Good Sunday to Everyone!
Yesterday, I was in a conversation with some friends (one of which is also a songwriter and had a band in the 70s for a while), and we got on the subject of A.I. and songwriting. I had heard a recent song a colleague at work was sharing from a tool called Suno. I was expecting it to sound pretty trashy and forgettable. It wasn’t. It was a touching ode to a friend that had left work to pursue a life-long dream of hiking the Appalachian trail. The production? As good as any other song on the radio. The vocal? Reminiscent of any male folk singer-songwriter in rotation. It was pleasant, touching even. With a loose idea, some prompts, and some time you can create a song you could post to Spotify. In one way, it’s chilling for anyone who writes songs. In another way, it begs the question: Where’s the human in this?
Humans? We are flawed. We can’t sing perfect all the time. Our playing can be scrappy. I’m sure you could ask an A.I. to try to sound a little scrappy — maybe mimic someone like Bob Dylan. But there’s a little something extra in a human performance. Call it charisma or attitude or sex appeal, but A.I. ain’t got it yet. It’s getting there, though. Ever hear of this A.I. singer-songwriter? Anna Indiana: https://www.dexerto.com/entertainment/ai-singer-songwriter-launches-first-song-to-mixed-reactions-2400842/
“She” has a kind of generic Kelsea Ballerini approach (especially that song “Front Row Seat”). But this is the kind of thing any non-musician who just wants to hear some songs might like. There’s nothing particularly wrong with it. It might be a bit boring. There are no background vocals. It’s too “in the pocket” of radio-friendly arrangements. I suppose it mostly lacks a thing we call “heart” and another thing called “originality.” But it’s not incompetent. It’s just run-of-the-mill. However, this is the first stab at creating an A.I. artist. It’s only up-hill from here.
But is this some clarion call to give it all up and move on to other creative endeavors? Let’s say you’d rather write novels? Well, no, that’s been invaded by A.I. as well with similar feedback. What about poetry? Again, A.I. is there as well. Anything basically “written,” A.I.’s got us covered.
Is it any good? Yes and no. It’s polished. In fact, some people have serious difficulty differentiating the A.I. poem from an actual poet. But these A.I. train on huge amounts of data. They are basically mimicking, copying and then spewing forth similar prose. They become imitations of someone else as a result — some might say even “pale” imitations of someone else.
Is it copyright infringement? This is where the law is woefully behind. If I write a song and put it out in circulation, it becomes A.I. territory. I could be imitated at any moment… if the A.I. can find me! It would not take my lyrics or sounds or melody specifically. That would be plagiarism and would definitely violate copyright laws. But it could write “in the style of” and come up with something remarkably close. The A.I. is only as good (in a way) as the material it can feed on.
And this where a different question gets asked: Is it ethical? The A.I. train left the station years ago before anybody was able to clarify an answer. One could argue that the human singer-songwriter is writing “in the style of” other artists ALL THE TIME. We train on other singer-songwriters, right? We are a sum total of our influences. An A.I. is no different. No one’s music exists in some vacuum in space somewhere. Some might try — Bjork anyone? But we are social animals with influences and expectations and all kinds of rules all over the place. A.I. is just trying to learn.
It might sound like I’m defending A.I.? But I see this debate in shades of grays not black and whites. I think an A.I. assistant for a human singer-songwriter could be a brilliant and liberating tool. But it’s a tool just like a hammer 🔨 or a blender. It’s not taking over, not yet. What I think needs to happen is a collaboration between a favorite A.I. helper, your own mind, and your own experiences. An A.I. doesn’t have the emotional tapestry of a human. It’s a predictive model, a landscape of 0’s and 1’s. With some collaboration, though, we could end up with songs that have both a soul and an attractive production in a half hour, not three months. I suppose if we’re looking for speed and efficiency, this might be the answer.
Does it cheapen the result, though? 🤔 I wonder. Production is ALSO quite an art. For instance:
- The sounds we pick for the arrangement.
- The parts we create.
- The hooks in the tune… the way the guitar riff hits in bar 16?
These are all part of the creative stew as well. And if I use an A.I., I’m leaning on it to define those sounds for me. It’s hallucination becomes MY production. My guess is that I’ll find it generic and boring. With some training on what I do, though? It could get spookily close.
So as an experiment 🧪 (in April), I’m going to take a scratch pad of a song I’ve written and see what Suno does with it. I’m curious.
More to come in future posts.
🎹🎶🎸