AI and The Common Song

A note on how this began. A few weeks ago, two people who did not know each other argued in a comment thread under an article about the Fourth Turning. The argument was good. It kept going — through one essay, then a response, then a series of exchanges in which disagreement reliably made the subject sharper rather than smaller. This piece came out of that. It is an attempt to think through, together, what AI might do to one of the oldest things human beings share. We have left some of the disagreement in.

I. The song that was not there

A few days ago I heard a new song by one of my favorite artists.

Everything was in place. The voice was his. The phrasing was his — that particular way of arriving slightly late on a line. The production had the texture I associate with his records. The harmonic vocabulary was the one he has used for twenty years. I listened to it twice. I liked it. I made a small mental note that he was still good.

Then I found out it had been generated.

I want to be precise about what happened next, because the obvious reactions did not occur. I did not feel that the song had become worse. It had not changed. The same air moved the same way. If I had never learned the truth, the song would simply have joined the others, and in ten years it might have carried a memory of this summer.

What I felt instead was disorientation about the object itself. Not is this good but what is this. And underneath that, a stranger question:

What exactly had I been listening to when I thought I was listening to him?

That question turned out to be a door. Behind it was not a debate about the music industry, or about artists’ rights, or about whether machines can be creative — all of which are real, and all of which are being argued elsewhere by people better placed to argue them. Behind it was something about culture itself, and about a structural condition of musical life so constant that I had never once thought to notice it.

To get to it, I have to go back through the instruments.

II. The long stability

For most of human history, musical instruments changed slowly in their fundamental logic.

You blew air through something. You struck a stretched skin. You plucked, bowed or hammered a string under tension. Wood, metal, gut, skin, air. The variations were endless and the principles were few. A violin remained recognizably a violin across many generations. An oud remained an oud. A guitar remained six strings over a resonating body.

And yet the music kept changing enormously.

Listening waypoint: Paco de Lucía, “Entre Dos Aguas.” The flamenco guitar, still rooted in an old physical instrument, already stretching the language around it.

Flamenco players kept finding more inside essentially the same instrument: more rhythm, more silence, more violence, more precision. Romani musicians did extraordinary things with violin, cimbalom and double bass. Tango found an entire emotional grammar in bandoneón, violin, piano and bass. Arab music carried an immense expressive world — maqam, ornament, microtonal shading — through oud, qanun, ney and percussion.

The instrument stayed relatively stable. The accumulated intelligence went somewhere else: into the human being. Into technique, phrasing, timing, memory, interpretation, lineage. A tradition was a very long conversation about what could still be extracted from a fixed set of objects.

This is worth holding onto, because it is the baseline against which everything after it should be measured. For most of the history of music, the machine was simple and the person was the sophisticated part.

III. Electricity carries the sound

Then electricity arrived, and at first it did something modest. It carried sound farther.

A microphone does not fundamentally change a flamenco guitar. It allows a larger room to hear it. A recording does not change what the musician played; it detaches the performance from the moment and lets it travel. These were enormous social changes — they created the record industry, the star, the global hit, the possibility of hearing a dead musician — but they were not changes to the instrument. The decisions still happened where they had always happened.

Although one thing did shift, quietly, and it matters later: amplification changed how people sang. Once a voice could be carried, it no longer had to be projected. Crooning became possible. Intimacy became a public sound. Even the modest phase of electricity was already reshaping technique.

IV. Electricity enters the instrument

Then electricity went inside.

The electric guitar was not simply a louder guitar. Pickup, amplifier, valve saturation, feedback and effects became part of its voice. Distortion was not a fault to be minimized; it was material. Feedback was not noise; it was a note you could hold and bend.

Listening waypoint: Jimi Hendrix, “Machine Gun,” Band of Gypsys, Fillmore East, 1970. Feedback used as an instrument rather than tolerated as a fault.

Then the synthesizer, which removed the requirement that there be an acoustic event at the beginning at all. The musician no longer selected an instrument; the musician designed the sound. Timbre stopped being inherited and became authored.

Listening waypoint: Jean-Michel Jarre, “Oxygène, Pt. 4,” 1976. The sound world itself is electronic.

