The Symphony of Progress

In 1982, a branch of Britain’s Musicians’ Union called for an outright ban on synthesizers.

The immediate concern was not philosophical. It was practical. Barry Manilow had toured the United Kingdom using a bank of synthesizers in place of an orchestra, and musicians quite reasonably saw jobs disappearing. The union’s own history now recounts the episode as one of many moments when a new musical technology set off alarms about whether a machine was replacing a musician.¹

The ban never became lasting policy. Synthesizers did not destroy music. They became music.

That story is tempting to use as an easy analogy for artificial intelligence: See? People were afraid of synthesizers too, and everything turned out fine.

But AI deserves a better argument than that.

A synthesizer usually does not compose an entire song, choose its arrangement, write lyrics, invent a vocal performance, mix the track, and hand you a finished recording in seconds. Modern generative systems can automate an unprecedented amount of musical production. The disruption is real. Musicians are not irrational for worrying about it.

Still, the synthesizer story exposes something important: the line between “real music” and technological mediation has always moved.

And it keeps moving because music has never been defined by the purity of the tool.

It has been defined by what people do with tools.

The moving boundary of “real music”

Every generation inherits musical practices that feel natural because somebody else already fought about them.

Recorded music itself once severed sound from the physical presence of a performer. Electric amplification changed the relationship between the hand, the instrument, the room, and the audience. Drum machines replaced some kinds of live percussion. Synthesizers created sounds no acoustic instrument could make. Samplers turned existing recordings into raw material. Sequencers automated timing. MIDI separated musical gesture from sound generation. Digital audio workstations made cutting, copying, moving, quantizing, stretching, correcting, and layering sound routine.

None of this is obscure technology anymore. It is simply how music is made.

The strange part is that yesterday’s violation of authenticity has a habit of becoming tomorrow’s tradition.

The Musicians’ Union itself records earlier disputes over whether skiffle players were “real musicians,” whether Hammond-organ players deserved membership, and whether synthesizers should be allowed at all.¹ In the early 1980s, the recording industry also warned consumers that home taping threatened music. Scholars looking back at those campaigns have noted how new media technologies repeatedly trigger conflicts over labor, ownership, production, and even who gets to be recognized as creative.²

This does not mean every panic about technology is foolish. New tools really do eliminate some jobs, create others, redistribute power, and change what skills are valuable. The correct lesson is narrower:

If that were the standard, most recorded music made in the last half-century would already be disqualified.

The music we already call human is engineered

Consider what happens inside a modern studio.

A vocal performance may be assembled from multiple takes. Timing can be adjusted. Pitch can be corrected. Drums can be replaced or reinforced with samples. Guitars may be re-amped. Entire orchestral arrangements can come from software instruments. A songwriter may bring in a melody, another person may rewrite the chorus, a producer may change the harmony, a beatmaker may construct the rhythm, and an engineer may radically alter the sonic character of the final recording.

That does not make the song fake.

It makes the studio part of the instrument.

This is where some criticism of AI music becomes inconsistent. We generally do not demand that a conventional recording prove its humanity by presenting an untouched acoustic performance. We accept enormous technological mediation because we have learned to recognize the decisions behind it as part of musicianship.

The point is not that an AI generator and a digital audio workstation are the same thing. They are not.

The point is that technological mediation cannot be the test, because we abandoned that test decades ago.

So we need a better one.

Where does creativity actually live?

We tend to imagine creativity as production: the hand strikes the drum, the fingers play the chord, the singer moves air through a vocal tract.

Those acts can be profoundly creative. But creativity has never lived only in physical execution.

A producer deciding that a song needs silence before the chorus is making a creative choice. A filmmaker choosing one take out of fifty is making a creative choice. A photographer deciding where to point the camera is making a creative choice. A DJ selecting, sequencing, and transforming recordings is making creative choices. A composer may create music without personally performing any of it.

Creative agency includes intention, selection, rejection, arrangement, revision, transformation, context, and taste.

That distinction matters enormously with AI.

There is an obvious difference between typing a generic prompt, accepting the first result, and declaring yourself the musical genius behind it—and spending hours or days directing generations, rewriting lyrics, changing structures, extracting parts, editing performances, replacing sections, manipulating audio, combining outputs, adding human performances, and deciding what survives.

