2022's best place for Cybersecurity Insights and Advice for Everyone

#1 - Signup to our list and get regular insights and advice on how to be cyber safe.

By entering your email address you agree to receive emails from EveryDayCyber. We'll respect your privacy and you can unsubscribe at any time.

What is a deep fake video and how to spot It

by Cybergal | Last Updated | July 17, 2022
CyberSecurity - Consumer|CyberSecurity Insights

Artificial intelligence has been growing rapidly. It has done much to improve the quality and efficiency of many industries and our daily lives. 

However, not all AI creations are beneficial to society. Some technologies are developed that are later exploited by criminals. 

People can easily be fooled into believing they are seeing or hearing something that has no basis in truth in a culture filled with misinformation and deception. Deepfake videos have added to the uncertainty. Many are intended to trick the viewer or misrepresent the person in the video. Deepfake videos have forced viewers to wonder whether or not what they are seeing or hearing in a video or audio recording is real.

What is a Deepfake Video?

A deep fake video is a video in which a person’s face or body has been digitally altered to make them appear to be someone else. It’s usually used maliciously or to spread misinformation.

The term “deepfake” is derived from the underlying AI technology of “deep learning.” Deep learning algorithms are used to swap faces in video and digital content to create realistic-looking false media. Deep learning algorithms train themselves on how to solve issues when given vast volumes of data.

“Deepfake” is a combination of “deep learning” and “fake.” The “deep” part of the deepfake definition refers to deep learning, which is a method of teaching computers to think naturally like a human brain. The “fake” portion of the definition highlights the deceitful nature of deepfake media.

Identity Protection - McAfee Total Protection
Buy Now
We earn a commission if you make a purchase, at no additional cost to you.

Deep learning is an AI technique that includes performing a task over and over again, until it figures out the best approach to get a desired result. To create a deepfake, the AI is taught how to produce an animated rendition of a person’s face by feeding it hundreds of reference photographs and images.

How Do Deepfakes Work?

Machine learning algorithms can be used to create deepfake software in various ways. These deepfake algorithms can create content based on the information provided. If a software intends to generate a new face or replace a portion of a person’s face, the program must first be trained. The program is fed a large amount of data, which it then uses to learn to generate new data on its own.

The creation of synthetic videos is based on two common methods: autoencoder and generative adversarial network(GAN). 

Autoencoders

Autoencoders are artificial neural networks that learn to duplicate their own input unsupervised. They compress data in the same way that they were trained on. The output will not be identical to the input.

Autoencoding for deepfake video creation

An autoencoder has three components:

Generative adversarial networks

In machine learning, generative modeling is an unsupervised activity that includes autonomously detecting and learning the regularities or patterns in incoming data. The model can then be used to produce new examples that could have been drawn from the original dataset.

The GAN approach involves two sub-models: 

The two models are trained in an adversarial zero-sum game. The purpose of the generator is to fool the discriminator. When the generator defeats the discriminator around half of the time, it indicates that the generator model is producing plausible output.

#2 So here we are at the middle of the post. We still think it's a good idea to signup.

By entering your email address you agree to receive emails from EveryDayCyber. We'll respect your privacy and you can unsubscribe at any time.

How are Deepfakes Created?

Deepfakes are created by feeding a large number of images of a subject to variational auto-encoders (VAE). The goal is to teach the VAEs how to capture a wide range of matchable features, such as lighting conditions, positions, shadows, and emotional expressions. This allows the AI to distinguish visual features that are unique and features that can be replaced.

How deepfakes are created

Here are the basic steps in making deepfakes:

  1. The AI requires two sets of images: the genuine source (Person A) and the deepfake target (Person B).  Computers can be trained on an array of images of a specific person. 
  2. The AI creates the output images. To make the deepfake look real, the AI must identify and retain subtle signs that are unique to the targeted individual. 
  3. To carry out the face swap, the AI algorithm merges the output image with the data of the deepfake’s target face. The encoder then reassembles the movements and emotional expressions of Person A with Person B’s face. The AI must do the tasks frame by frame for the deepfake video to be truly believable.

