Remember, we're trying to discern the difference in expression. Any kind of music, whether it is written by a human or by AI sounds vastly different when it is played through biological fingers. You can't escape the rigid modality of studio produced timings and sounds. This is but one of the reasons people are perhaps unaware of their attachment to live performances over studio "recordings" - in the sense that the performance embodies the entire essence of the musician (including their flaws) but makes them unique in a way that midi and studio processing would wipe clean.
RE: Deep learning in music?