I disagree - there is probably more usable music than usable text - but music has a Time dimension, that's a hard problem see moving images and music is really sensitive to errors a wrong tone destroys everything while our visual system is quite forgiving. We like AI pictures even if they are totally incongruous but we probably ditch music if there is a screeching sound added. Music is quite mathematical in nature ML can take a shot after it mastes the five finger problem ;)