What are the Uses of Deepfakes?

Artificial intelligence technology is unsettling to many people. It’s loaded with ethical issues and unknown future implications. The production of deepfakes is perhaps one of the more concerning capabilities afforded by AI advancement.

Disconcerting uses of deepfakes

Deepfakes cause us to lose faith in the information we see. Furthermore, they provide several options for mischief and criminal use. It’s difficult to imagine how we could ever hope to portray such functioning in a positive way, or how anyone could ever trust it. Here are some of deepfakes’ appalling uses:

Scams

Deepfake technology can be used by cyber criminals to create scams, bogus claims, and hoaxes that weaken and destabilize businesses.

For example, a crook could fabricate a video showing a senior executive engaging in unethical behavior or making false statements about the organization’s activities. This could have a significant impact on the company’s brand and public reputation, in addition to spending time and money to disprove.

Identity theft

Fake videos can be used to establish new identities by stealing the identities of real individuals. Attackers can use the technology to manufacture bogus documents or create deepfake audio to imitate their victim’s voice, allowing them to open accounts and make purchases in their name.

Blackmail against celebrities

Fake videos are dangerous media in the hands of people who want to extort or defame others. Our society, particularly the older generation, is still unaware that technology exists that can replace a face in a film.

Politicians are among the most frequently targeted by deepfake videos. Candidates’ ratings can be severely harmed by compromising videos implicating them in elections.

Deepfake pornography is a big concern that deepfake poses. Pornographic videos account for 96% of deepfakes on the Internet. The majority of deepfake technology focuses on revenge porn on celebs.

Automated spread of disinformation

Deepfake video can also be used to distribute automated disinformation, such as conspiracy theories and inaccurate political and social beliefs. A fabricated video of Facebook founder Mark Zuckerberg claiming ultimate control over billions of people’s data is an obvious example of a deepfake being used in this way. 

Social engineering

Audio deepfakes have been used in social engineering frauds to deceive people into believing trusted individuals have said something they did not say.

For example, the CEO of a British company was duped into thinking he was dealing with the CEO of the energy firm’s parent company in Germany. The person with the deepfake voice persuaded him to transfer €220,000 to a fictitious Hungarian supplier’s bank account.

Positive deepfake applications

Deepfakes do have beneficial uses. Deepfakes can be quite empowering when applied correctly. Because of advances in data science and artificial intelligence, new ideas and capacities for empowerment have evolved. AI has the potential to open doors for everyone, regardless of who they are or how they speak, communicate, or listen.

Education

Deepfakes can help a teacher give more interesting lectures. These lessons go beyond the usual visual and media formats.

Synthetic media created by artificial intelligence can bring historical people to life in the classroom. Educational deepfakes could make lessons more engaging and interactive. The impact of a synthetic movie of reenactments or a voice and video of a historical figure will be higher. 

Anonymity and privacy

Human rights advocates and journalists can use deepfake tools to remain anonymous under totalitarian and harsh countries. Citizen journalists and activists can use deepfake technology to report injustices on traditional or social media platforms. Deepfake can be used to safeguard the privacy of voices and faces.

The Arts and entertainment

Deepfakes can be used for artistic purposes. Entertainment companies have used high-end  deepfake technologies to create artificial but realistic worlds for fascinating storytelling. The video gaming industry is also fast catching up with AI-generated graphics to create a more engaging experience for gamers.

Medicine

AI-backed deepfakes can also provide benefits to the healthcare industry. For example, deepfake patients can be created at hospitals. They can use realistic patient data for testing and experimenting that does not endanger real patients. As a result, instead of using genuine patient data, researchers can employ deepfake patients.

How to Spot a Deepfake

How will you tell if the material you’re seeing or listening to is real or deepfake? Here are various self-help and AI-assisted methods to spot fake videos:

Awkward facial feature placement

You should be dubious of a video’s legitimacy if someone’s face points one way and their nose points another.

