An AI and Music Manifesto

As you might know, I have been working with AI and classical music since 2018, when I was appointed composer in association with the BBC Philharmonic and RNCM PRiSM for my PhD to explore the creative and disruptive uses of this technology. Since then, I’ve tried to let the music do the talking.

However, AI is increasingly dominating the cultural landscape, and a specific view of what AI is and does is ascendant in the media and public discourse. I thought it is time I set out my personal, artistic, manifesto on this technology that I have developed over the last few years. These are things I strive towards and improve on in every piece – mileage may vary in how successfully I’ve achieved them.

AN AI MUSIC MANIFESTO

1 – Make Amazing Sounds

2 – Don’t Steal

3 – Reject the Mediocre

4 – Reclaim Experimentation

5 – Centre the Human

Helpful Notes to Myself

MAKE AMAZING SOUNDS

AI is capable of making incredible, breathtaking, complex and, above all, new sounds in so many ways. I love finding sounds I have never heard before and imagining where they might go, or come from, in my work. It’s deeply disappointing that many people aren’t even aware this is possible. Generating an entire piece of music, end-to-end, by typing a prompt into a startup’s webpage won’t make an amazing sound; not that you would know, because you won’t listen past ten seconds. The world is overflowing with AI-generated music that nobody will ever hear. Forget imitation; never replace, only augment. Make something you’ve never heard before – that couldn’t exist before. This doesn’t only mean creating new noises with AI. Think about how it can reframe how to play an instrument, or what an instrument is, or what the distinction between music and sound is. Use those thoughts in your work. Make amazing sounds. This is essential. Without this, what are we even doing here?

DON’T STEAL

We are told, often, that AI cannot function without scraping music from the Internet without creator, or owner, consent. This is not true. AI can be trained on any data, including very small datasets. My work with the BBC Philharmonic is trained on recordings solely owned by the orchestra of music that is out of copyright. My work with soprano Stephanie Lamprea used recordings of just her voice, improvising. My work with International Contemporary Ensemble used publicly available recordings of a single songbird. This goes beyond an ethical stance. By using specific, consensually shared music/sound the work becomes more local. The technology acts in service of the human, of the project, place, person or time. The music is better, more meaningful, more thought-provoking, when you don’t steal.

REJECT THE MEDIOCRE

While this point follows naturally from the first two, and is always a priority for any artist, mediocrity has a special relationship with AI. AI is a profoundly mediocre technology. Its foundational principle is that of averages, of finding common patterns in data, in linking different ideas together into one line. We’ve all noticed its natural tendency when generating text, or images, or music, is to create statistically sensible – predictable – content. It should go without saying that this mediocre AI should be rejected when using AI in art. Who wants mediocre art? We can also use AI to reject the mediocre. AI can reveal trends in complex data. By listening to what it creates, we can hear a computer’s understanding of what is mediocre. We can then make work that rejects this average. Maybe in a few years, the mediocre trends will actually be quite interesting.

RECLAIM EXPERIMENTATION

Musicians are experimenters. Look at radio orchestras – institutions set up a hundred years ago to push the boundaries of music with new technology. Experimentation is exciting. It’s how we learn more about ourselves. All things were once an experimental new technology; the orchestra is a museum of musical technologies. AI is a vast technology, full of contradictions. We shouldn’t be denied the joy of experimenting with it, of using it to express ourselves more deeply and to more people, even if others are using it for terrible things. Loud, joyful experimentation is one of the most powerful ways to resist technology simply being something that happens to us.

CENTRE THE HUMAN

Think about human bodies. Where possible, make AI something you can touch, move, practice, perform with, express yourself with. Listen to what musicians want – not just what they want from AI (or don’t). Design your tools to be useful to musicians. Music is not data to be consumed; it is an experience in time. Remember that this technology is doing things to us. It is shattering our attention spans; it is radicalising us online; it is directing traffic; it is saving lives; it is destabilizing the nation-state; and it is accelerating. When you put AI in your music, you are saying something about being a human in this kind of world.

I might write a little more on each of these in the future, pointing to work by other artists that I think exemplifies these ideas.

The more of these tenets a project engages with, the more interesting it is to me. Generating an entire piece of music, end-to-end, engages with none. It certainly doesn’t centre the human nor encourage any form of meaningful experimentation as the user has no real control over the algorithm (even if they are given a few sliders to push). These systems are designed to replicate existing genres and artists, making them by definition mediocre and unlikely to produce an amazing sound. They are built on non-consensual data usage. I think we can do so much better.

if you’ve come here from Twitter, know that I’m not really planning to use that platform anymore. If you want to hear from me, follow me on Bluesky

https://bsky.app/profile/robertlaidlow.bsky.social

Report update