deep artificial intelligence.
The Dark Side of Deepfakes
However, although this technology has created benefits for many industries and businesses, on the other hand, deepfake technology has been used to create a lot of trouble. For example, fraudsters use such technology to create videos or audio clones of stock experts in Thailand. One of them is "Dr.
Niwet Mewachirawarakorn" and then deceived the general public into investing.
In addition, in recent years, many fraudsters have used deepfake technology for illegal purposes, including to create pornography without their consent, until the US Federal Bureau of Investigation had to issue a public notice in June 2023 to warn the public about the dangers of generative AI and their use of those tools to create nude content. Cyber sexual extortion and harassment
In 2017, a Reddit user named "Deepfakes'' created an account to spread pornography that uses the faces of celebrities. According to Deeptrace, pornography accounted for 96% of deepfake videos found online in 2019.
In addition, deepfakes are being used for non-sexual criminal activities, such as in 2023 involving the use of deepfake technology to: Imitating the voice of a lady's child to intimidate and extort her.
The technology is also used in politics. For example, in 2018, a Belgian political party published a video of former US President Donald Trump. In a speech calling on Belgium to withdraw from the Paris climate agreement. However, Trump never gave that speech, and it was all a fake.
A similar incident occurred on the Users create videos of national leaders. Like Barack Obama, Donald Trump, Joe Biden or Vladimir Putin. The above points have led many analysts to believe that deepfake technology will change the landscape of global politics because people will have to question more and more about how to make a difference. Are the images and videos they see or hear real?
How to detect Deepfakes
1. Does the video detail look blurry or vague? One of the techniques for detecting deepfakes is to look for skin, hair, or face that may look blurrier than the environment in the video, or to consider skin that looks unnaturally soft.
2. The light doesn't look natural, right? From the beginning, it was explained that a deepfake video is made up of two or more videos, so the lighting in each video will not be the same. Therefore, such a way to check a video is to look for an anomaly in the direction of light.
3. Do the words or sounds match the picture? The audio may not match the person's mouth shape in the video, especially if the video is spoofed, and the pronunciation often does not match.
4. Does the source look credible? One technique that journalists and researchers often use to determine the true source of an image This can be done instantly by searching for the image backwards on the web like Google Lenses, as well as checking who posted the image. Where to post it, and does it make sense to post the video?
However, the issue of detecting deepfake videos is still a matter of debate among people in the tech industry because the development of AI is constantly moving to the extent that Wall Street Journal reporter Create your own avatar and test it with the bank's Voice Biometric system by saying your first and last name, followed by the journalist's email with your avatar. After all, the voice, which is generated by Deepfake technology, can also pass through the voice authentication system.
This issue has led many analysts to worry that if in the future artificial intelligence systems can create clones of people and can verify their identities through facial recognition and voice recognition, it may be a tool for fraudsters to use to commit illegal activities in the future. For example, deceiving or impersonating celebrities through video or audio cloning and then deceiving money from innocent people.
Deepfake Regulatory Measures
In order to prevent the potential damage caused by the advancement of artificial intelligence technology last year, the Chinese government enacted laws regulating the use of deepfakes, including: The consent of the model must be obtained, the video must not be used to disseminate fake news, the system must be strictly verified before creating a simulated video, the producer must clearly watermark the video as a simulated video, and it is forbidden to create video content that violates national security laws.
in The case of Thailand Legal scholars are of the opinion that the Computer Crime Act can be used. In Sections 14 and 16, it is stipulated that the use of a photograph of another person, regardless of whether the image is created, edited or added, or by any means. If it is republished and causes embarrassment, it will be punishable by imprisonment of up to 3 years and a fine of not more than 200,000 baht.
Meanwhile, the European Union (EU) was the first country to enact legislation to regulate the use of AI in general, which has come to control concerns about the use of deepfake technology to verify identities and conduct transactions on behalf of real people.
The EU's AI ACT Law prohibits six major prohibitions: Biometric Categorisation Systems, Untargeted Scraping of Facial Images, Emotion recognition, Social Scoring, AI designed to guide human actions and restrict freedom, and AI for the purpose of exploiting a person's imperfections and vulnerabilities.
Terminology
- Deep Neural Networks = Learning characteristics of artificial intelligence that mimics human learning through data recognition. The system repeats the pattern of a trained dataset to predict future trends.
- Machine Learning = Making machines, which are computers or electronic devices, able to learn information on their own through training with big data.
- Generative Adversarial Networks (GAN) is the learning and media creation of artificial intelligence through the combination of original images and deep learning of the machine brain and then created into A new set of photos, videos, or songs.