Then the studio itself became an instrument. Dub is the clearest case anywhere in music. The band finishes playing and the record is not finished — the producer keeps composing, at the mixing desk, out of echo, subtraction, reverb, bass and silence. The most important compositional act becomes removal: deciding what disappears and when it comes back.

Listening waypoint: Jacob Miller, “Baby I Love You So,” then King Tubby and Augustus Pablo, “King Tubby Meets Rockers Uptown.” The same recording, twice. Listen for what is taken away.

Then disco and its engineering, where the record was built as an object rather than captured as a performance.

Listening waypoint: Cerrone, “Supernature,” 1977. A record built as an object rather than captured as a performance.

Then software absorbed the physical instrument entirely. One laptop now contains drums, pianos, orchestras, synthesizers, and instruments that never existed physically at all. Techno, house, hip-hop and everything downstream of them were made largely by people who were not “playing” in the older sense.

Each of these was described, at the time, as the end of something. Each was, in fact, a migration: the intelligence moved from the fingers to the signal chain, from the hand to the desk, from the room to the machine.

But look at what survived all of it.

Somebody was still deciding.

Hendrix was still Hendrix. Jarre still chose the sounds. The dub engineer still decided when the voice vanished and when the bass returned. The techno producer, playing nothing, was still saying: this sound, here, now; repeat it; take it away; bring it back.

Every revolution up to this point changed the tool. None of them touched the decision.

V. The boundary

This is the boundary I think AI crosses, and why it does not sit comfortably in the sequence above as merely the next step.

AI does not enter the instrument. It enters the decisions that made the musician a musician.

Melody. Harmony. Rhythm. Arrangement. Instrumentation. Timbre. Production. Style. And — this is the part that produced my disorientation — the performer. The voice. The particular way a specific human being arrives slightly late on a line.

The synthesizer let you design a sound that nobody had heard. It did not design for you. The distinction sounds small and is not. A synthesizer expands the space of possible decisions; a generative model occupies it. What is automated now is not the production of sound but the production of choices, and choice was the last thing left that was unambiguously ours.

I want to resist overstating this. A person still prompts, still selects, still rejects, still curates. There is a real argument — and I will come back to it — that prompting is simply the next migration of intent, no different in kind from a dub engineer who plays nothing. I am not certain that argument is wrong.

But my interest turned elsewhere. Because when I asked what I had actually been listening to, the answer that came back was not about authorship at all.

At first, I thought it was about scarcity.

VI. What music is actually for

Music has never only been entertainment, and treating it as a product with quality attributes is what makes most of this debate miss.

Music stores culture.

It carries language, poetry, geography, religion, class, migration, labor, mourning and celebration. It is one of the few technologies humans have for moving an emotional structure across a century without writing anything down.

You can often hear ten seconds and know where you are. A particular oud tone and a particular ornament and you are in an Arab musical world. A fiddle, a guitar and a certain vocal placement and you are somewhere in Texas. A flamenco guitar puts you inside a history of specific people in specific towns. A Russian melodic contour is recognizable long before anyone says the word Russia. None of this information is in the lyrics. It is in the sound, and it got there because thousands of people made similar decisions over a long time in the same place.

And music does something stranger still. It creates collective memory in real time.

Put Piano Man on in the right bar and people who have never met begin singing the same words at the same moment. Fifty unrelated lives occupy the same piece of memory for four minutes. Nobody organized it. At a wedding, the first two bars of one song can make a room react before the chorus arrives. Parents play music in a car; children complain; twenty years later the children hear it and the car comes back, and the parents, and the specific quality of light on a specific road.

A song, when enough people know it, becomes a public container into which millions of people deposit private memories. That container is not made of sound. It is made of the fact that other people have it too.

VII. The floor nobody noticed

Here is the thing I had never thought about.

For all of recorded history, music was hard and slow to make.

A musician had to spend years learning. A composer had to write. A band had to rehearse. Recording required a room, equipment, engineers, money and time. Distribution required physical objects or licensed channels. Even in the era of the laptop, when the cost fell dramatically, a song still cost a human being something: attention, hours, a decision to finish.

So the number of musical objects entering a society was finite. And because it was finite, millions of us encountered the same ones. Repeatedly. Over years. Long enough for the objects to accumulate history, and for the history to become shared.

