Date: 04/05/2024

What are Deepfakes?

Deepfakes are instances where powerful techniques from machine learning (ML) and artificial intelligence (AI) are leveraged to manipulate or generate visual and audio content with a high potential to deceive.

These artificial images or videos are generated through a process called deep learning. This type of learning is a subset of machine learning that utilizes deep neural networks to display complex decision-making power in an attempt to mimic the human brain. Through this method, it “feds” examples from existing parts of the internet, depending on what the AI is being trained on, and produces a compilation of what it has learned.

How are they made?

Deepfakes utilize two algorithms: the generator and the discriminator. These algorithms create content as the generator builds a data set based on the user’s desired output, letting the AI create the fake content. The discriminator analyzes the content produced and decides how realistic it looks. As this process is repeated, the algorithms keep perfecting the craft until it is no longer corrected and deemed realistic. This creates a Generative Adversarial Network(GAN). This network uses deep learning to find patterns in images, which are then used to make fakes. For videos, GAN takes the initial video and studies the behavior, movement, and speech patterns, then applies that to its system to create a realistic version of its video.

Image From Blockchain can help combat the threat of deepfakes. Here’s how

Image from Deepfakes: Trick or treat?

Effects of Deepfakes

With the rise in AI advancement, there has been increasing difficulty in being able to differentiate between real and fake images or videos which makes deepfakes potentially dangerous.

In 2018, researchers in the United States demonstrated that deepfake faces didn’t blink like humans do. However, as soon as the study was published, deepfake creators started fixing this, making it even more difficult to detect deepfakes (Britt). It is still possible to spot subtle changes in a very well-made, deep fake video or image. It is not impossible to spot all the signs that could show if a video is fake or not. The software used, which is easily bought online, can produce these videos without any trouble and make dozens of them. These videos can have any AI-generated figures with any accent, gender, age, and clothing.

Types of Deepfakes

Multiple types of deepfakes are found online. Some of these types include blackmail, entertainment, fraud, and misinformation:

Blackmail: Blackmail can be targeted at a certain person or group of people who look like they are performing illegal and inappropriate things to ruin their reputation and make them look bad in public, mainly to cyberbully or as revenge.
Entertainment – Using deepfakes for entertainment has been in the industry for a while now, especially when there are hard scenes to create or an actor has grown too old to play their role. A lot of deepfakes are also used for satire and parody when it comes to entertainment. A great video demonstrating this can be found in this video taken in 2023 of Dwayne Johnson as Dora the Explorer. The link can be found here.
Fraud – Fraud is a big problem when it comes to deepfakes, as it can be used to impersonate a reliable person or a company to gather personally identifiable information or PII. PII can range from bank account, credit card, and social security number information. This is also seen as a major cybersecurity threat.

Since deepfakes can be anyone, many videos and photos that are deepfakes are used to spread misinformation. Many politicians or trusted sources are misrepresented using deepfake techniques to change public opinion, create confusion, and get negative responses. The most dangerous effect is for people to act based on misinformation, which could endanger lives.

Deepfake Apps

Finding deepfake apps online is straightforward. Each app takes a different approach when it comes to making a fake video or image. Some apps only allow photos to be created, while others only support videos. Each app that offers to make deepfakes requires a subscription, while also offering different methods for creating and distributing these requested fakes. There are many apps to choose from but only two will be referenced below:

Deepfakes Web: This service allows the creation of deepfake videos and uses deep learning to understand complex human expressions or facial data. Unlike some of the other apps, Deepfakes Web will take a few hours for the AI to learn and train through various images and videos. This usually results in a better outcome than a quick 30-minute swap with other apps. Deepfakes Web also has a premium version, which will split the time it takes to make a video, but it costs more than four dollars an hour.

Deepbrain: This is another piece of software that allows users to create deepfake videos, but they can only be found online. This website stands out from the rest because it takes a whole different approach. The user can use the already-created AI characters to make a PowerPoint or news presentation on whatever topic they want. The website creates these deepfake videos through text, as the user gives a prompt through text, and the software then creates a video through the sent message. PPTs, PDFs, URLs, and articles can also be turned into videos with a generated script. Since the character models are already trained, it takes a much shorter time to produce a video. ChatGPT is also included in this website and can be used to create text for the AI avatar to say. This method of creating deepfakes is popular, but the regular subscription costs twenty-four dollars a month, and a premium subscription is 180 dollars per month. The website also does not mention or use watermarks or any methods to stop the spread of malicious content.

Deepfakes Laws and Regulations

There have been a multitude of deepfake videos that have surfed the internet, and it feels almost impossible for the really bad videos to get taken down or have any sort of action taken against them. However, already existing laws can challenge these deepfake videos and images and battle against them, while new laws can be created to specifically target this type of concern.

Privacy laws vary by state, but California’s Right of Publicity Law “protects individuals from unauthorized use of their name, voice, signature, photograph, or likeness” (Dickinson). Although these laws were put in place for commercial purposes rather than deepfake, they can still be used against malicious videos. Creating deepfakes with a user’s likeness without permission or acknowledgment infringes on their rights to control their image and likeness. The only way to avoid that law is to make sure the individual gives express consent, which is when the person is clear and direct when agreeing to an app, software, or website to use their likeness.

