Deepfake

What is a Deepfake?

A deepfake refers to computer-generated videos, audio recordings, and images that are used to portray individuals saying or doing things they never actually did or said. Essentially, deepfake technology uses artificial intelligence (AI) and machine learning (ML) to generate synthetic digital content that looks and sounds as if it is authentic.

While deepfakes can be created for benign purposes, such as in filmmaking or satire, they have gotten a negative connotation because the most publicized applications of deepfake technology have been fraud-related. They have gained attention for their ability to mislead viewers into believing that falsely depicted events or statements are real.

The technology’s ability to be misused has raised ethical, legal, and social issues, particularly regarding misinformation, privacy violations, and the manipulation of public opinion.

Techopedia Explains the Deepfake Meaning

The deepfake definition is a portmanteau of “deep learning” and “fake.” In this context, deep learning is a type of machine learning, and fake is a synonym for fabricated or synthetic data.

It’s important to remember that while all deepfakes use synthetic data, not all synthetic digital content qualifies as deepfakes. The key difference lies in the intention behind the creation of the content and its potential for deception.

How are Deepfakes Created?

Deepfake models can be created with generative adversarial networks (GANs), autoencoders, or variational autoencoders. Once the model is sufficiently trained, it can be used to create deepfakes by inputting new training data or prompts.

What is Deepfake-As-A-Service?

Ten years ago, if someone wanted to create convincing deepfake content, they needed to have a strong background in mathematics, data science, and computer programming.

Today, people can use free or low-cost apps and cloud services that can create convincing deepfakes from just a few reference images or videos. For example, Tencent has a commercial deepfake service that can create high-definition, realistic deepfake humans using just three minutes of live-action video and 100 spoken sentences as source material.

Unfortunately, this has made it easier than ever for threat actors to create deepfakes too. In the last five years, there have been notable instances where deepfakes have been used to spread misinformation, commit financial fraud, create non-consensual adult content, and unduly influence political campaigns.

The quality of deepfakes created with inexpensive software is inconsistent, however. That is why deepfakes with noticeable flaws or inconsistencies are often referred to as cheapfakes or shallow fakes (the opposite of deepfakes).

Use Cases of Deepfakes

Deepfakes have been used in various contexts, ranging from benign and entertaining to controversial and malicious. Here are some notable examples (both good and bad).

Who is Making Deepfakes?

Some of the initial work on deepfake technology was conducted in academic and research settings to explore the potential for using AI in film. Practical applications of early deepfake technology included matching an actor’s lip movements to audio recorded in another language, de-aging actors, or replacing one actor’s face with another’s in a specific scene without having to film the scene again.

Today, a significant number of deepfakes are created by hobbyists and technology enthusiasts. Deepfake technology accessibility has increased with the widespread availability of user-friendly deepfake software and cloud services, and people with varying levels of technical skill can create realistic deepfakes.

How to Spot a Deepfake

Although it’s getting harder to identify deepfake images, audio, and video as the technology improves, there are still some telltale signs and techniques you can use to help identify deepfakes.

Deepfake Detection Tools

Today, there are a number of technical tools and services that people can use to detect inconsistencies and artifacts introduced during the deepfake creation process.

• Sentinel: According to their website, Sentinel works with governments, media, and defense agencies to help protect democracies from disinformation campaigns, synthetic media, and information operations.
• Deepfake Detector: The company’s AI Voice Detector can help users detect if an audio or video clip is a deepfake.
• Sensity: Sensity’s proprietary application programming interface (API) can accurately identify AI-altered visuals 98.8% of the time.
• Intel FakeCatcher: According to the Intel website, FakeCatcher analyzes blood flow in video pixels to determine a video’s authenticity.
• Resemble AI: Resemble AI services include a cutting-edge AI Voice Generator and robust deepfake audio detection.

It’s important to remember that while deepfake detector tools and services are continually improving, the technology behind deepfakes is also advancing. This is why deepfake detection is often categorized as being a cat-and-mouse game.

The Impact of Deepfakes on Society

While deepfakes can serve as powerful tools for entertainment, education, and social commentary, the technology’s potential for misuse in phishing scams, identity theft, and financial fraud has made it a significant security concern.

Deepfake technology has the power to erode trust in media, facilitate misinformation campaigns, fuel political polarization, and pose a serious threat to individuals’ reputations and emotional well-being.

In society, the creation and dissemination of deepfakes are raising questions about consent and privacy, as well as questions about the potential for technology to impact humanity in a harmful manner.

Examples of Deepfake Misuse

Here are some real-world examples of deepfake technology misuse:

• A finance worker in Hong Kong transferred more than $25 million to scammers after being deepfaked by a video conference call.
• In the United States, thousands of people received a deepfake robocall from President Joe Biden telling them not to vote.
• An energy firm’s CEO was tricked into transferring €220,000 after he was deepfaked into believing he was on the phone with his boss.
• A deepfake video of Japan’s Prime Minister Kishida Fumio made it seem as if the Prime Minister was uttering vulgar remarks.
• A video of Microstrategy executive chairman Michael Saylor urged viewers to send him Bitcoin with the promise that he could double their crypto investment.

Ethical Implications of Deepfakes

As deepfake technology becomes more sophisticated and accessible, it is raising questions about how to verify the authenticity of source content.

Other questions concerning the ethical use of deepfakes include:

• Do we need new laws or regulations to regulate the use of deepfake technology?
• Can we use AI watermarks to reliably identify deepfake content?
• Who should be held responsible if synthetic content damages someone’s reputation?
• Do existing laws and regulations address the challenges in courts of law where video and audio evidence was once considered reliable?
• Can blockchain technology be used to authenticate legitimate content and establish the provenance of a digitized asset, including text, audio, and video?

Legal Implications of Deepfake Technology

Deepfake technology is creating challenges in courts of law where video and audio evidence were once considered reliable.

In many countries, existing laws and regulations are not addressing the nuances of deepfake technology adequately, and this has led to calls for new regulations and legal frameworks.

Several countries and jurisdictions have begun to introduce laws and regulations specifically designed to hold deepfake creators and distributors accountable for harmful impacts.

• In the United States, at least ten states have passed laws that criminalize the creation and distribution of deepfake pornography without consent and deepfake videos that aim to interfere with elections.
• China has strict laws that specifically prohibit the production of deepfakes without user consent and require content generated with artificial intelligence to be clearly labeled.
• The EU’s Artificial Intelligence Act includes provisions that require anyone who creates or disseminates a deepfake to disclose the content’s artificial origin and provide information about how the content was created.

The Bottom Line

Deepfake technology itself is not dangerous. It can be used to engage learners, lower production costs in film, and streamline content adaptation for audiences who speak different languages.

The ease with which threat actors are using the technology to create convincing fake videos and video clips, however, is helping to undermine people’s trust in digital media.

FAQs