Lately, the world has been inundated with DeepFakes, photos and videos manipulated by AI technology. A recent app called DeepNudes, which used AI to “undress women,” created a controversy that resulted in its creator deleting it.
It is clear that DeepFake will become more common as time goes on, and if we do not take steps to control it, it could become a new threat. To address this possibility, researchers at the University of California have developed a deep neural network that can detect DeepFake.
This network, developed to detect patterns in raw data and modeled similarly to the human brain, was fed a set of images, including manipulated and unmanipulated photos, into the neural network.
The researchers knew which photos were doctored and which were not. To train the network, the researchers highlighted pixels along the edges of digitally added elements in the photo. DeepFake images are known to have lighter pixels in the artificially added parts.
While most of the time, it is not possible to detect altered photos with the naked eye, a computer that could examine photos pixel-by-pixel can use this capability to detect deepfake images.
After the neural network also looked at images outside the dataset previously fed to it, the network was able to detect deepfakes “most of the time.”
The results were satisfactory, but the neural network currently only works for photos. The researchers are trying to figure out a way to apply it to videos.
However, we cannot say that it is an answer to DeepFake, as the neural network is not 100% accurate.
Nevertheless, this development is a very good investment that we currently have to detect DeeFake.
