New layer of realism in the fake videos generated by AI, which now feed the detection of real human heartbeat, according to a new study. “Now, just because a person has a measurable pulse in a video, that doesn’t mean we can assume they are real.”
With each passing day, artificial intelligence (AI) becomes more realistic, and Deepfake technology has just reached a new, scary milestone to avoid detection tools, further destabilized the line between good and evil.
According to a study this Wednesday in Frontiers in Imagingthe deepfake technology generated by AI can now imitate real human heartbeat.
To confirm, by “fighting” detection methods based on physiological signals, the technique brings a whole New layer of realism To videos that we often no longer be able to distinguish from reality today.
Currently, the detection tools that prove more effective for detecting these fake videos and images depended on the identification of subtle skin color changes caused by blood flow. But the new study by the Fraunhofer Telecommunications Institute in Germany ensures that some deepfakes can now simulate these signs precise enough to detect detection systems.
The investigators, led by Peter Eizeddeveloped a detection tool designed to analyze heart rate signals in videos, based on medical imaging technology called remote photolaphing, highly accurate, according to the.
To test their effectiveness, they recorded real videos of participants at the same time monitored their heart frequencies. Then they digitally replaced participants’ faces with altered versions. And the worst scenario was confirmed: the detection tool identified realistic pulsation signs in the Deepfakes, similar to those of the original filming.
“Now, just because a person has a measurable pulse in a video, that doesn’t mean we can assume they are real,” warns Hany Farid of the University of California in Berkeley, which has not participated in the study. But it also points out that “just because one or some deepfake generators can reproduce this physiological sign, that doesn’t mean all Deepfake generators can do it.”
Faced with this new obstacle, researchers have already begun to explore new but similar methods of deepfakes, such as the detection of others unique patterns of blood flow on the face or from Microscopic Image Detailshow Pixel glow variationsmost often imperceptible to the human eye.
“As the heart beats, the blood flows through the blood vessels and reaches the face,” Eisert told the BBC. “Then it is distributed throughout the facial area, and there is a small delay in this movement that we can detect in real images.”
For now, good and evil continue in the race to step ahead of the other.
Tomás Guimarães, Zap //
