There are many recent issues in the public mind that concern, effectively, "what is copying?".
A.i. visual art has revealed curious biases towards ownership of style. In the visual domain, the premise of an a.i. system stealing a *style* is a concept, whereas in music it's around the premise that a.i. systems are copying, verbatim, copyrighted music.
I won't get into that. Instead, how you train YOUR dataset is on you. Meaning, GIGO - garbage in garbage out.
If you never look at how your favorite artists were influenced to create a style that they have, your dataset will NOT have the prerequisite information required for YOU to do anything akin to what they did.
Which is to say, understand antecedents.
Led Zeppelin gets burdened endlessly about how they've "ripped off" classic examples of blues songs. The problem being, for most of the songs if you take out the lyrics the average person would probably never be able to equate the "questionable" song with an example of it's origin. Different tempos, beats, chords, melodies.
But, that isn't to say there isn't a connection. Understanding how they got to *there* from over *there* is important. That comparison is what allows you to do the same. Because there is an intellectual difference between being *influenced*, and *copying* -
...but you can't argue it, or understand the difference, without having studied it.
You don't have to go more than 2 generations back. Meaning, if you're into a metal/rock band, chances are they liked some aspect of Led Zeppelin's catalog. You figure that out - that would be one generation back, then from there you figure out what the generation was that influenced that.
In that process, you see the lineage of creativity. There isn't a perfectly unique creation from nothing, that doesn't exist. You study the antecedents, and see how the antecedents depended upon other antecedents, and get *influenced by the processes that made those connections happen*.