The deep fake technology first emerged in the mid-2010s and soon gained popularity on Reddit by 2017. The technology is used to create synthetic media with AI, and it has grown exponentially since then. While deep fakes have several legitimate use cases, like every other piece of technology, they also have drawbacks.
The technology soon became exploited by cybercriminals, and today, despite being an important part of various industries, it poses serious privacy and security risks. In this article, you will learn in depth what deep fake is, how it works, and the impact it has on technology.
What is Deep Fake?
The term deep fake comes from “deep learning” and “fake,” referencing the neural networks that power the technology. In simple terms, a deep fake is a form of synthetic media, such as an audio, video, or image, that has been digitally manipulated using AI to mimic real people, actions, or speech.
Unlike traditional editing, deepfake media is generated by feeding AI models with real data, such as voice recordings, photos, and video frames. This helps create hyper-realistic content that is almost indistinguishable from reality.
How are deep fakes created?
The term deep fake first made an appearance in the public domain in 2017, after a Reddit user with the username “deep fakes” posted pornographic videos on the site. This user created those videos using Google’s open-source, deep-learning technology to swap celebrity faces onto the bodies of pornographic actors. The modern version of deepfake is a direct descendant of the original code used to create those videos. Now there are two main methods to create deep fakes, which are the GAN and the Autoencoder:
GAN
The most popular way to create deepfakes is through generative adversarial networks (GANs). This method trains itself to recognize patterns using algorithms that can be used to create synthetic media. A GAN consists of two networks:
- A generator that uses an autoencoder to create images and fake videos that look real
- A discriminator that evaluates whether the videos or images created are fake or real.
These two networks train together in a competitive loop. The generator continues to improve its fakes while the discriminator gets better at spotting them. Eventually, the generator becomes so good that the discriminator can no longer spot the fake.
Autoencoder
An autoencoder is a type of neural network designed to learn an efficient representation of data. It consists of two parts:
- An encoder that encodes input data into a similar representation
- A decoder that reconstructs the original input from its compressed form.
In deepfake, the autoencoders are trained on two individuals: one is the source face, and the other is the target. The encoder learns to extract facial features, and the decoder learns how to reconstruct each individual’s face. Together, the encoder for the source and the decoder for the target are used to generate realistic images of the target person.
The impact of deep fakes on security
The deep fake technology is indeed impressive. However, quite like how it originated, the technology is still widely used to conduct sinister acts and poses a deep harm to online privacy and security. Here are some of the ways deep fakes are used to harm society:
Political interference
Political manipulation through deep fakes is one of the most significant threats of this technology. Often, opposing party members create advanced AI-generated content to create highly convincing videos of political figures, saying or doing things they have never said or done.
Moreover, at times, party members also boost their election campaigns by creating fake images or videos of celebrity figures or influencers cheering for a certain party to boost its popularity. All these acts are especially common during the election cycles and are done to change public opinion. The problem here is that it could erode people’s trust in politics, creating a liars’ dividend where even original videos are dismissed as fakes.
Scams and hoaxes
Cyber criminals can use deep fake technology to create scams. False claims and hoaxes to destabilise organisations or cause public unrest. For example, an attacker could create a false video so a senior executive admits to committing a crime or exposing false “truths” about the organization’s business activities. Such acts could not only cost money but also dent an organisation’s reputation and share price.
Cybercriminals could also cause public unrest by creating deepfake videos of politicians, influencers, or public figures supporting controversial positions, thereby inciting public outcry. Such videos often lead to riots and public unrest.
Identity theft and financial fraud
Deep fake technology can be used to create new identities and steal the identities of real people. Cybercriminals can use the technology to replicate a voice, allowing them to buy things and create multiple accounts in that person’s name.
It is very easy to use deepfake technology to spread automated disinformation, such as conspiracy theories and incorrect ideologies about political and social issues. This could cause public outrage and unrest. This could also be used to defame an important person, such as a politician or a celebrity.
Social Engineering
The most common use of deepfake technology is in social engineering scams. Attackers use the technology to replicate fake audio and videos to conduct illicit activities. For example, they could create a fake audio of the CEO urgently demanding a large sum of money from the accounts manager in an offshore account. If the accounts manager falls for the scam, it could lead to significant financial losses for the company.
