When you consider that the people who develop this stuff consider that a good way to write a song might be to get an AI to generate a complete generic, formulaic song and then have a musician fix it, it's not so hard to understand why they think it might work for the law.
Because that isn't how you write good songs any more than it's likely how you write good legal filings.
Depends on your definition of good. All popular songs have been highly formulaic for a long time. That's what makes them popular, they aren't challenging to listen to and sound roughly like everything else. I'm not going to get into the nitty gritty of music theory, but any serious musician can tell you that writing a good song is striking a balance between boring and novelty. I think AI generated music does in fact get you 95% of the way there. The other 5% is already being in the elite levels of the music industry.
That's even more confused. They never said formulaic songs become popular, they said there is a formula to make popular songs. Those are different statements, and the latter is clearly not true.
A "hook" is a pretty pure example of musicianship, because it has to be "hooky".
Gregory Bateson said that "information is the difference that makes a difference".
For a musician as for music fans, hooks are like this: a thing will stop being hooky when everyone uses it.
It just becomes part of music language. Indeed you could argue that many of the fundamental qualities of popular music are hooky qualities pushed down a layer or two.
Carly Rae Jepsen's "Call Me Maybe" is an example of this. It's full of very obvious, well-understood musical tricks -- syncopation, breakdowns, subtle pacing changes.
But it's also full of maddeningly effective hooks that are incredibly, even deviously well-crafted. People write songs like this to feel proud of writing songs that get into everyone's heads, but also because it gets into their heads.
Since generative tools don't know how to focus on craft -- or even of its existence -- a good hook is something a generative algorithm will always struggle with, because it requires innovation, and because it is a complex, fragile element in itself.
The rest of it, as you say... it's likely quicker for an experienced musician to just do the work rather than keep poking a generative tool until it is right.
I bet you all of this holds true -- innovation, craft -- even to some extent in quite banal legal filings.
> but there is already a significant amount of automation for the mechanical parts.
Right -- and there can be ordinary, boring, individual scrutiny of what those pieces do, and data/code fixes for them.
I mean, it's no small amount of irony that the systems lawyers use are close to the kinds of "expert systems" that dominated AI development after the first AI winter.
Because that isn't how you write good songs any more than it's likely how you write good legal filings.