Originally Posted by etcjoe
I lost the thrust of my point, I think we all create music, using what we learned from other music, in a similar way that our AI engineers here at my work describe to me what is going on.
When you create music, you don't draw from the audio signatures of millions of songs to generate not only replicate chord progressions, but instruments, vocals, and studio effects.

Quote
It just has the capacity to ingest 1000's of those songs and use it to generate its output.
Not "1000's", but millions of songs.

Songwriters remember songs.

AI doesn't remember anything - it stores the information in a neural network and uses that to create songs that are based on that information.

AI is designed as a tool that learns the elements of millions of song - including the voices and the instruments - and then creates new songs by mixing elements of those songs. Every song created is a remix of a prior song.

That's entirely unlike how people learn and replicate music. As a small example, AI isn't going to include ideas inspired by a T.S. Eliot poem, the death of a loved one, or a seeing a rose in a garden.

Every word that AI writes is taken from lyrics from someone else's song, using voices copied from other singers.

Quote
Do I have to attribute Chuck Berry when I use a double stop lick in my playing? Nope, but everyone that hears it and knows Chuck Berry can pretty much know it was lifted from him.
But AI doesn't know whether something it replicates is copyright infringing or not, because it's all potentially copyright infringing. Everything it does is created by a process of copying. The difference is that the process of replicating the sounds can usually hide the source material because there's so much of it.

Quote
I know in my experimenting, I can't get it to create anything that sounds like something else I already know except in style and genre etc...
That's not because it can't do that, but because the designers have intentionally hidden those controls from you. But AI can very much create songs with Elvis' voice, or riffs created by Chuck Berry.

Quote
...which it nails almost every time. does it use chord progressions that are prominent in those styles? Sure it does just like every songwriter on the planet. ii V I anybody?
Because it mixes together inputs, AI songs tend to be an average of the training data.

So it will tend to make generic choices, except when it drops signature sounds, like guitar licks and vocals.

Equating the process that AI uses with the process songwriters uses is a false equivalence.

AI doesn't have its own voice, the ability to play guitar, or the ability to consider the emotional impact of its lyrics.

Quote
It will certainly be decided by the lawyers and the courts. The big "record" companies are mad because they think they should be the only ones exploiting artists. how dare the computer guys get in on their game. If anybody is going to rip off an artist they want it to be them and nobody else!! LOL.
Unfortunately, this is likely to be 100% accurate.

It seems that companies that own the copyright aren't interested in preventing AI from competing with artists, but are more interested in making sure they get a piece of that pie.

-- David Cuny

My virtual singer development blog
Vocal control, you say. Never heard of it. Is that some kind of ProTools thing?

BiaB 2025 | Windows 11 | Reaper | Way too many VSTis.