The Puppet Masters of the Digital Age: Deepfakes and the Fight for Truth
What Are Deepfakes
Imagine a world where a celebrities’ words can be twisted with a click, sparking outrage and fracturing trust. Deepfakes, hyper-realistic forgeries, are the ultimate misinformation weapon, threatening to tear apart communities and destabilise nations.
See for yourself:
Deepfakes use cutting-edge artificial intelligence (AI) concepts such as computer vision, to create videos, images or even audios so convincing, they would fool even the most discerning eye. Think of it as a next-level illusionist, weaving a web of convincing deception. The science behind deepfakes is impressive, but the potential for harm is terrifying. Anyone can become a puppeteer, pulling the strings on an individual’s virtual image, making them say and do anything. And as this tech gets more sophisticated, spotting the fakes can become a nightmare.
Who Can Create Deepfakes
Deepfakes are becoming alarmingly easy to create. Once the exclusive domain of programmers, advancements in technology have opened the door for anyone to use them. User-friendly tools like Reface are just the tip of the iceberg, simplifying complex AI processes for the average user. Others include: Wombo, Faceswap and Deepfakes Web. This surge in accessible deepfake creation tools raises concerns about the potential misuse of this powerful technology.
How Are Deepfakes Detected
Remember that video of a deepfaked Barack Obama? Well now watch this:
Play close attention to the eyes and mouth in the two videos… notice anything? The way the mouths move does not quite seem to match the words being spoken, and the eyes…well, they do not look quite natural, do they? These inconsistencies can be signs of deepfakes.
While some inconsistencies are clear, others are becoming increasingly sophisticated. Here is where AI computer vision comes in. Just as AI computer vision is used to create deepfakes, it can also be harnessed to detect them. This powerful technology can analyse subtle details beyond human perception, acting as a stronger and smarter defence against manipulated videos.
AI Computer Vision To Unmask Deepfakes
Blurring:
Deepfake creation hinges on flawlessly merging the source face onto the target face. This process, however, is not foolproof. Misalignments, especially around the edges like the hairline or jawline, can introduce subtle blurring. Here, the textures and details from both faces are not seamlessly combined, creating a faint blur. In other instances, there might be artefacts like pixelation or colour inconsistencies where the software couldn’t perfectly blend the images. AI computer vision excels at spotting these discrepancies. It analyses the image pixel-by-pixel, comparing details with hawk-like precision. Even the slightest mismatch around the edges can be a red flag for AI computer vision, signalling a potential deepfake.
Lighting Inconsistencies:
Flawless lighting harmony is a hurdle for deepfakes. Merging a brightly lit face onto someone in the shadows creates a lighting paradox. The source face’s highlights on the forehead and cheeks would not align with the target face’s shadows, leading to inconsistencies. AI computer vision acts like a light meter, meticulously inspecting the image. It compares lighting on the face with the background, searching for giveaways. These lighting mismatches expose deepfakes that try to pass off manipulated reality as genuine.
Abnormal Movements:
Manipulating expressions in deepfakes is a minefield for AI computer vision. Superimposing new smiles or frowns can lead to telltale unnaturalness around the mouth and eyes. AI computer vision analyses these areas with hawk-like precision, detecting mismatched lip movements or irregular blinking patterns. But it does not stop there. Computer vision goes beyond the face, scrutinising body language throughout the video. Jerky head movements or hand gestures that clash with facial expressions expose inconsistencies. By analyzing every frame for these deviations from natural human movement, AI computer vision helps us sift truth from fabrication in deepfakes.
Nevertheless, by training on massive datasets of real and deepfake videos and images, AI computer vision models can become highly accurate deepfake detectors. They can also analyse large amounts of visual data rapidly, making them ideal for real-time applications. Perhaps the most crucial aspect of AI computer vision is its ability to adapt and learn. As deepfakes become more sophisticated, AI computer vision can evolve to identify even the most realistic forgeries.
Can Deepfakes Be Legally Combatted?
Deepfakes pose a complex challenge for the law. Striking a balance between protecting people from their harmful effects and upholding freedom of expression is no easy feat. Laws need to be able to differentiate between a harmless parody video and a malicious deepfake designed to damage someone’s reputation or sway an election. This requires clear definitions of what constitutes a deceptive deepfake.
Further complicating matters is the global reach of deepfakes. They can be created and disseminated across borders, making it difficult for any one country’s laws to effectively address the issue.
Despite these challenges, there are some legal tools currently available. Existing slander and libel laws can be applied in cases where deepfakes spread demonstrably false information that harms someone’s reputation.
Withal, the future of deepfakes and the law likely involves the establishment of legal precedents through court cases that set standards for how these manipulated media creations are handled. International cooperation between governments and tech companies will also be crucial for effectively addressing this global issue. The ultimate goal is to find a way to protect people from the potential harms of deepfakes without stifling the important right to free speech.