Those are not the same creative process.

AI music therefore makes more sense as a spectrum of agency than a binary category.

At one end, AI can function almost like a vending machine: request song, receive song. The human claim to detailed authorship may be weak.

Further along the spectrum, AI can act as a collaborator: proposing melodies, textures, arrangements, voices, or variations that a person evaluates and redirects.

Further still, it can act as one component inside a larger production process controlled by a human artist.

This is not merely philosophical hair-splitting. The U.S. Copyright Office reached a related conclusion in its 2025 report on AI-generated works. It said that using AI as an assistive tool does not bar copyright protection, while protection depends on sufficient human-authored expressive elements; merely providing prompts is not automatically enough.³

Copyright law and artistic judgment are not identical. But the principle is useful:

The uncomfortable industry argument

This is where the debate gets more politically interesting.

The recording industry is correct about something essential: artists deserve meaningful control over their work, and companies should not be allowed to treat every song ever recorded as a cost-free resource simply because machine learning requires enormous datasets.

But the entertainment industry also has a complicated history when it presents itself as the singular guardian of creativity.

Copyright is meant to encourage creative production by giving creators enforceable rights. That protection matters. Without it, powerful businesses could simply copy the work of less powerful artists and monetize it.

Yet copyright can also constrain new creation when permission becomes expensive, uncertain, or impossible to obtain.

Sampling is the clearest musical example.

In Bridgeport Music v. Dimension Films, the Sixth Circuit adopted an exceptionally strict approach to digital sampling of sound recordings—summed up by the famous instruction to obtain a license or refrain from sampling.⁴ Legal scholars and music researchers have argued that the resulting licensing environment changed the aesthetics of hip-hop itself by making dense sample collage dramatically harder and more expensive to release commercially. The book Creative License, based on extensive interviews with musicians and industry figures, argues that the licensing system has constrained creative choices and estimates that landmark sample-heavy albums such as Paul’s Boutique or Fear of a Black Planet would be extraordinarily difficult to clear under modern licensing economics.⁵

At the same time, American copyright law contains another tradition. In Campbell v. Acuff-Rose, the Supreme Court emphasized that transformative uses can further copyright’s underlying purpose by creating new expression, meaning, or message.⁶

Those two forces—protection and transformation—have always existed in tension.

That is why I am skeptical whenever the AI debate collapses into a simple story where established rights-holders represent “human creativity” and new technology represents its destruction.

Sometimes the fight is genuinely about protecting creators.

Sometimes it is also about controlling the terms under which culture may be reused.

Usually, it is both.

AI is not “just another synthesizer”

A serious defense of AI music has to admit where the analogy to older technologies breaks.

Generative AI can collapse stages of creation that previously required multiple people and years of accumulated skill. A user can describe a musical idea in ordinary language and receive a convincing result without understanding harmony, orchestration, microphone technique, synthesis, mixing, or performance.

That is qualitatively different from buying a new keyboard.

It raises uncomfortable questions.

If someone delegates virtually every meaningful musical decision to a model, how much personal authorship remains?

If models are trained on musicians who never consented to that use, what obligations do developers have to them?

If a system can deliberately imitate the recognizable voice of a living singer, does that cross a line even if the resulting song is technically new?

If millions of nearly effortless songs can be generated every day, what happens to the already brutal economics of attention?

These are not hypothetical concerns.

In 2024, major record labels sued Suno and Udio, alleging that the companies had copied copyrighted sound recordings without permission to train their systems.⁷ The U.S. Copyright Office’s 2025 report on AI training rejected simplistic answers in either direction: it concluded that some forms of AI training may qualify as fair use and others may not, depending on the works, sources, purposes, outputs, and effects on existing markets. It specifically warned that commercially exploiting vast collections of copyrighted expressive works to produce competing content—particularly when the works were obtained unlawfully—can go beyond established fair-use boundaries.⁸

And abundance itself is becoming a real problem. In 2026, Deezer reported that AI-generated tracks had at times exceeded half of all new music uploaded to its service. The company also said consumption remained relatively low and that a large share of streams involving fully AI-generated music showed signs of fraud.⁹

So yes: AI can produce spam.

