It’s an image created by an algorithm that smashes pieces of existing art in its database to create a “close enough” image to what the prompt is. It isn’t really “learning” at all, and the “AI” part is just a marketing buzzword. I see it like those NFTs and things, it’s a cool fancy new technology thing, with no real practical applications right now, so people are being convinced of some hypothetical future benefit while the only people actually benefiting at present are grifters.