I’m pretty much repeating what you said but I really just felt like writing this down ^^’

Music tends to be shared with an intent for someone else to get something out of it. A kind of communication. When people are listening to music they begin to think about what the music is (or the artists are) trying to say, or what feelings or reactions they are trying to evoke. Sometimes that resonates a lot.

“AI” output generalizes everything. Instead of just getting a hyperpop song, you get something closer to the most median of all hyperpop songs (in the dataset). You end up with no particular meaning or direction, and very little feelings besides what the genre’s general aesthetics are. Like a paragraph of text that only exists to be text, not actually meant to be read.

I chose hyperpop as example because artists tend to go a very specific direction; Usually taking a single element of pop music and heavily exaggerating it. Think heavy vocal manipulation, ambient audio design, sex appeal, or even the simplicity of house music. Taking the median of it all completely defeats the purpose.

I don’t have a link but there was a great video about an “AI rice cooker” and then a breakdown of the quite simple fuzzy logic actually inside