Advanced systems of artificial intelligence can interact in order to create original, ultra-realistic images and sounds something that machines have not been able to do yet.
This provides machine with something similar to imagination, which even if can lead them to be less dependent to humans, makes machines alarming powerful machines for counterfeiting.
This ability is based on the GAN theory. Two neural networks are trained in the same database. One, known as the generator, is responsible for creating modifications in images already seen – for instance, a man with three arms. The second one, known as the discriminator, has to identify whether it is similar to the images it has learn during its training or a fake image made by the generator.
The generator learns about the structure of what it sees and listens so well that the discriminator may not perceive counterfeits.
GAN have been used to generate realistic sounds and fake images, but their results are not always perfect. Sometimes, they can see objects where there are none and faces in wrong places.
By now, they do not have so much imagination, but the GAN are able to create images and re-imagine them. It seems that this neuronal network battle is not close to be finished.