You are describing the limitations of current technology.
You haven't seen what AI can do. You have only seen what it can do today.
It's not magic now, and it won't be magic 10 years from now, also what you're describing isn't here yet, it might plateau tomorrow for the next 20 years.
35mm is reliable, and has the actual image on it, not a generic mathematical model.
Also, I don't care what it can do 10 years from now, imo, I look at a natural image and admire it for a reason, not to admire something a computer can shit out.
An artist makes an image, and a computer merely shits it out with no effort.
The costume designers, the prop masters, the set designers, the actors, the cinematographer, Lighting...
All of these people worked to produce a beautiful crisp textured image on 35mm film.
No AI model can surpass 35mm film if it's working with analog tape.
There's a wide shot from Emissary with Picard, and it looks like a video game NPC talking and not Picard.
There's another one from "Q-less", where John DeLancie's face is the same.
There are wide shots of Odo, where it would smear detail as Odo spoke, the 35mm would have more than enough fidelity and information there to see Rene Auberjonois give that performance.
Plenty of other shots get ruined by AI smearing it.
There's a shot from Voyager, where Tuvok's face looks very feminine, in low res you can tell it's Tim Russ, in High Res, it smeared it into some androgynous mess.
Maybe it can get good enough to see what I see and reconstruct it, but that would take an academic time to identify and name those issues, much less develop code and models that can predict and guess how that works, and quality is like an onion.
With 35mm film, you bypass those problems entirely, because, there's no guesswork, it's just scan what we got that day.
That's reliable, simple, and easy to do.
The AI requires a lot of other steps that are far more complicated.
That being said, for missing footage it's a good tool.
For automating the assembly of a new master, once you've scanned the image, "iConform" is a tool that's been used since 2012.
It uses the original tape master as a template, and with some human oversight, can identify the correct take, and overlay that film scan onto the tape, and produce a high definition master.
It was used to do X files, The Shield, The Wire, Baywatch and TNG season 2.
To say a computer should ever replace that, disgusts me, and should disgust anyone.
We admire the beauty of an image, because it wasn't exactly made by a factory (There's a more complicated argument we can have...DS9 is fiction, Hollywood is an industry.)
But IMO, to let a model alter that right now, is insulting.
Again, in 10 years an AI might be good sometimes, but here's the thing any moron can do it.
Only the studio can take care of a legitimate asset.
I don't want to watch AI Wizard of OZ, I want the real thing.