Superresolution in consumer cameras is definitely adding detail to the image using prior information about the universe (hair has a pattern, what looks like an edge has to be sharper than what the image says, etc). This is definitely the questionable artsy aspect of this superres boom. But the question I answered was more specific which supposed how you can make up info which doesn’t exist, but that’s definitely not true either. Modern superres tech (especially the non-deep learning kind) can extract more info from the image if you systematically account for the psf of the camera, distortions etc.

What makes you so confident that this "AI" doesn't end up just doing deconvolution?

Because the images being enhanced are in focus