Deep fake

Deep fake

Subject: Science and Tech

Context: In a new documentary, Roadrunner, about the life and tragic death of Anthony Bourdain, there are a few lines of dialogue in Bourdain’s voice that he might not have ever said out loud.

Concept:

Deep fake technology can seamlessly stitch anyone in the world into a video or photo they never actually participated in. Deep fake technologies new automatic computer-graphics or machine-learning systems can synthesize images and videos much more quickly.

Deep fake making,

• Deep fake video of someone, a creator would first train a neural network on many hours of real video footage of the person to give it a realistic “understanding” of what he or she looks like from many angles and under different lighting.
• They’d combine the trained network with computer-graphics techniques to superimpose a copy of the person onto a different actor.
• The addition of AI makes the process faster than it ever would have been before, it still takes time for this process to yield a believable composite that places a person into an entirely fictional situation
• AI encoder runs several face shots of two persons. Then, the encoder learns of similarities in the faces and compresses the images.

The rise of ‘deep-fakes’ and threats to truth

• As the faces are different, one AI decoder picks the first person’s face, and another decoder picks the second person’s face. Then, the encoded images are fed into the “wrong” decoder to perform the face swap. This process is done on every frame to make a convincing video.
• In certain other cases, the AI is trained to create new images and videos from scratch using Generative Adversarial Network (GAN).
• GAN, with the help of Machine Learning, makes two neural networks to contest against each other in a game with a given training set. This helps generate new data and output related to the training set.
• A class of deep-learning algorithms called generative adversarial networks (GANs) will be the main engine of deep fakes development in the future. GAN-generated faces are near-impossible to tell from real faces.

Effects

• Destructive effect on the social fabric,
• Non-consensual pornographic deep fakes other problematic forms.
• Deep fakes may well enable bullying