Can AI simulate fake inputs, train itself on them, and elaborate them like they were real ones generated from an outside source?
This idea is similar to how human brain works, indeed it can "think" whatever feeling, such a voice, a sound, an image, a smell, a flavour without really get them from outside. It can work on it like the real ones and give an output. It looks to me this is how the consciousness works.
Can AI simulate fake inputs, train itself on them, and elaborate them like they were real ones generated from an outside source?
Yes, this is exactly what generative adversarial networks do. Take a look at "A Beginner's Guide to Generative Adversarial Networks (GANs)" The GAN has a generator and discriminator. The generator "simulates fake inputs". The discriminator (police) tries to determine if it is fake or real. When the discriminator catches the fake ones, the generator learns from its mistakes. The generator keeps creating fake images until it starts fooling the discriminator into thinking they are real. In other words, the AI (i.e., the generator/discriminator combo) "trains itself" to create fakes that are so good that they start looking real
