AI-generated videos have turned out to be more convincing and common in this new age and it’s turning out to be quite scary.
What is a Deepfake?
There is a video of Barrack Obama calling Donald Trump a “complete dipshit”; or the one where Mark Zuckerberg admits to “total control of billions of people’s stolen data”. These are some of the deep fakes that we have out there.
Deepfakes use a form of AI called deep learning to come up with images of fake events. Do you want to sabotage your favourite politician, dance like a professional or star in your favourite movie? Then it’s your time to shine with deep fakes.
What are they for?
In recent days, they have been created as pornographic content. People who create them use their favourite celebrities’ faces as well as porn stars to create content. In some cases, they have been used as a form of revenge porn. Beyond the terrible porn, there’s plenty of satire out there.
Are Deepfakes about Videos only?
No, with deep fake technology, you can come up with images from scratch. Moreover, audio can be deep faked too. People have come up with clones or voice skins of public figures.
How are Deepfakes made?
In just a few steps, you can face swap someone. First of all, you run thousands of face shots of the two people through an AI algorithm known as an encoder. Encoder learns about the two faces and reduces them to their shared features and compresses the images.
A second AI algorithm called a decoder is then used to recover the compressed images. You train a decoder to recover the two different faces. To make the face swap you feed the encoded images into the “wrong” decoder.
A good example is that you take a compressed image of person A and feed it to the decoder trained on individual B. The decoder reconstructs the face of person B with everything and even expressions.
You can use various other ways too like a generative adversarial network (Gan).
How to spot a Deepfake
With technology improving by the day, it gets harder for one to spot deepfakes because as soon as a weakness is revealed then they make the corrections. However, poorly made deepfakes are easy to know. The lip-syncing is poor and the facial skin can be seen to have patches. There’s also flickering around the face
