I’m Not Sure How I Feel About AI

I’m not quite sure how I feel about AI.

As a composer for commercial media, it would make sense for me to be against music AI. AI seems to threaten most immediately, library music (but its coming for the rest of us, sooner or later). A not-insignificant proportion of my royalty income comes from this sort of music that I’ve written in the past. I expect I’ll continue to write for libraries well into the future… unless AI takes over completely (which I predict it will eventually).

Writing for library music can be really fun – I’ve thoroughly enjoyed the production library albums I’ve worked on – but it tends to require quick writing to ensure a profitable output where originality and creativity sometimes become secondary to the need for placements and any ensuing royalties. The more you write, the faster you write, and the more categorisable it is, the more likely you’ll make a decent living from it.

By their nature, libraries thrive on having vast catalogs. This increases the probability they’ll have just the tune that the client is looking for.

However, composers who write for those catalogs will likely do better in more boutique, smaller, specialist libraries – which has mostly been my experience to date. Any client arriving at that particular library will more be likely, purely through the library’s specialisation and the fact that there are fewer other tracks to compete against, to pick that composers track. There’s a clear, economic, unspoken tension between these two positions.

In theory (and I’m sure in practise very soon) AI could rapidly produce vast catalogs of music in any given style using a dataset to learn from. Though not original, it could rival much of the music found in libraries today. I have a sneaking feeling this could happen within the next year or two.

In this scenario, AI might easily replace the human composers of library music, cutting off a source of revenue for many early-career composers especially. It’s unclear whether the composers whose data has been used to train the AI will be compensated, though I think the most visible companies will make some effort, the way Shutterstock has said its photographers, artists and visual designers will be compensated for clients using Shutterstock’s own in-house AI and from the sale of its visual dataset to other machine-learning entities.

Even if they are, I suspect the royalties generated will likely be small because, let’s face it, the point of using AI is to save costs. Machine-learned, neural network AI is cheaper to use than people. This is primarily why the big companies are so so very excited about AI: the bottom line (understandably).

Library composers appear to be the most immediately at-risk, but others that are likely next in the firing line are assistant composers (who are known to ghost-write for lead composers without credit) and composers for any media who specialise in pastiche styles – those who can write in any given musical style, analyze and synthesize any musical pattern, structure, or format, and create something still new enough to avoid legal issues and also familiar enough to avoid scaring off an audience that enjoys most what it’s familiar with.

And yet, and yet…

I really just want to input all my previous works, including both the classical and commercial ones, into an AI that can learn my style and generate new ideas. This AI would analyze and synthesize my work, and suggest new concepts that I may like. I just want to know what it’d spew out!

I expect I’d really rather enjoy collaborating with this AI to create something unique and unexpected, like a patchwork quilt of all my previous work and the AI’s interpretation of it. I’d take this as raw material to create something further, a more finished piece. I’m quite sure I’m not the only person to think this way – for this reason, I suspect AI collaboration is an area within composition that will become inevitable.

But the ethical considerations bother me. I think of the notion of ‘appropriation’ – situations where the less powerful are taken advantage of and their cultural creations are stolen for the personal entertainment and gain of the capitalist oppressor. I’m reminded of the ways we in the West have appropriated cultural artefacts and styles in the past (and continue to do so, the kleptomaniacs that we are).

If I could guarantee that the AI I was working with was using only my music to train itself, quite frankly I’d be there in a second. My ability to code is limited to html and css so I’d need an interface such as a plugin or a standalone app. I don’t think that technology is far from becoming a reality though.

Although I am conflicted and will continue to be so, I suspect… I still want to see what AI can do.