In today’s digital age, technology is constantly pushing the boundaries of what’s possible, and one of the most intriguing yet concerning developments is the rise of deepfakes. Have you ever come across videos where famous figures appear to say or do things that seem entirely out of character? If so, you’ve witnessed a deepfake. Deepfakes represent the 21st century’s answer to image manipulation, employing a specialized form of artificial intelligence known as deep learning to create convincing but entirely fabricated content. In this comprehensive guide, we’ll delve deep into the world of deepfakes, exploring their purpose, creation process, potential impact, and how to spot them.
What is a Deepfake?
Deepfakes are a product of deep learning, a subset of artificial intelligence. They utilize advanced algorithms to manipulate images and videos, swapping faces and voices to create deceptive content. The term “deepfake” itself derives from the combination of “deep learning” and “fake.”
What Are They Used For?
1. Pornographic Content: Unfortunately, a substantial portion of deepfakes involves explicit content. In fact, a shocking 96% of deepfake videos discovered by the AI firm Deeptrace in September 2019 were pornographic in nature. Even more concerning, 99% of these videos featured faces of female celebrities superimposed onto adult film actors. With the increasing accessibility of deepfake creation tools, the risk of deepfake-related revenge porn and exploitation is on the rise.
2. Spoof and Satire: Beyond the realm of explicit content, deepfakes have been employed for purposes like spoofing, satire, and mischief. They can be used to create content that appears as if it’s coming from public figures, leading to potential confusion and controversy.
Is it Just About Videos?
No, deepfake technology extends beyond videos. It can create entirely fictional photos from scratch. For instance, the existence of a non-existent journalist, “Maisy Kinsley,” with profiles on LinkedIn and Twitter, was likely the result of a deepfake. Additionally, deepfake technology can manipulate audio, leading to the creation of “voice skins” or “voice clones” of public figures. This audio manipulation can have significant consequences, as seen in a case where a UK subsidiary executive transferred a substantial sum of money to a Hungarian bank account after receiving a phone call from a fraudster who mimicked the German CEO’s voice. While deepfakes often focus on videos, they also have the capacity to affect written and audio content.
How Are Deepfakes Made?
Deepfake creation involves a series of intricate steps:
1. Data Collection: Thousands of images of the target individuals are collected.
2. Encoding: An AI algorithm, known as an encoder, identifies similarities between the faces and compresses the images, focusing on shared features.
3. Decoding: A second AI algorithm, the decoder, is trained to reconstruct faces from the compressed images. Separate decoders are trained for each target face.
4. Face Swapping: To create a convincing deepfake video, encoded images are input into the “wrong” decoder, causing the decoder to generate the face of one person with the expressions and orientation of another. This process is repeated for each frame in the video.
Another method uses generative adversarial networks (GANs), where two AI algorithms, a generator and a discriminator, compete to create and detect realistic images. With continuous feedback and training, GANs produce highly convincing deepfakes of entirely fictional individuals.
Who is Making Deepfakes?
Deepfake creation is not limited to a specific group; it involves a wide range of individuals and organizations:
1. Academic and Industrial Researchers: Experts in AI and image manipulation research deepfakes for various purposes.
2. Visual Effects Studios: Professionals in the film industry and visual effects studios are constantly pushing the boundaries of what’s possible with deepfake technology.
3. Enthusiasts: Amateur enthusiasts with a passion for AI and image manipulation create deepfakes as a hobby.
4. Governments: Governments have been known to dabble in deepfake technology as part of their online strategies to discredit and disrupt extremist groups or engage with targeted individuals.
What Technology Do You Need?
Creating high-quality deepfakes typically requires robust computing resources, often including high-end desktops with powerful graphics cards or cloud-based computing power. These resources significantly reduce processing time, from days or weeks to mere hours. While advanced tools and expertise are necessary, several companies and even mobile apps are available to assist in deepfake creation. For instance, the mobile app Zao allows users to add their faces to a list of TV and movie characters for which the system has been trained.
How Do You Spot a Deepfake?
As deepfake technology evolves, spotting them becomes increasingly challenging. However, there are still some telltale signs to look out for:
1. Blinking: Deepfake faces often do not blink naturally, as the majority of training images feature open-eyed subjects. Some early detection methods revolved around this, but deepfakes have since evolved to include blinking.
2. Poor Quality: Lower-quality deepfakes may exhibit issues like unsynchronized lip movements, patchy skin tones, flickering around transposed faces, and difficulties rendering fine details such as hair, jewelry, and teeth.
3. Lighting and Reflections: Inconsistencies in illumination and reflections on the iris can also serve as indicators of a deepfake.
Efforts to detect deepfakes are ongoing, with research initiatives supported by major tech companies like Microsoft, Facebook, and Amazon. While progress is being made, the continuous improvement of deepfake technology keeps the challenge alive.
Will Deepfakes Wreak Havoc?
The potential impact of deepfakes is far-reaching, and their misuse could lead to various consequences:
1. Misinformation: Deepfakes can harass, intimidate, demean, undermine, and destabilize individuals and institutions.
2. Stock Prices and Elections: Plausible deepfakes could influence stock prices, voters, and even international relations.
3. Undermining Trust: The insidious effect of deepfakes is the erosion of trust in digital content. A society unable to distinguish truth from falsehood becomes highly susceptible to manipulation.
4. Legal and Personal Security: Deepfakes pose risks in legal proceedings and personal security. They can mimic biometric data, potentially leading to scams and fraud.
What’s the Solution?
Artificial intelligence itself can be part of the solution. AI-based detection systems are being developed to flag deepfakes as soon as they emerge. Additionally, a blockchain-based online ledger system is being explored to maintain tamper-proof records of videos, images, and audio, allowing for the verification of their origins and authenticity.
Are Deepfakes Always Malicious?
Not all deepfakes are harmful. Some are purely for entertainment or helpful purposes. For example, voice-cloning deepfakes can restore lost voices due to diseases, and they can enliven galleries and museums. Technology also allows for improving the dubbing of foreign-language films and even resurrecting deceased actors for new roles.
What About Shallowfakes?
Shallowfakes, a related phenomenon, involve videos presented out of context or manipulated with simple editing tools. They may not be as sophisticated as deepfakes, but they can still have a significant impact, influencing public perception and causing controversy.
In conclusion, deepfakes represent a powerful and
