Better, But Still Not Good

After using AI for music the past month, some thoughts/takeaways:

• Music isn’t a toy. While it's amusing to hear how songs could be reinterpreted, those interpretations usually can’t easily be played on a guitar or piano without lots of rewrites. This is because it’s all a sonic confection. (Music is shockingly unforgiving without some degree of skill. Postmodernist approaches have always let us get away with just about anything).
• Quite expensive for very little control over musical elements. “Write a song about—-“ is not what a songwriter does. AI music generation is more a form of passive listening but is more “cool” in a McLuhan sense.
• Depressed mood while working with AI. Performing music uses different parts of the brain.
  
• A jukebox/slot machine of incomplete songs.
• Songs seldom make musical or syntactic sense. On a number of generations it used the chord changes for the chorus on a repeat of the first verse. That's kind of interesting but a human would never do it that way.
• The prosody and rhythm of the vocal samples is incredibly annoying, with a  tendency for singing words awkwardly on upbeats when they’d naturally be sung on downbeats by a human vocalist.
• The algorithms work more efficiently if the lyrics are already singable.
• AI as producer, finding the best session "players".
• Culminated in an "album" (playlist) of decent AI-generated songs, so not all for naught.
• AI music might be a dangerous insidious distraction, another example of Aesthetic Erosion where the land you once lived on falls into the sea. You have to continuously keep moving back from the cliff.