I am still learning, but I have so far only applied inverse methods to systems where the physics is well-defined and the model already accepted as "real". For example, CT scanner algorithms which model the photon counters and attenuation through matter (although this can get complicated...), with the output being data in well-defined format (like the distribution of matter). I do see people using inverse solvers or AI to derive something which isn't an "image" but a model of the physics (does that make sense?). The extreme version of this which I have seen once before, is a paper using AI to extract algebraic forms of PDEs from videos of fluids. To what extent does the AI user have to understand/"pre-model" the physics, and to what extent does the AI generate understanding in some way? Apologies if this is misinformed as I am still outside the research community, and don't very actively read papers.
Currently I am trying to learn background to understand electromagnetic properties of materials with an interest to apply this to computer graphics. I eventually want to learn what processes are currently done for extracting material properties from scans, etc., along the lines of the classic gonioreflectometers. In this space, from the papers I look at, it looks like AI is unavoidable and I am sure an extremely useful tool but I definitely want to be careful not to get lost in misunderstanding.