Thoughts

I’ve been trying to find AI music.

That sounds ridiculous given the amount we’re told is being generated. Deezer has reported tens of thousands of fully AI-generated tracks arriving on its platform every day. Suno can produce a finished song from a prompt in seconds. AI music is supposedly everywhere.

But where is it?

Maybe I’m hearing it all the time and simply don’t know. Maybe it’s good enough now that I’m being fooled. Or maybe the slightly more paranoid parts of the music industry are overestimating how much people actually want to listen to it.

Recently, I got an answer of sorts in a theme park.

Every ride was absolutely hammering out music. Not cinematic Disney soundtracks. Think more UK travelling funfair: Waltzers, dodgems, flashing lights and enormous speakers pumping out pop-tech-house, pop-and-bass and commercial dance music.

A surprising amount of it appeared to be AI-generated.

And it was fine.

Actually, that’s understating it. It was doing exactly what it needed to do.

Nobody was standing next to the Waltzer discussing the production credits. Nobody cared about the artistic intention behind the drop. The music needed to be loud, energetic, contemporary and make being thrown around in a fibreglass car feel more exciting.

AI music might never make someone’s favourite song.

But standing there, I realised it doesn’t need to.

Music nobody needs to love

There is an enormous amount of music whose purpose isn’t to become anyone’s favourite record.

Library music. Corporate videos. Advertising. YouTube beds. Background music. Cheap sync. Hold music. Music in shops. Music on rides.

It has a job.

And that’s potentially where AI becomes enormously disruptive.

If you need 90 seconds of uplifting electronic music for a corporate video, the AI version doesn’t need to be better than the work of a professional composer. It needs to be good enough, available immediately and substantially cheaper.

That’s a very different proposition from asking AI to make the next Blue Monday.

It’s also where I think some of the conversation about AI replacing musicians gets confused. I don’t believe AI is going to replace our desire to make or love music. I absolutely believe it could replace many of the reasons we currently pay people to make music.

That’s a commercial crisis rather than a creative one.

And I should acknowledge that it’s easy for me to make that distinction. I make music because I love making music. I don’t rely on it to pay the mortgage.

If you’re a composer whose income comes from library music, sync or commercial commissions, the threat isn’t philosophical. If a client who paid you last year types a prompt into a machine this year, you’ve lost work. It doesn’t particularly matter whether the result has artistic meaning.

It did your job.

Does it actually matter who made it?

There’s an uncomfortable question underneath all of this.

If the music is good, does it matter where it came from?

AI-generated tracks have already appeared in the electronic music ecosystem without immediately announcing themselves as such. That’s particularly interesting when DJs are involved.

A DJ goes through new releases, hears a tune, likes it, buys it and plays it to a room full of people.

Then they discover it was generated by AI.

Was their original judgement wrong?

I don’t think it was.

If it sounded good yesterday, it still sounds good today. Deciding not to play it again because of how it was made is completely legitimate, but that’s a decision based on principles rather than music.

AI has separated two questions that, until very recently, we didn’t really need to separate:

Is this good music?

And:

Do I value the fact that a person made it?

I’m not sure they have the same answer.

Maybe listeners aren’t the undiscerning ones

It’s tempting for musicians to assume people who accept AI music simply aren’t listening closely enough.

I don’t think that’s fair.

We’re all undiscerning listeners sometimes.

I care enormously about music, but I don’t need to know who produced every track I hear in a shop. Someone on a fairground ride can be extremely discerning about whether the tune banging out of the speakers is making the experience better while having absolutely no interest in who made it.

Maybe musicians have simply overestimated how much listeners care about the things musicians care about.

We care about production. Influences. Equipment. Technique. Sampling. Scenes. Process. Originality. Who made the record and why.

Most people, most of the time, are asking a much simpler question:

Do I like this?

Streaming has already moved us some distance in this direction. Music increasingly arrives as mood and function: focus, workout, chill, party, sleep.

You don’t necessarily choose an artist. You choose what you want the music to do.

AI fits disturbingly well into that world.

The Velvet Sundown didn’t fool me

The Velvet Sundown is often held up as the great example of AI music fooling everyone.

I’m not convinced.

It arrived relatively early in this current wave of AI music, and to me neither the music nor the imagery was particularly convincing. People were already questioning whether the band was real before the story properly exploded.

Then being AI became the most interesting thing about them.

The mystery generated coverage. The coverage generated listening. The revelation generated more coverage, which generated more listening.

At some point, people weren’t listening despite it being AI. They were listening because everyone was talking about the AI band.

