WHAT IS A DEEPFAKE?
Deepfakes are fake videos or audio recordings that look and sound like the original but are not. It combines the concept of machine or deep learning to create something that is not real.
The underlying Artificial Intelligence (AI) technology is used to manipulate both video and audio to accurately impersonate individuals.
Deepfakes have grown increasingly sophisticated over the years. They have caused serious financial damage, manipulated public opinion with fake videos and images, promoted political agendas or gain competitive advantage, and caused reputational damage to high-profile individuals.
HOW DO DEEPFAKES WORK?
Deepfakes tech uses Generative Adversarial Networks(GANs) and two Machine Learning(ML) models to generate the forgery.
The auto-encoder tries to reconstruct the input image. It learns a lower representation of the input image and uses it to swap with the target. One model trains a dataset and then create video fakes while the other model attempts to detect the fakes.
ML Model 1: Trains dataset creates forgeries
ML Model 2: Detects forgeries created by Model 1
- The two models consist of neural networks that engage in a complex interplay that interprets data by labeling, clustering, and classifying
- There is a checkpoint for when a fake is detected by the second model
DEEPFAKE ALGORITHM
The ML algorithm examines the images at a pixel level. The goal of the autoencoder is to convert an image to an image vector.
Each pixel is compared to the ones directly surrounding it.
The pattern from each pixel to the next is traced by arrows from where the image is light to where the image gets darker across the entire image.
The image is then broken into smaller squares and the total direction of arrows is counted. Each square is replaced with the direction that has the most counts. This turns the original image into a basic representation that captures the structure of the face.
The image’s pattern is compared to another image and if they match, the face is detected. The model now has a dataset to train with.
USES OF DEEPFAKES
Deepfakes made with malicious intent capitalize on the fact that, for most people, seeing is believing. However, the movie industry has used its technology for special effects and animations. Examples:
Karaoke impressions of your favorite singer.
Editing facial impressions to make them photorealistic. Bring portraits to life.
Dubbing audio for movies in different languages.
IMPACT OF DEEPFAKES
Deepfakes have had negative impacts such as
- Invasion of privacy
- Social engineering
- Scams and hoaxes
- Celebrity pornography
- Election manipulation
- Automated disinformation attacks
- Cybercrime such as identity theft & fraud
Malicious deepfake technology use is a growing cyber threat
