I would go with, if the song could be reasonably performed live by the human publishing it, and have a similar sound to the recording, then its fine to keep in. The issue I have is songs like this one, where there is no way anyone could even try to perform it
https://open.spotify.com/track/0jGJtiDfEO9syfSL8AshBF?si=b92...
That doesn't really work for electronic music though.
Electronic or not, whether messing with a buchla or producing via vsts and changing knobs around still ends up with a human feel and human choices in a way AI music doesn't.
And what about the line between triggers or samples? I can play some impossible AI music if I sample the impossible parts and just say it's a sample played on a synth or whatever.