What We Got Wrong About Music and AI in 2017

In 2017, I was working in digital innovation at Warner Music UK through Firepit Tech, looking at new technologies and trying to work out which ones might have an impact on the business.

A founder came in to demonstrate some machine learning technology. We weren’t calling it AI then. The example he used was a Michael Bublé track. The technology could analyse someone’s streaming behaviour, understand their musical taste and create a version of the track designed to appeal to them. Someone who listened to punk might hear a punk version. Someone else might get house or garage.

I was blown away. It was far more advanced than I expected technology to be at that point, and I left the meeting thinking about the implications rather than the demonstration itself.

I started talking about it with people across the business and with some artist managers, and the response was pretty consistent. There was discomfort around the idea of taking something an artist had created and changing it according to what an algorithm believed an individual listener would prefer.

I understood the concern. Artists put an enormous amount of themselves into their work. The writing, performance, production and decisions that happen along the way are part of a creative process that can be intensely personal. There is something about that process that deserves protecting. But I played devil’s advocate. If technology could make someone more likely to listen to a song, and that led them to discover the artist, hear more of their work and potentially become a fan, wasn’t there a case for exploring it?

Much of the conversation around AI in music has focused on whether machines can create music. That makes sense. It is the most obvious challenge to the creative process, particularly once technology becomes capable of generating music, voices and performances.

But what struck me about that demonstration wasn’t the idea that a machine could make a piece of music. It was that technology could understand an individual listener well enough to change how that person experienced the music. That turned out to be a much bigger proposition.

Technology now touches almost every part of how music is created, marketed, discovered and consumed. It can help artists make things, help companies understand audiences, automate work that once took enormous amounts of time and influence what gets made and what gets heard.

The demonstration I saw in 2017 was one early example of that shift, and we just didn’t know how far it would go. The concern wasn’t simply about technology. It was about control.

Who gets to change an artist’s work, and who decides what version gets heard? Does the artist even have a say? And if technology can produce endless variations of something, what happens to the value of the original?

Those questions are considerably more complicated now because AI is no longer sitting outside the creative process. It is becoming part of it.

I am fiercely loyal to the creative process. I think music is magical, and I don’t use that word lightly. A song can change your mood, transport you somewhere, remind you of somebody or make you feel something you can’t quite explain. I don’t want that reduced to a data point simply because technology allows us to optimise it.

But music has always evolved with technology. Recording changed what was possible, synthesizers changed what musicians could do, and sampling changed how people created a track in the first place. Streaming then transformed how music was distributed and consumed altogether.

AI is going to be part of that story too, and the question is what we choose to do with it.

There is a temptation with every new technology to focus purely on efficiency: how much faster, how much cheaper, how much can be automated away.

Those are legitimate business questions, but in music we also need to ask what happens to the value created by those efficiencies.

If AI allows a company to produce more music with fewer musicians and the financial benefit largely stays with the company, I’m not sure we’ve moved the industry forward. If it gives artists better tools, removes costs around creativity, helps them reach more people and creates more opportunity to earn from their work, that is a different proposition.

Perhaps we were asking whether AI was good or bad when the better question was what we wanted it to do.

In 2017, I was fascinated by a machine learning system that could take a Michael Bublé song and make it more likely to appeal to an individual listener. The people I spoke to afterwards were worried about what that meant for the integrity of the artist’s work.

I could see why and I could also see the possibility of reaching people who might otherwise never listen.

Nearly ten years later, I still think both arguments have merit.

What has changed is the scale. AI is no longer an interesting demonstration brought into a meeting at Warner Music. It is becoming part of the tools, platforms and businesses surrounding music, and the industry is still working out how it wants to use it and will continue to do so as it evolves.

I don’t think that means choosing between technology and creativity. It means being much more deliberate about where technology belongs, who benefits from it and what we are prepared to give up in return.

I saw the tension in 2017. Nearly a decade later, I still think it is one of the most important questions facing music.