I learned Basic so I could program a Yamaha CX5M music computer. This was the mid-1980s, and the machine could do something that felt, at the time, almost transgressive: a subroutine that randomly generated notes for you. After entering the notes, you could "humanise" them by generating data that would add errors and variety to pitch, duration, velocity. Making the music sound more "human". It sounds easier than it was, and it begs the question: why does human = error?
That was my first encounter with the machine as creative collaborator, and I've been thinking about it, in different forms and different levels of complexity, ever since. I've used computers to generate material, to manipulate material, to model material, to calculate long lists of numbers to synthesise material. I've done this for myself and for the composers I've worked with over thirty years. The machine has always been part of the process.
So when people ask me what I think about AI and music, I'm not starting from zero.
A few positions I hold, offered without full development for now:
We would be crazy not to use it. Running a race now in shoes from the 1960s, or doing research using only your local library, would make no sense today. AI is a tool, and refusing tools on principle is not a creative position. Preferring inconvenience and dressing it up as ethics is not a position. It's a posture. The translation of the 2002 paper now on this site was prepared with the assistance of Claude. The intellectual work — writing the original, knowing what the terms mean, deciding what it should sound like in English — was mine. The mechanical labour of converting French sentences to English sentences accurately was the tool's. There is a distinction worth making explicit: made by AI and made with AI are not the same thing. One is a product; the other is a process. The confusion between them is where most of the bad arguments live.
But there is a large "but". Tools that provide turnkey creative solutions are not the same as tools that require mastery. A perfect Beatles tune generated by a machine and a perfect mix generated by an intelligent assistant are not the same kind of thing, even if both involve AI. One replaces the creative act; the other extends it. The distinction matters. Most of the current conversation about AI in music collapses it.
Transparency and a code of conduct are not optional. This is where I feel most strongly. The question is not whether or not to use AI. Too late, we are and of course we should, but how to use it with honesty and to be transparent about what it contributed is important. The collaborative act between humans is not the same as the machine assistant. The human collaborator is "flawed", embodied, historically situated, and has intimate, unique, memories of existence. That makes them irreplaceable, even when the machine is faster or more accurate.
The interface is still the question. It always was. My work for thirty years has been about the space between human intention and machine output: how to make that space musical, how to give a performer genuine control over it, how to ensure the human remains the author of what comes out. AI doesn't change that question. It makes it more urgent... and it might help solve some of those problems.
I was invited to speak on this at a conference in Venice in late 2022. I'll develop that talk into something longer. For now, the question I'm sitting with: where exactly is the line between a tool that serves creation and one that replaces it? I'm not sure I can define it precisely — but I know it when I hear it...