Artificial Intelligence (AI)-generated fake videos that can easily manipulate regular users are now a common thing, These videos have emerged as modern computers have gotten much better at simulating reality. For example, modern cinema depends a lot upon computer-generated sets, scenery, characters and even visual effects. These digital locations and props have replaced the physical ones as these scenes are hardly distinguishable from reality. One of the latest common things in computer imagery, deepfakes are designed by programming AI to make a person look like another one in a recorded video.
What are deepfakes?
The term “deepfake” has derived from a form of artificial intelligence called deep learning. As the name suggests, deepfakes use deep learning to make images of fake events. Deep Learning algorithms can teach themselves how to solve problems involving large sets of data. This technology is then used to swap faces in videos and other digital content to make realistic-looking fake media. Moreover, deepfakes are not just limited to videos, this technology can be used to create other fake content like images, audio, etc.
How do they work?
There are multiple methods for creating deepfakes, however, the most common one depends on using deep neural networks that involve autoencoders to apply a face-swapping technique. Usually, these are made on a target video that is used as the basis of the deepfake and then AI uses a collection of video clips of the person you want to insert in the target to replace the actual person in the video.
The autoencoder is a deep learning AI program that can study several video clips to understand what a person looks like from different angles and situations. By finding common features, it maps and replaces the face of the person with the one in the target video.
Generative Adversarial Networks (GANs) are another type of machine learning that can be used to create deepfakes. GANs are more advanced as they make it harder for deepfake detectors to decode them as it uses multiple rounds to detect and improve flaws in the deepfake. Experts believe that deepfakes will become far more sophisticated as technology develops
Nowadays, generating deepfakes is even easy for beginners as several apps and softwares help in creating them. GitHub, a software development open source community, is also a place where a huge amount of deepfake software can be found.
How can you detect deepfakes?
Online users have also become more aware and attuned to detecting fake news. For cybersecurity to enhance, more deepfake detecting technology needs to emerge to prevent misinformation from spreading. Previously, deepfakes were detected by following the blinking of the person in a video. When a subject never blinks or blinks very frequently or unnaturally there is a possibility for the video being a deepfake. However, newer deepfakes were able to overcome this problem. Another way of detecting a deepfake is by monitoring skin, hair or faces that may seem blurrier than the environment in which they’re placed and the focus might look unnaturally soft.
Sometimes, deepfake algorithms retain the lighting of the clips that were used as models for the fake video. The poorly matched lighting in the target video can also give away a deepfake. If the video is faked and the original audio is not as carefully manipulated, the audio might not appear to match the person.