DeepFake

Introduction

Deepfake is comprised of two words 'DEEP' and 'FAKE', where Deep stands for Deep Machine Learning, a type of AI tool, and Fake specifies the generated data. Deepfake is basically generating or manipulating fake videos, images, or speeches.  The concept behind generating or manipulating images or videos lies in feature recognition. In feature recognition, a machine reads and understands the features. Facial features like skin color, nose, ear, eyebrows, lips, etc, and for voice it focuses on pitch, density, word speed, amplitude, etc. For example on a base level, the filters in Snapchat do the same, it reads the facial features like eye, nose, eyebrows, forehead, and glasses, and applies the filter exactly in position but deep fake is far above this because in deep fake it almost impossible to recognize that which one is real and which one is fake by the human eye.

Types

• Faceswaps: It is clear from the name itself that it swaps between the faces of two humans, or it applies the face of a person on another just like copy-paste.

• Facial Attributes and Facial Expression Manipulation - With this technique facial features like skin color, gender, and age can be altered and it is also possible to alter facial expression using this tool.

• Face Synthesis - In this technique it is possible to create a  new inexistent face using the facial features of an existent one.

How It Works

It is a way complex technology but talking about the technical terms it usages " Encoder and decoder Neural Network. A neural network is a network of perceptron that can mimic a human brain, it reads and understands a form of data and enhances it's performance on its decisions and working.

• Encoder - Breaks down the small details of an image called compression.
• Decoder - This neural network built the manipulated or fake image from the compressed data passed by the encoder.

It also uses another technology called GAN (Generative Adversarial Tool ) which is used to create a nonexistent image. It is called Generative AI in lame language to create a deep fake video 1000 images of an individual taken from different angles are fed to the machine, and then the machine will reconstruct a suitable result using an encoder and decoder neural network.

Advantages and Disadvantages

Advantages: 

• The advertising industry can benefit a lot from this technology as it reduces costs and saves a lot of time in making ads. It can also make customizable advertisements that will be shown according to viewers' preferences and recommendations. For example, the Cadbury ad in which Shahrukh Khan promoted stores in Pan India and same with Hritik Raushan's Zomato ad.
  

• The entertainment industry can also benefit because as per the data, the videos present on the internet which is generated by deep fake for entertainment purpose. It can be used for content creation by active YouTube channels and Instagram handles, so they can generate and post on a regular basis and can also save time and money.

• In the movie industry, deep fake can be used to generate any model's role without being available for shooting, this can be done using their old footage. 

• It will also help to reduce the cost by replacing local voice-over artists who are required to dub a movie in any regional language fact deep fake will give a more real experience to the audience.  example - In Star Wars Peter Cushing played the role of ' Commander Moff Tarki" but the amazing face is Peter Cushing died in 1994. 
  

Disadvantages:

• The most dangerous aspect of deep fake is that it blurs the line between truth and lie. The world today is full of rumors or fake information and the advancement of deep fake technology will be proven catastrophic for truth.

• Cyber crimes and financial fraud will also increase due to deep fake videos. For example, a fraudster can target a victim by using their voice fakes for money the same has happened in 2019, where a fraudster was successful in making a CEO transfer 2 lakh pounds into a Hungarian account using a deep fake voice of the CEO of Germany based head office.

• Due to deep fake non-consensual pornographic content has also increased, as per the data 96% of non-consensual videos are fake.

• In this era where people have lost their patience and tolerance, this is a harmful tool that someone can think of. Nowadays a squabble upon ideology, politics, or religion turns into a violent clash or riot in no time deep fake will help to escalate this issue and it will definitely distort the social fabric.

Solution and Way Forward 

• It is easy to detect low-quality deep fake videos because they may have unnatural facial color, Different earrings in both ears, and Static hair. 

• MicroSoft, Google, and Adobe are investing in Deep fake detection technology. 

• As scientists are working to find a way to detect a deep fake video, there are new tools emerging in the market for deep fake editing, it will be a race between generation and detection of deep fake. So the solution lies here in finding a way of detecting real videos of the heap of fake videos.

• The government can also take some steps on the policy level, as the government will aware people of deep fake and its adverse effects under the digital media literacy program. Authorities can be set up to monitor deepfakes using blockchain.