look
I think AI will soon be doing this to a level that no one would be able to tell a difference. It will be similar to an AI 'deep fake'. The AI will be trained with millions of film images. It will then 'know' what film looks like. It will be given a digital image then make it look identical to a variety of films. It will not fail like filter effects fail, as editing a digital image is making various adjustments what the AI will do is re-imagine the image as a film shot image and output that. It will not be very fun though!
Physical filters would still be making the work of the AI much easier. You can’t fake data.
AI will contribute to film in a more important way: Grain removal. We all love grain as photographers. But interpolation and even artificially raising resolution is hard when grain is involved.
Grain is not really noise in the Shannon - Nyquist sampling sense.
That makes it hard for conventional algorithms to identify and remove/filter out grain, with the aim of raising contrast and interpolating detail.
C-41 film has an insane amount of detail south of MTF 20, north of 100 lppmm, partially masked by grain.
Real resolution, that could come in handy for any number of applications. Unique to film.
CMOS sensors has low contrast in those ranges too, but due to the predictability and lack of a pseudo stochastic physical substrate structure, contrast is easily pulled up (and of course an incredible amount of research has gone into that by now).
Look at some of the sparse sampling ray tracing examples from the last few years to get an idea of the real immediate potential.
Won’t that take out the uniqueness of film? Not at all!
It will only give us more data to work with.
Of course we will need the inexpensive, good scanner alluded to above first, to be able to extract the detail without all the artifacts and plain inability of decades old scanner technology.