Different Kinds of Knowing

 

Eventually, music AI will become more sophisticated, and we will be able to select more complex chord changes, choose keys and meters, and so on. But if we can simply select those things from a drop-down, is there a difference between the simple selection of options and intimately knowing something? For example, if you generate a jazz head that is composed of ii-V-I patterns through several keys, and all the dominants are altered, or have complex extensions, what would be the result? Would it sound like the jazz that humans actually compose and play? 

Music composition with LLMs, while interesting in its syntactical connection with music and language, doesn't rely on any "procedural knowing": you're simply clicking on things in an interface or dashboard, letting it generate iterations and selecting the ones you like. It's like photography in that sense, choosing a decisive moment from 10 or 12 captures. If we look at creativity being untethered to a domain, where a domain doesn't (always) require procedural knowing, perhaps the results can be more interesting. 

Yesterday I was generating music based on translations of my lyrics to Esperanto out of curiosity. It's something that was never possible before, especially generating many possible versions of someone singing your lyrics in Esperanto--or any other language. While it's interesting, there are certain aspects coming from procedural knowing, or intimate knowledge or skill that create a cognitive dissonance, even frustration. Why am I stuck with cliche chord changes, or is it that ultimately to listeners it doesn't matter? AI is already making that decision for us, where we just capitulate to the constraints of the tools we're given rather than use all the tools (and knowledge) that were always available.