It’s a thorny issue and one of the most polarising subjects in music today! And in art. And 3d. And filmmaking. And coding. And writing. And web design. And. And . . .
When people get passionate about AI, they are often talking about generative AI - i.e. the creation of new content. But there is also assistive AI, i.e. augmenting workflows to help with a task.
Personally, I don’t use AI to generate new ideas or sounds for the simple reason that I don’t need to. That sounds a bit pompous, which isn’t my intention. What I mean is I have enough musical sense (just!) to get huge satisfaction from noodling on a guitar or tinkling on a keyboard until something interesting falls out, which I then work up into a song. AI would take that away from me and make the songwriting process less satisfying.
That said, I am talking about generative AI, such as Suno. Of course there are plenty of assistive AI tools that provide utilities such as stem separation or mastering. I have used these with varying degrees of success, mostly just about acceptable, but not enough for me to want to carry on with them. I’ve also used Harmony Bloom and Eternal Arps - not AI as such (I don’t think, although it could fall under the assistive AI umbrella) but still, it wasn’t me creating the output. Certainly, I took whatever those plugins spat out and developed it into a song, but those elements were not created or imagined by me. Audio repair? Noise reduction? Should I spend hours doing it manually? Nah.
For me, where the debate gets really tricky is when you consider using AI to generate, for example, a single instrument part (or the arps I mentioned above). Some would argue that if you can’t play it (or programme it) yourself you should hire someone to do it for you. Others would argue that doing so might be out of reach for them financially, or practically. Does that mean they don’t deserve to have bagpipes in their song? Or should they take an ethical decision not to write a song with bagpipes if they can’t afford a player? What about AI vocals? To me, they’re horribly generic as an up front voice in a song, but what about for demos, or as a guide track before your vocalist turns up? Or for backing vocals? I can’t afford to pay three backing vocalists, and my voice isn’t strong enough to do the job. Does that mean I shouldn’t have backing vocals in my songs? (In fact, I run a theatre company and I have access to whole bunch of singers that help out with backing vocals, but you get the point).
And then there’s the AI -v- samples/loops debate. I very rarely use samples and the only time I used a loop (that I didn’t create) was for a conga track. I didn’t create it or perform it or even come up with the pattern. It was a free loop I found online. Would it have been wrong for me to use AI to create that part instead? What’s the difference? Or should I not have had the congas and ended up with a lesser song?
I think one of the most important things in the AI debate is disclosure: making it clear if a song is AI generated. But how do you enforce that? And does that mean you have to disclose the use of AI if a pad track or backing vocal was generated by Suno? If so, why wouldn’t you have to disclose that you mastered a song with Landr, or mixed it with Ozone? Or used Midjourney to create your album cover? or ChatGPT to write some blurb for your release? Or Gemini to remove the bass player from your band photo after you fired him?
Like I say, it’s all so polarising, and people will have very valid arguments for and against everything I’ve said above. At this point in time, there are more questions than answers. The most important thing is that as musicians, or songwriters, or producers - of any standard - we have a choice to use or not use AI. At the moment we also have a choice whether to listen to AI generated music or not. But that choice is rapidly diminishing as the lines become more blurred.