It can enable fraud.

It can imitate people in unethical ways.

It can be used by corporations to replace labor rather than empower creators.

None of those facts answer the artistic question.

They answer different questions—important ones—about markets, consent, ownership, identity, and power.

Ethics are not ontology

This distinction is crucial.

A creative process can produce art while operating inside an unfair economic system.

The reverse is also true: something can be perfectly legal and still be creatively worthless.

If a musician makes an extraordinary AI-assisted piece using a tool whose training practices later turn out to have been unethical, that training controversy does not magically erase the musician’s creative decisions. It tells us that the infrastructure needs better rules.

Likewise, recognizing that AI can be used creatively does not excuse an AI company from compensating creators when compensation is legally or ethically warranted.

We should be able to hold both thoughts at once:

AI can be a legitimate creative instrument.

AI companies can still owe artists better consent, transparency, licensing, attribution, or compensation.

That is not a contradiction. It is what maturity looks like.

The Copyright Office’s approach is instructive here precisely because it does not treat “AI” as one legal use. It analyzes different uses separately and acknowledges that outcomes depend on context.⁸

The music industry itself is beginning to move in the same direction. By September 2026, Suno was releasing new models through partnerships with Warner Music Group and BMG built around licensed material and opt-in participation for artists.¹⁰ Whatever one thinks of the specific deals, the development demonstrates something important: even the conflict between generative music companies and established rights-holders is evolving from an argument about whether AI music should exist into an argument about the rules under which it should exist.

That is progress.

Access is also part of creativity

There is another side to AI music that gets surprisingly little respect.

Not everyone with musical imagination has the technical vocabulary required to realize it.

A person may hear arrangements internally but never have learned an instrument. Someone may write extraordinary lyrics but have no access to musicians. A disabled creator may face physical barriers to traditional performance. A teenager may understand exactly what atmosphere a song should create without owning anything more sophisticated than a phone. An experienced musician may use AI simply to audition twenty arrangements before recording one conventionally.

Historically, new tools have repeatedly reduced the cost of translating ideas into sound.

Affordable multitrack recorders did it.

Samplers did it.

Home computers did it.

DAWs did it.

Software synthesizers did it.

AI pushes that democratization dramatically further.

There is a legitimate concern that making creation effortless can devalue craft. Learning an instrument develops forms of discipline, embodied knowledge, collaboration, and musical intuition that typing prompts will not replicate.

But expanding one route into musical expression does not require destroying the others.

Photography did not make painting meaningless.

Drum machines did not abolish drummers.

Electronic music did not erase acoustic music.

Recorded music did not eliminate concerts.

New forms tend to rearrange culture rather than cleanly replace what came before.

The question should therefore not be whether AI lets somebody “skip” skills that another artist spent decades learning.

The question is whether the person used whatever skills they do possess—taste, language, structure, storytelling, editing, listening, emotional judgment—to make something intentional.

If the answer is yes, dismissing the result solely because the instrument involved artificial intelligence starts to look less like a defense of creativity and more like a defense of credentialing.

The flood is real. So is the filter.

Critics are right that generative abundance changes the economics of attention.

When producing a competent-sounding track takes seconds, the world can be flooded with competent-sounding tracks.

That might make human curation more valuable, not less.

The future value of a creator may increasingly lie not in proving that every note was generated through manual labor, but in demonstrating a recognizable sensibility: this person consistently chooses interesting things.

Taste becomes visible.

Direction becomes visible.

Context becomes visible.

Identity becomes visible.

This is not entirely new. Music fans have always cared about more than waveform originality. We care about scenes, personalities, histories, performances, communities, visual language, politics, record collections, liner notes, influences, live experiences, and the strange emotional relationship we form with the people behind sound.

AI does not automatically supply those things.

Humans do.

That is one reason the claim that AI will simply “replace artists” feels incomplete. It may replace some kinds of musical labor. It may commoditize some kinds of production. It may fill some functional markets—background cues, stock music, anonymous mood playlists—with synthetic output.

But an artist has never been merely a machine for emitting audio files.

An artist is also a point of view.