I had always assumed that “the artist,” “the song,” “the album,” “the hit,” “our song,” “the music of that summer,” “our generation’s music” were natural categories — features of what music simply is.

I now suspect some of them are consequences of production scarcity.

Not the music itself. The commonality. The floor under musical culture was never a value or a tradition. It was a constraint. There were only so many songs, so we shared them, and sharing them is what turned them into culture rather than audio.

At first, that looked like the floor.

VIII. Before AI, the common song was already thinning

There is a complication here, because AI did not invent this direction of travel.

The common song was already becoming less common.

For a long time, musical life contained a great deal of forced overlap. The radio played something and everybody in the car heard it. A record was put on in a house and the children heard what the parents had chosen. A jukebox belonged to the room, not to one person. Television put the same performance in millions of homes. A handful of stations, shops, labels and charts determined far too much of what could reach us — and that concentration had obvious costs — but it also produced something we did not know we were receiving.

Overlap.

You heard music you had not selected. You heard it repeatedly.

You heard what your parents heard, what your neighbors heard, what the person beside you at school had heard. Not everybody, not everywhere, and never as completely as nostalgia likes to remember it. But enough people encountered enough of the same objects that a repertoire could become generational.

Then the listening device moved closer to the individual. The family stereo became the personal player. The radio became the playlist. The playlist became the recommendation. The speaker became the headphones.

A family can now occupy the same car while inhabiting several musical worlds.

And something else accelerated at the same time: replacement. The next song is always available. Then the next. Then another playlist. Another recommendation. Another release. The cost of moving on has approached zero.

That matters because culture is not made by exposure alone.

It is made by recurrence.

A song becomes part of a life because it survives long enough to meet that life repeatedly. You hear it before you meet somebody. You hear it with them. You hear it after they are gone. The recording has not changed, but you have, and each return deposits another layer inside the same object.

Perhaps that is why the music of parents can feel so strange to their children now, and why even people of the same age increasingly cannot assume that they know the same songs. It is not simply that taste has diversified. Musical time itself has accelerated. The object has less time to become old enough to become ours.

Generative AI does not begin this process. It removes its remaining brake.

Streaming gave us an effectively unlimited shelf of existing music. Generative systems can give us an unlimited supply of music that did not exist a moment ago.

At that point the question is no longer only whether we will hear the same songs.

It is whether any song will be given enough time to become the same song twice.

IX. The listener becomes the prompt

Streaming already personalizes, but it personalizes selection. Spotify knows me and recommends an existing song that another human being made for reasons of their own. However narrow the algorithmic corridor gets, I am still being handed objects that exist independently of me and that other people are also being handed.

Generative systems personalize creation.

The question stops being which song should I listen to and becomes something I can simply describe:

I am nostalgic tonight, but I do not want to be sad. Something with an old Russian atmosphere, a little tango phrasing, an oud somewhere in the distance, no vocals, eighty-two beats per minute, twenty minutes long.

And the system says: here.

There is another transformation hidden inside that prompt.

Russian. Tango. Oud.

Things that once arrived attached to places, communities and histories can now arrive as parameters. A musical inheritance becomes a texture I can request.

That does not erase the tradition it came from, and borrowing between musical cultures is hardly new. Music has travelled for as long as humans have. But there is a difference between encountering another tradition and asking a machine to extract its recognizable properties for my mood tonight.

Provenance can become optional. The sound of a place can remain after the place has disappeared from the encounter.

The song did not exist before I asked. It may never be heard by another person. Tomorrow my wife describes a different evening and receives a different perfect song. You receive another.

Extend that honestly, without cynicism, and you arrive at a genuinely strange destination. Everyone may have extraordinary music — better matched to their mood, memory, nervous system and moment than anything a stranger in another decade could have made for them.

But we may no longer have the same music.

Perfect music, and no common song.

X. The room

The sharpest test is still the dance floor. But I think I understand better now what dancing together actually means.

It does not merely mean that several hundred bodies happen to be moving in the same room.

They are moving inside the same time.

A beat arrives and everyone hears it. A break opens and everyone enters the same absence. The return has not happened yet, but the room already knows something is coming. Then it comes — for everyone at once.

For a few minutes, hundreds of nervous systems have accepted the same clock. That may be one of music’s oldest social functions.

