Earlier this month, the arrest of a Taiwanese YouTuber drew attention to a technology known as deepfake. A YouTuber known as Hsiao Yu and two others were arrested on charges of making deepfake sex videos targeting celebrities and politicians. Deepfake is a technology that can make realistic videos of people doing things they never did. It uses deep machine learning to make fake videos so convincing, they can fool not only the naked eye, but also the software designed to detect them. Today in our Sunday special report, we take a closer look at this rising technology and its alarming implications for democracies worldwide.
Every year on Dec. 25, Queen Elizabeth II of England delivers a message to her people.
Elizabeth II
Queen of England
For nearly 70 years, I have kept a tradition of speaking to you at Christmas.
Sharp-eyed observers might find that this Christmas message is not quite like the rest.
Elizabeth II
Queen of England
Like many of you, when I wasn’t settling down with my husband to Netflix and Phil, as I like to call it, I was perfecting my moves for TikTok.
The monarch tells a joke before beginning a dance routine. Believe it or not, she is not the queen. Behind the curtain is actress Debra Stephenson, who little resembles the royal in appearance or in age. She is transformed through the miracle of modern technology.
National Cheng Kung University’s Hsu Chih-chung許志仲 is one of the few scholars in Taiwan who study face recognition and synthesis technology. He said the queen’s alternative Christmas message was created with a tool known as deepfake. The defining feature of deepfakes is that it’s not important who the actor is. If enough data is fed to the computer, any person can become anybody they want to be.
Prof. Hsu Chih-chung
National Cheng Kung University, data science
In most cases of face swapping, it doesn’t take much for an actor to satisfy the basic requirements. The two people do not really need to look alike. But some basic features should be similar. For example, light and shadow. For example, their skin color might need to be the same. Their appearance shouldn’t be too different.
Lee Chun-yi and Chen Kuan-lin are both Professor Hsu’s graduate students. They show me how it’s done. Images and videos of their faces are loaded into a facial synthesis program, which analyzes and deconstructs the data. Once the computer has learned each of their facial features, it’s able to superimpose one student’s face onto another.
Through the power of artificial intelligence –and specifically what’s known as deep learning – an ordinary laptop can generate sophisticated fake videos that could easily lead the way to fraud.
This tech also exists in the form of mobile apps that let the user replace their face with a celebrity’s. It’s fun to play around with, but it could also be dangerous in the wrong hands. Once a program is trained on a large dataset of real images, it can create footage of you saying things you never said.
In this video, actor Jordan Peele impersonates Barack Obama, making him give a speech about fake news. The video was made to warn about the dangers of deepfake technology, which has been used to create fake celebrity porn.
。
A sultry Emma Watson appears in this video that’s nearly impossible to spot as a fake. But it is. Created by a computer, this is one in a tide of sexually explicit deepfakes that can ruin lives and reputations.
Prof. Hsu Chih-chung
National Cheng Kung University, data science
Photos and videos of public figures are very easy to obtain. So using synthesis tools, it is very easily to achieve a realistic synthesis. Besides, when we’re watching videos or looking at digital information, we usually just glance at them. We don’t examine the details with care. So this information can propagate very easily to all corners of the world – social media makes it so easy. A piece of information can suddenly spread out in a swarm, and people don’t have time to distinguish whether it’s real or fake.
Public figures have public influence. When that influence is misappropriated by a deepfake, it can have far-reaching social implications.
Hu Yuan-hui
Chung Cheng University, communications dept.
During an election, if someone makes a deepfake video and distributes it for a political purpose or some other purpose, that’s very likely to have an impact on the election. It could give rise to an unfair election.
Chen Hui-min is editor-in-chief of the Taiwan FactCheck Center. After the latest U.S. presidential election, she and her team noticed an online video accused of being a deepfake. It was footage of Trump conceding his defeat, and not a few of his supporters were convinced it was created by AI. But because this video was distributed by major U.S
#台灣新聞 #TaiwanNews #民視新聞 #FTV新聞 #Taiwan