Legal use cases of deep fake technology
Although deepfake technology is mostly known for its illicit uses, it also has several legal uses. The entertainment industry has a significant use of the technology, with directors and producers using it to shoot movies and TV shows. It can be used to create realistic stunts without harming anyone.
Apart from that, professional and amateur historians use deepfake technology to recreate old photos and paintings. The technology can also be used to demonstrate technical events and demonstrations for news reports.
Is it possible to spot deep fakes?
While deepfakes are used to mimic reality and the technology has advanced rapidly, it is still possible to spot deepfake-generated media. All you need to do is look out for unnatural movement or activity, such as:
- A synthetic video created using this technology will have unnatural eye movement, as people’s eyes are hard to replicate.
- The deep fake technology has yet to master the natural human action of regular eye blinking. Synthetic videos contain people blinking less than an average human does.
- The technology mostly focuses on replicating a person’s face, so the videos often have people with unnatural body shapes.
- Deepfakes still struggle to replicate natural skin color, and the resulting videos or pictures often have slightly unnatural skin tones.
- A synthetically created video will likely feature lip-syncing that does not align with the words being spoken by the people in the video.
- These videos often have people with awkward-looking head and body positioning, such as jerky movements.
If you look closely, there is a good chance you’ll recognize these issues and figure out whether a video is synthetically generated (e.g., a deepfake).
How to stay safe?
Deepfake technology is growing rapidly, and everyone must take the necessary steps to stay safe. The governments and big tech companies have already put in place various precautions that are being followed, such as the following:
Deepfake videos widely circulate on social media platforms, which is why Facebook hired researchers from universities to help build a deepfake detector. This detector enforces bans on deep fakes. Similarly, Twitter has policies in place to prevent the spread of fake content, and it tags deepfake images so they are removed immediately. YouTube also blocks deepfake content, especially when it relates to elections.
Research lab technologies.
Researchers have been developing data science solutions to detect and filter deep fake content. However, so far they have not been able to come up with anything rock-solid as attackers’ technology continues to evolve.
Filtering programs
Various filtering programs are working to prevent deepfakes. The AI firm DeepTrace has invented a program that mimics an antivirus or a spam filter. It quarantines fake content. Similarly, other programs tag manipulated content before it causes any damage.
Best Practices to safe and secure.
While these best practices are in place, the best way you can stay safe from synthetically generated media is by being informed and adopting cyber hygiene. Here are some of the practices you must do to stay safe:
- Avoid oversharing personal photos and sensitive details on social media and tighten privacy settings, limiting who can see the content you share.
- Strengthen account security by using appropriate measures such as two-factor authentication and unique passwords.
- Maintain digital literacy. Question emotionally charged or sensational content before reacting to it harshly or sharing it with others.
- Pause and inspect before clicking any links on the internet to verify legitimacy.
- Independently verify requests for money or sensitive information through a trusted channel instead of blindly trusting a familiar voice.
By combining these digital safeguards and remaining sceptical, it is possible to stay safe from deep fake fraud.
Key Takeaways
The deep fake technology has rapidly been on the rise, specifically with the advancements of AI. While the technology does have several use cases, it is a severe danger to online privacy and security. Not only does it create mistrust, but it can also cause significant harm to people, organizations, or even governments. Therefore, it is crucial to remain educated about deep fakes and enforce best practices to prevent them from causing significant harm.
FAQs
What are the current deep fake laws?
Deepfake laws vary by country, but many jurisdictions now criminalize non-consensual deepfake pornography, election interference, fraud, and identity misuse, with stricter AI transparency and labeling requirements.
How can you protect yourself from deepfake threats?
You can protect yourself from deepfake threats by limiting public sharing of personal media, enabling strong privacy settings, using reverse image searches, and relying on verified sources to confirm suspicious audio or video content.
What are the legal consequences of deep fakes?
Legal consequences of deepfakes include fines, civil liability, reputational damages, and imprisonment for offenses such as defamation, fraud, harassment, identity theft, and distribution of non-consensual explicit content.
What are the ethical and security concerns associated with deep fakes?
Deepfakes raise ethical and security concerns, including misinformation, political manipulation, reputational harm, cyber fraud, erosion of digital trust, and threats to personal privacy and biometric security.