That’s not particularly compelling evidence that AI had quietly conquered music.

Show me something more boring.

Show me an anonymous track that thousands of people bought, played, danced to or heard in the background without ever wondering who made it.

That’s much more interesting.

Because that’s how I suspect AI music actually arrives.

Not with an AI band standing on a virtual stage announcing the future of music.

It just quietly comes out of the speakers on the Waltzer.

Everyone can make music now

I’ve also spent a slightly embarrassing amount of time lurking in Suno Facebook groups.

One thing I’ve found genuinely interesting is how many people appear to be making music primarily for themselves.

And I don’t mean that disparagingly.

Someone who can’t play an instrument, program a drum machine, mix a record or arrange a song can describe something they want to hear and, seconds later, it exists.

Maybe it’s about their wife. Maybe it’s a joke between friends. Maybe nobody else will ever hear it.

That’s actually quite lovely.

AI has democratised access to something resembling music-making. People who would previously have been entirely consumers can now participate in creating something.

Where it gets stranger is the number of posts that very quickly move on to:

How do I make money from my AI-generated music?

That seems incredibly naïve to me.

Making a finished piece of music was never the scarce resource.

There are already vastly more extraordinarily talented human musicians making brilliant records than there is money or attention available to support them.

The hard bit is making anybody care.

If AI lets millions more people generate essentially unlimited quantities of competent music, it doesn’t solve that problem. It makes it dramatically worse.

AI has democratised the ability to make music at precisely the moment it is destroying the scarcity that might have made that ability commercially valuable.

The streaming problem all over again

Unfortunately, I think we’ve seen this before.

Streaming is incredible for consumers.

For roughly the cost of an album a month, I can access almost the entire history of recorded music instantly.

It’s also been extremely good for the companies operating the platforms.

Whether the economics have been equally good for the people actually making the music is considerably more contentious.

AI feels like the same equation pushed even further.

Consumers get effectively infinite music, potentially tailored precisely to what they want.

Businesses get cheap music without commissioning or licensing it in the traditional way.

AI companies get another enormous market for their technology.

Musicians get an explosion in supply and the potential disappearance of some of the less glamorous work that helped people earn a living from music.

And why would everyone else choose a worse deal for themselves simply to protect musicians?

I don’t think they will.

That’s perhaps the most uncomfortable part of this.

Musicians can be completely right about what AI will do to their livelihoods and still lose the argument.

Streaming didn’t win because it was the best possible economic model for musicians.

It won because it was incredibly convenient for everybody else.

AI music may win for exactly the same reason.

I use AI to make music

None of this comes from a position of technological purity.

I use AI as part of making music.

It has made parts of my workflow easier. I’m perfectly happy for technology to remove tedious processes. I’ve used samplers, DAWs, software instruments and countless other technologies that allow me to do things that would once have required far more time, skill, people or money.

I don’t think difficulty is inherently virtuous.

But there’s something strange about describing the ability to prompt a finished song into existence as democratising making music, because it assumes the finished audio file was the entire point.

For me, it isn’t.

I’ve spent hours chopping samples that don’t quite fit. I’ve put drums underneath loops and discovered they groove in completely the wrong way. I’ve broken things, fixed things and occasionally discovered that the broken version was better.

That’s not friction preventing me from making music.

That is making music.

I don’t make music because I need more music.

There is already more music in existence than I could listen to in a thousand lifetimes.

I make music because I like making music.

AI can make that process easier. I don’t particularly want it to remove the process altogether.

Your favourite song

I still believe the best music will always be made by humans.

Not necessarily because AI will remain technically incapable of making something that sounds brilliant. Given how quickly the technology is improving, I’m increasingly uncomfortable making predictions about what it will never be technically capable of doing.

It’s because our relationship with music isn’t purely a relationship with sound.

It’s people. Scenes. Places. Artwork. Gigs. Fashion. Stories. Mistakes. Personalities. The circumstances in which something was made and the circumstances in which we discovered it.

A favourite record becomes attached to a particular summer, relationship, night out, club, group of friends or period of your life.

The mythology matters.

Perhaps AI will eventually generate an audio file that, in a blind test, I would prefer to something made by a human.

If that happens, I’ll have to concede that it’s good music.

But I’m not convinced that’s the same thing as becoming my favourite music.

And perhaps that distinction doesn’t matter very much to the future of the music industry anyway.

AI doesn’t need to make the next generation’s favourite record to transform the economics of making music.

It just needs to make the thousands of other pieces of music they hear every day.

I think the best music will always be made by people.

I’m much less convinced we’ll continue to pay people to make most of the music we hear.