For most of the history of recorded music, making a song involved some kind of obstacle. You needed to know how to play something, or know someone who did. You needed a microphone, a studio, a computer, a producer, a certain amount of money or simply enough time to learn how to turn whatever was in your head into something other people could hear. None of this guaranteed that the music would be good, but it meant that getting from an idea to a finished recording required effort.
That barrier is disappearing surprisingly quickly.
Today, someone who has never played an instrument can describe a song they have in mind and hear something resembling it within minutes. They can ask for a melancholic piano piece with a female vocal, a slow build and an enormous chorus, change the mood, alter the instrumentation and try again without ever touching a piano or opening a recording studio.
It is difficult not to find that impressive.
It is also difficult not to wonder what happens to music when the difficult part is no longer making the song.
We have been here before, in a way
There is a tendency to talk about AI music as though it arrived from nowhere, as if music suddenly went from human hands directly into the hands of machines. It didn’t.
Music has been absorbing technology for a very long time.
The synthesizer changed what a musician could do with sound. Drum machines changed the role of rhythm in popular music. Sampling allowed producers to take fragments of existing recordings and turn them into something entirely different. Digital recording eventually meant that an entire studio could sit inside a laptop, making it possible for someone in a bedroom to produce music that once would have required a room full of expensive equipment.
Every one of those changes made some people uncomfortable, particularly when a new technology appeared to remove part of the skill that had previously been considered essential.
Yet musicians didn’t disappear. They adapted, experimented and, eventually, turned the new technology into part of the musical language itself.
AI feels different because it isn’t merely giving us another way to produce a sound. It is beginning to take part in decisions that used to belong almost entirely to the person making the music. It can suggest melodies, harmonies, arrangements, lyrics and entire musical directions, sometimes with remarkably little human intervention.
That changes the relationship between the person and the song.
The strange part is how quickly an idea can become a recording
Think about what normally happens when somebody has an idea for a song but doesn’t know how to make it.
Maybe they hear a melody while walking home. They hum it into their phone, forget half of it, try to play it on a keyboard and discover that the version in their head doesn’t quite work. Perhaps they eventually find a musician to help them, or spend months learning enough to turn the idea into something usable.
There is something valuable in that process, even when it is frustrating. The limitations of the person making the song inevitably become part of the song itself.
AI removes some of those limitations.
That can be liberating, particularly for people who have always had musical ideas but never had the technical ability to express them. A person who cannot play guitar can now experiment with guitar arrangements. Someone who cannot sing can hear a vocal interpretation of a melody. Someone who knows nothing about orchestration can begin playing with sounds they would never have been able to create alone.
In that sense, AI could make music more democratic.
The interesting question is whether making music more accessible will also make the music itself more interesting.
We have never needed difficulty in order to make great music
It would be easy to romanticise the old barriers and assume that the more difficult something was to make, the more authentic it must have been.
Music doesn’t really work that way.
A technically brilliant musician can make something completely forgettable, while somebody with very little formal training can stumble upon a sound that stays with people for decades. Some of the most interesting moments in recorded music came about because somebody didn’t have the right equipment, didn’t know the accepted rules or simply made a mistake and decided to keep it.
The history of music is full of accidents that became style.
That is why I don’t think the fact that AI makes music easier to produce is, by itself, a problem. Difficulty has never been the measure of artistic value.
What matters more is why the choices were made.
A musician might spend an entire afternoon deciding whether a guitar should enter eight seconds earlier. A producer might throw away a technically perfect vocal because another take has something slightly broken about it. A songwriter might change one line because it reminds them of a person they once knew.
Those decisions are not necessarily visible in the finished recording, but they are part of what gives music its character.
AI can produce options. It cannot tell us why one of them matters.
This is where the question of human experience becomes unavoidable
There is something peculiar about hearing an AI-generated song about heartbreak.
The system can understand what heartbreak tends to sound like. It has encountered thousands upon thousands of songs built around loss, longing, separation and regret. It knows that certain chords, tempos, instruments and lyrical images tend to produce a particular emotional effect.
But it has never waited for someone who didn’t come home.
It has never found an old photograph in a drawer and suddenly remembered what it felt like to be twenty.
It has never driven past a familiar street after a relationship ended and felt the strange physical sensation of being somewhere that suddenly belonged to another version of itself.
A human songwriter can bring those experiences into a song, even indirectly. They don’t have to tell their own story, and they don’t have to have lived exactly what the listener is living, but somewhere behind the music there is usually a person who has experienced something.
AI knows the patterns of those experiences without experiencing them.
That doesn’t mean an AI-generated song cannot move us. We respond to sounds, melodies and words, and those things don’t stop affecting us because we know a machine helped produce them.
But it does make the experience different.
When we hear a song that feels painfully specific, part of the pleasure is often the suspicion that somebody else has felt something similar. We recognise ourselves in another person’s attempt to make sense of their life.
That human exchange is difficult to reproduce.
