There's people doing both types. Look up survey of mechanistic interpretebility of language models and survey of explainable AI for neural networks. Those will give you many techniques for illustrating what's happening.

You'll also see why their applications are limited compared to what you probably hoped for.