Mason  🏃‍♂️ ✂️

Yeah. (e.g. play arpeggios of equally spaced notes from random chords all from the same randomly-selected key)

I think you can create something that *sounds* good in that way, however that is because we have heard it before so it is not something new. There is a difference between music *sounding* good and being objectively good. Creating the latter is hard

Again: I admit vast ignorance here. But it seems analogous to AI visual art. You may get pretty patterns via following an algorithm (I recall when "visualisers" were popular in Apple and Microsoft music players). I mean: pretty to look at. But: original art? Actual creativity?

Not creative currently, whether it is generative art or generative music. The programming that causes such pretty phenomena cannot critique its own efforts - past, present, and future - unless a skilled enough programmer supplies the prerequisite "instruction from without".

Yeah, the question isn’t really about whether a program can make something that sounds good. It’s “What does the program do when it confronts a musical problem - when something sounds bad?”

I'm having a really difficult time mapping any of these concerns/assumed limitations to the actual process of e.g. training a neural net. Do you need human feedback? Yes. But that's exactly what platforms like Instagram & Soundcloud are for

It’s a philosophical question: can an algorithm be written such that largely *absent frequent human interventions* good music can be produced (in other words: some kind of automated composition). No problem with the idea that *with human creative input/feedback* it’s possible.

Humans broadly suck without external human feedback, too