Techno makes the paradox especially visible because it is among the most machine-mediated music we have produced. The sound may originate in circuits and software; the pulse may be more exact than any human drummer could naturally sustain. Yet in the club the result is profoundly collective. The machine does not isolate the listeners. It synchronizes them.

Now take personalization to its logical end. A thousand people in a room, each receiving a perfectly optimized stream generated live for their physiology, memories and mood.

A thousand perfect songs.

What does together mean now?

They could still look at one another. Copy movements. Respond to the same lights. Perhaps the systems themselves could communicate, matching tempos so that everyone appears to occupy the same dance.

But something has changed. There is no longer one musical event happening to the room.

The person beside me moves because something happened in his song. I move because something different happened in mine.

We are adjacent.

Whether we are dancing together is no longer obvious.

But there is another possibility, and it may be more likely. The AI does not generate for each person.

It generates for the room.

It watches the floor. It feels, through sensors and behavior, when energy is rising and when it is falling. It changes the rhythm, extends the build, delays the drop, shifts the bass, finds whatever this particular collection of people responds to tonight.

Now everyone is dancing together again. Perhaps more successfully than ever.

But the song has never existed before this moment. And when the night ends, perhaps it disappears.

That gives us something new:

a shared present without a shared past.

Everyone in the room experienced the same music. Nobody outside it did.

There may be nothing to replay at the wedding ten years later. Nothing to put on in a bar and discover that a stranger knows too. Nothing a child can find in a parent’s collection and ask: What was this?

The collective experience survives. The cultural object may not.

And that distinction may matter more than whether the composer was human.

XI. What actually has to survive

This changes the question. Perhaps the danger is not AI-generated music at all.

Suppose a machine generates a song tomorrow. It is beautiful. Somebody gives it a name. Millions of people hear it. They play it again. Couples attach themselves to it. It appears at weddings. Children hear their parents playing it. Twenty years pass and the opening bars suddenly return an entire period of life.

At that point, what exactly has failed?

The composer was artificial. The culture was not.

The song acquired history because human beings gave it recurrence.

So perhaps scarcity was never the deepest floor.

Scarcity created something else accidentally:

repetition.

There were only so many songs available, and therefore we heard some of them again and again. And repetition produced memory.

Which means the cultural danger of generative abundance is not that machines can make songs. It is that the machine never needs to stop making the next one.

The next song can always fit me slightly better. The next one can be newer. The next one can respond to who I am tonight rather than who I was yesterday.

Nothing needs to fail for the previous song to disappear. It is simply replaced by something marginally more appropriate.

A civilization accustomed to infinite generation may therefore discover that the scarce thing is not music.

It is the decision to keep something.

To hear it again.

To let it become imperfectly familiar.

To allow children to become tired of it in the back seat.

To play it at several weddings.

To hear it badly sung in a bar.

To give it enough years to become larger than the moment for which it was made.

The machine can produce another song immediately.

Culture may begin when we decide not to ask for one.

XII. The strongest objections

I do not want this to become an anti-AI argument, partly because I do not believe it, and partly because the anti version is boring. If AI music is bad, nothing here matters; nobody’s culture is threatened by something nobody wants to hear.

The difficult future is the one where it is beautiful. Where the machine really does understand what moves us. Where it makes something better suited to my Tuesday evening than the artist I was going to play. Where I never again have to tolerate anyone else’s musical choice, because mine can be manufactured on demand.

So here are the objections I think are strongest, including one that may be fatal to my central claim.

1. Commonality may be a want, not a by-product.

This is the serious one. My argument began by assuming shared songs were produced by scarcity. I am less certain of that now.

People may not sing together because there were few songs. There may have been few songs, and people used them because singing together was already something they wanted to do.

If synchronization is a human want rather than an accidental residue of scarcity, abundance does not destroy it. It only means we have to choose what to share.

Football chants survive when millions of other songs are available. Families repeat particular songs because they belong to the family. Weddings deliberately return to old music. People travel across countries to stand together while one artist plays something every person in the room already knows.

Perhaps the common song survives because humans keep making common songs on purpose.

