This is where I'm at as well, where it differs IMO is: the throughput/variety and the zero barrier to entry—even the most gifted 'shopper could only work so fast and with whatever available media. It would be possible to deconstruct the image to know face from X, background from Y. And still, virtually all photoshops have "tells" — they're "off" with regard to lighting, perspective, etc. and it varies from artist to artist.
Tools exist that can tell which pixels were "disturbed" or otherwise inconsistent with the rest of the image. Being able to spot a Photoshopped image is an internet joke and true life skill—"it's a 'shop. I can tell by the pixels and from years of seeing 'shops"
AI-generated images removes the maker's mark, and despite hilariously struggling with anatomy, seldom gets lighting or coloring "wrong" in a way that a human might.