And yet, maybe the human doesn’t have to disappear
There is a temptation to divide the future into two camps: human music on one side and AI music on the other.
I don’t think it will be that simple.
A musician might use AI to generate a hundred possible arrangements and reject ninety-nine of them. Another might use it to find an unexpected chord progression and then spend hours reshaping it until it sounds nothing like the original suggestion. Someone else might use an AI-generated vocal as a placeholder before recording their own voice over it.
In all of those cases, the machine is part of the process without becoming the artist.
This isn’t fundamentally different from the way musicians have always worked with tools. A synthesizer doesn’t decide what a musician wants to say. A recording program doesn’t know why a particular take feels better. A guitar pedal can create a sound the player couldn’t make without it, but the decision to use that sound still belongs to the person.
The difference with generative AI is that the tool is much closer to the creative decision itself, which means that the person using it has to become more deliberate, not less.
If a machine can give you fifty choruses in a few minutes, the interesting question becomes why you chose the one you kept.
That choice may ultimately tell us more about the human behind the music than the original prompt ever could.
Then there is the problem of where AI learned all of this
There is another part of the conversation that shouldn’t be buried beneath the excitement about what the technology can do.
Generative music systems are built by learning from enormous amounts of existing material, and much of that material was created by people who never agreed to become raw material for an AI system.
This raises difficult questions about copyright, consent and compensation, and those questions are still being argued across different legal systems and within the music industry. They are not going to disappear simply because the technology is impressive.
There is a particular irony here.
The better AI becomes at sounding human, the more obvious it becomes that its ability comes from learning from human music.
That doesn’t automatically make every use of AI wrong. Artists have always learned from other artists, and musical traditions have always developed through imitation, borrowing and transformation. But there is a significant difference between a young musician hearing a hundred records and developing a style of their own and a commercial system ingesting vast quantities of recorded music at a scale no individual human could approach.
The question isn’t simply whether AI can make music.
It is who gets to benefit from the music that taught it how.
What happens when there are simply too many songs?
This may be the part that interests me most.
If AI makes it possible for almost anyone to produce a finished song, we are going to have an extraordinary amount of music.
Streaming already gave us more music than any human could realistically listen to. AI could take that abundance to another level, where creating a new song becomes so easy that the world could be flooded with music designed for very specific moments: music for working, music for sleeping, music for a rainy Tuesday, music that sounds like a memory you can’t quite place.
At some point, the problem stops being production and becomes attention.
We are already seeing this in streaming. There are millions of albums available at the touch of a screen, yet most of us continue returning to the same relatively small collection of artists and songs. Having more choice does not necessarily make us more adventurous. Sometimes it simply makes the familiar more comforting.
If there are eventually billions of AI-generated songs competing for attention, discovery may become even more valuable.
And that could make human curation more important, not less.
The person who says listen to this one may become more valuable than the machine that can generate another thousand.
Maybe taste becomes the scarce thing
For a long time, being able to make music was a valuable skill because making music was difficult.
In a world where machines can help almost anyone produce something polished, technical ability will still matter, but it may no longer be the only thing that separates one creator from another.
Taste will matter.
Judgement will matter.
Knowing when something is finished will matter.
Knowing when it isn’t will probably matter even more.
A musician who can recognise the one strange, imperfect sound worth keeping may have more to say than someone who can generate endless technically flawless tracks.
This isn’t really a new idea. Producers, editors and artists have always made their reputations partly through knowing what to leave out. The difference is that AI could give us vastly more material to choose from.
And when everything is possible, choosing becomes a creative act in itself.
I don’t think AI will make human music disappear
If anything, I suspect it will make us more conscious of what we value about human-made music.
There will be people who don’t care how a song was created as long as it sounds good, and there is nothing particularly wrong with that. There will be listeners who prefer AI-generated music because it can give them exactly the sound they want. There will be musicians who use AI extensively and others who refuse to touch it.
All of them can exist at the same time.
The more interesting music may come from the people who find ways of using the technology without allowing it to make all the decisions for them. Just as sampling became an art form, just as electronic production became a language, AI may eventually become another part of the vocabulary musicians use to make something that could not have existed before.
But there will still be something we look for in music that has nothing to do with technical perfection.
We want to feel that someone meant it.
We want to hear a voice crack at the wrong moment and realise that the imperfection is exactly what makes the recording work. We want lyrics that sound as though somebody had to get them out of themselves. We want strange decisions that make sense only after we’ve lived with a song for a while.
AI may become extraordinarily good at giving us music that sounds convincing.
The harder thing will be making music that feels necessary.
And perhaps that is where the human will remain easiest to hear.
Not because humans will always make better sounds, or because machines will never understand musical structure, but because a song can be more than a successful arrangement of sounds. Sometimes it is a record of somebody trying to say something they couldn’t quite say any other way.
As long as people still have something they need to say, there will be a reason for them to make music.
And that, more than the technology itself, is what I think will decide what survives.