Unnatural facial expressions

When anything about a face doesn’t appear right, it could be a sign of facial modification. This can happen when a simple stitch of one image is done over another.

Stiff body, posture, or body movement

Another way to detect deepfakes is if a person’s body shape is unnatural, or if their head and body are positioned awkwardly or inconsistently. Because deepfake technology primarily concentrates on facial traits rather than the entire body, this may be one of the simpler anomalies to identify.

You should assume the video is phony if someone appears distorted or if they turn to the side or move their head. Or if their movements are choppy and disconnected from one frame to the next.

How to spot a deepfake

Unusual eye movement

You can also spot a deepfake with the following red flags: abnormal eye movements, a lack of eye movement, or the absence of blinking. It’s difficult to mimic the act of blinking in a realistic manner. Imitating a real person’s eye movements is similarly difficult.

Unreal skin color, hair, and teeth

Skin tone, discoloration, strange lighting, and misplaced shadows are all indicators that what you’re seeing is most likely fake.

Because fake photos are unable to generate these unique qualities, you will not notice frizzy or flyaway hair.

Individual tooth sets may be impossible for algorithms to track, so the lack of clear tooth outlines could be a clue.

Odd lighting and shadows

A deepfake video is prone to discoloration, misplaced shadows, and odd lighting, similar to the reasons behind artificial skin tones.

Inconsistent lip-syncing

A deepfake video will likely feature lip movements that are not in sync with the words being uttered by the people in the video.

Audio flaws

A person’s voice in a fake audio may seem weak and slurred when compared to an actual clip. A deepfake can also be identified by audio that is out of rhythm with the mouth’s actual movement.

Blurred areas

Another sign that you’re dealing with a deepfake video is soft or blurred areas, especially around the mouth. If you look attentively, you’ll observe hazy patches around the chin, cheeks, and jaws that become more evident during facial transitions and movement.

How to Fight Malicious Deepfakes

Deepfake tech presents a new type of challenge. It’s not always easy to spot a fake photo or video on the Internet, and detecting deepfakes will get more difficult.

Security and technology researchers have been working on detection technologies. Businesses are educating their employees on how to spot fake images and videos. Governments are crafting laws criminalizing malicious deepfakes, particularly those used for pornography, swaying public opinion, and influencing elections. 

How exactly can you protect yourself from deepfakes? Here are steps you can take:

Limit the information you share online

The simplest way to prevent someone from creating a deepfake of you is to limit the number of videos and photographs you upload on social media accounts and to restrict your privacy settings.

Educate yourself

Be vigilant and aware of recent technological advances and understand the challenges these technologies present. 

Check your sources

Make sure you’re using highly reliable sources. If you’re unsure, double-check the facts on several trustworthy platforms. When it comes to content that has surfaced from unknown sources, be skeptical and do more research.

Take advantage of reverse image search

Without getting too technical, reverse image search might help you recognize a fraudulent video or social network profile. Similar photographs may be found online that can help you get the real context of the stuff you’re looking into. 

Google reverse image search, also known as Google Search by Image, is a Google tool that allows users to search for images by starting with an image rather than a written or spoken search phrase.

Simply submit an image or provide a link to an image on the web, and Google will try to find similar images.

Our final thoughts.  Deepfake is a relatively new and exciting technology. Humanity is still getting to know it, and its full potential in our society has yet to be realized. Like many other technologies, it has benefits and drawbacks. It has the potential to hurt or benefit our world. 

We’ll need some time to figure out how to make the most of it in various businesses. In the meantime, let’s protect our privacy online. Our photos, videos, and sensitive information belong to us. Let’s keep it that way.

#3 You really should sign up. "Scouts Promise" ... it really will help you stay cyber safe.

By entering your email address you agree to receive emails from EveryDayCyber. We'll respect your privacy and you can unsubscribe at any time.