But that introduces another question: what used to happen automatically may increasingly require intention. Radio, television, the family stereo, the car radio, the chart, the album and the nightclub all created overlap without asking us to design it. Personalization removes more and more of those defaults.

Culture may survive abundance perfectly well. But it may have to become something we consciously maintain.

2. Generational fragmentation was already happening.

This is important because otherwise AI gets blamed for a change it merely accelerates. A child today can inhabit a musical world almost entirely separate from the parents sitting beside them. Two people born in the same year can receive radically different recommendations and never encounter one another’s repertoire. The category music of our generation is already less secure than it once appeared.

Generative AI does not create that fragmentation. It changes its possible scale.

Recommendation can separate us while still drawing from the same library. Generation can give each of us a different library.

3. Scarcity moves rather than vanishes.

Very likely true. If music is infinite, then what becomes scarce is everything the music cannot supply instantly: provenance, recurrence, reputation, authenticity, live presence, the knowledge of who made it and why, and perhaps simply the assurance that somebody else knows it too.

4. Live music may become more valuable, not less.

Almost certainly. Abundance of the reproducible has repeatedly increased the value of the unrepeatable. If every soundtrack can be generated, being in a room where something happens once — and happens to everyone there — may become a premium cultural good rather than a nostalgic one.

5. AI can make common songs.

Of course it can. And this may be the most important correction to the whole argument.

Nothing about an artificial origin prevents a piece of music from becoming culturally human afterward. If we keep it. If we replay it. If enough lives gather around it.

The question is therefore not whether AI music can become culture. It is whether a system optimized to generate endlessly gives anything enough time to do so.

6. We may choose recurrence precisely because it becomes scarce.

The disappearance of an automatic condition can make its value visible. Perhaps families will become more deliberate about shared music. Perhaps clubs will preserve particular sets. Perhaps artists — human or otherwise — will matter because their names provide continuity. Perhaps the cultural institutions of the future will be less concerned with discovering new music than with deciding which music deserves to remain.

That would be an unexpected reversal. In an age capable of producing infinite novelty, curation may become an act of memory.

XIII. What I still think survives

After all of that, this is the residue I cannot argue away.

Every previous revolution changed how music was made and left largely intact a chain that made music cultural:

a limited number of objects,

encountered by many people,

repeated often enough,

for long enough,

to accumulate shared history.

Amplification did not break that chain. The synthesizer did not. Dub did not.

The laptop weakened parts of it, and streaming weakened others, because listening became more individual and the catalogue effectively infinite.

Generative personalization reaches the final link. It can make recurrence unnecessary.

Not certainly. Not inevitably. Perhaps not predominantly.

But for the first time, the technically easiest response to I want to hear something may be not to retrieve something that already exists, but to make something else.

And the thing about a floor you have never noticed is that you do not miss it while it is being removed.

You notice later, in the specific places that depended on it:

a bar where fewer people know what comes next;

a wedding where the room does not react to the first two bars;

parents and children who lived in the same house but inherited different musical pasts;

a generation whose members have exquisite, highly developed taste and remarkably little repertoire in common;

a couple who spent thirty years together and discover that the machines scored those years separately.

None of this arrives as catastrophe. Nobody announces the end of anything.

There is simply a gradual reduction in the number of memories that belong to more than one person.

XIV. The bar

So let me end where the argument started, which is not with technology at all.

A bar, late.

Somebody puts on a song everybody happens to know. It is not a particularly good song. It has been played too many times. Some people in the room would never choose it if they were alone.

That may even be part of why the moment works.

The first notes arrive. Someone recognizes them. Then another person does.

By the chorus, a room of strangers is singing words they did not write, about a life none of them lived, in unison, badly.

Nothing in that scene is about audio quality. Nothing in it would necessarily be improved by a better song. It is not even entirely about music.

It is about the fact that the song arrived in all of those lives separately, over years, and left the same shape in each of them. And now, for four minutes, that shape is visible.

Held in common.

The machine can make a better song. It can make one more appropriate to every person in the room. It can make another three seconds later.

What it cannot manufacture instantly is the twenty years during which fifty strangers came to know the same one.

That time has to be lived.

That is the thing I want to know the fate of.

Not whether machines can make music. They can, and some of it will be beautiful, and that is precisely what makes the question hard.

Karim Spring

—