You are just anthropomorphizing the model by calling it teaching and then implicitly equating that it's the same thing happening in the human mind. That's your burden to establish when you say it's the same.
Actually, isn't the burden of the plaintiffs to prove their copyrights are being violated?
Whether or not it's identical to human brains isn't the matter, they'd need to prove how a small 5GB model trained from a huge dataset infringes their rights specifically.
I am not anthropomorphizing anything, because this is literally what happens. The model is taught, by having its predictions tested against examples, how images work.