Cybercrime and harassment laws also cover many parts of deepfakes through their creation and disruption. Usually, cybercrime applies when these images or videos are created through hacking or unauthorized access to data. Cases can also be formed if deepfakes are used to harass or stalk an individual or group of people, along with deepfakes creating explicit content without the permission of the person whose likeness is being used.

In the beginning of 2024, bipartisan lawmakers introduced a bill that will allow victims of inappropriate and damaging deepfakes to sue the people who created them. The DEFIANCE Act, or Disrupt Explicit Forged Images and Non-Consensual Edits, allows individuals to sue the deepfake creators if they knew or “recklessly disregarded” the victim’s consent. This law came to light a week after a celebrity named Taylor Swift had her likeness used in many inappropriate deekfake images without her consent, which were then sent to X, formally known as Twitter. As stated above, some laws can be used when it comes to victims being harassed or bullied online by deepfakes, but this will be the first federal law that will specifically protect victims of deepfakes. Luckily, many Americans support this federal bill, as “…84% say they are in favor of legislation” (Burga). Unfortunately, this law is still being introduced, but there is a lot of support and hope that this bill will pass through Congress and be registered as a federal law.

Conclusion

Deepfakes are artificial images or videos generated through a special process called deep learning. Two algorithms are used to create deepfakes: the generator and the discriminator. These algorithms are used to create and refine content as the generator builds a data set based on the user’s desired output, letting the AI create the fake content. Due to technological advancements, distinguishing between real and fake images or videos is one of the reasons why deepfakes are becoming more dangerous. They are also running rampant on the internet due to their popularity and easy accessibility from multiple apps trying to gain easy money. Users create deepfakes for many uses, such as blackmail, entertainment, fraud, and misinformation. Many deepfakes that use the likeness of others without consent to harm and bully them can be faced with laws. These established laws are mainly for cybercrimes or other forms of harassment, but they can still work for cases involving deepfakes. There is also a recent introduction to a potential federal law that allows victims to sue the creators of deepfakes, including inappropriate actions, without their consent. This introduced law already has a lot of support, and there is potential that it will be passed in the future.

Works Cited

Anonymous. “Blockchain Can Help Combat Threat of Deepfakes. Here’s How.” World Economic Forum, 12 Oct. 2021, www.weforum.org/agenda/2021/10/how-blockchain-can-help-combat-threat-of-deepfakes/.

Anonymous. “What the Heck Is a Deepfake?” What the Heck Is a Deepfake? | UVA Information Security, 2024, security.virginia.edu/deepfakes.

Barney, Nick, and Ivy Wigmore. “What Is Deepfake Ai? A Definition from TechTarget.” WhatIs, TechTarget, 21 Mar. 2023, www.techtarget.com/whatis/definition/deepfake.

BBC News. “Fake Obama Created Using AI Video Tool – BBC News.” YouTube, YouTube, 19 July 2017, www.youtube.com/watch?v=AmUC4m6w1wo.

Britt, Kaeli. “How Are Deepfakes Dangerous?” University of Nevada, Reno, University of Nevada, Reno, 31 Mar. 2023, www.unr.edu/nevada-today/news/2023/atp-deepfakes.

Burga, Solcyré “Deepfakes Are at the Center of a New Federal Bill.” Time, Time, 31 Jan. 2024, time.com/6590711/deepfake-protection-federal-bill/.

Jodka, Sara H. Manipulating Reality: The Intersection of Deepfakes and the Law …, Manipulating reality: the intersection of deepfakes and the law, 1 Feb. 2024, www.reuters.com/legal/legalindustry/manipulating-reality-intersection-deepfakes-law-2024-02-01/.

Kietzmann, Jan, et al. “Deepfakes: Trick or Treat?” Business Horizons, Elsevier, 24 Dec. 2019, www.sciencedirect.com/science/article/pii/S0007681319301600.

Mulvihill, Geoff. “What to Know about How Lawmakers Are Addressing Deepfakes like the Ones That Victimized Taylor Swift.” AP News, AP News, 31 Jan. 2024, apnews.com/article/deepfake-images-taylor-swift-state-legislation-bffbc274dd178ab054426ee7d691df7e.

Satariano, Adam, and Paul Mozur. “The People Onscreen Are Fake. the Disinformation Is Real.” The New York Times, The New York Times, 7 Feb. 2023, www.nytimes.com/2023/02/07/technology/artificial-intelligence-training-deepfake.html.

Sha, Arjun. “12 Best Deepfake Apps and Websites You Can Try for Fun.” Beebom, 6 Sept. 2019, beebom.com/best-deepfake-apps-websites/.

TheFakening. “Dwayne ‘the Rock’ Johnson as Dora the Explorer Deepfake.” YouTube, YouTube, 19 Nov. 2019, www.youtube.com/watch?v=OVV8jyh8woQ. “What Is Deep Learning?” IBM, 2024, www.ibm.com/topics/deep-learning.

Deep Fakes