Frighteningly enough, deepfakes have become extremely sophisticated.
Whether it’s doctored videos featuring political figures or unauthorized explicit content generated by artificial intelligence, these fabricated productions are advancing more rapidly than the tools designed to identify them.
Currently, researchers caution that the deceptive tactics have gained yet another advantage in this evolving cat-and-mouse dynamic.
A study published in
Frontiers in Imaging
According to research conducted at Humboldt University of Berlin, the most recent version of deepfakes can now simulate authentic human heartbeats, which was previously thought to be an infallible indicator of manipulated content.
Also read
Vietnam Bans Telegram
Medieval ‘Anatomically Correct Dagger’ Discovered in Sweden Lives Up to Its Name
Peter Eisert, the study’s author, states:
For the first time, we demonstrate that contemporary high-fidelity deepfakes can incorporate a convincing heart rate along with subtle alterations in facial pigmentation, making these videos significantly more challenging to identify.
Deepfakes are now psychically convincing.
In the past, numerous deepfake detection methods relied on a technology known as remote photoplethysmography (rPPG). This approach, initially utilized in telehealth applications, identifies minor fluctuations in facial skin color due to variations in blood circulation. It was believed that these indicators were too complex for generative artificial intelligence systems to replicate accurately.
However, in the recent study, deepfakes managed to deceive detectors designed explicitly to identify this particular indicator. The research team generated 32 AI-produced videos based on actual recordings of humans. These fabricated clips looked highly realistic and importantly triggered a false positive indicating a human heartbeat.
The researchers wrote:
Our tests showed that deepfakes can display authentic-looking heart rates, which contradicts earlier research.
According to
Popular Science
It seems the AI models acquired the heart-rate signals from the original source footage.
The deepfakes didn’t create fake pulses; instead, they reproduced the actual pulses that were inherent in the training data they used.
Fake videos, real consequences
The research reveals a significant weakness in our ability to identify tampered media. Although certain platforms are developing techniques that focus more on non-physical attributes, such as examining pixel-by-pixel anomalies, others continue to lean significantly on biometric indicators.
This research intensifies the push by major corporations such as Google and Adobe to integrate digital watermarks and verification information into media files, enabling users to readily distinguish between authentic and counterfeit content.
However, these very tools encounter difficulties amidst a swiftly changing arms race.
As Eisert puts it:
Minor differences in the actual person’s skin tone are preserved in the deepfake along with their facial movements, ensuring that the original heartbeat is recreated in the false video.
That’s a bone-chilling reminder: the next deepfake you encounter might not only appear realistic but could also seem genuinely authentic.
Also read
Research reveals the impact on your brain when you’re working too much.
Hundreds of Missing iPhones Located at Enigmatic Chinese Tower
Google is transforming entirely – 10 key points you should be aware of regarding the upcoming changes
