WHEN THE MACHINE ENTERS THE STUDIO AI, Computer Music, Human Labour, Lost Recordings and the Price of Making a Song Where Does the Tool End, Where Does the Creator Begin — and What Happens When the Machine Becomes Cheaper Than the Musician? By Gordon Lawrence Taylor — Voice In The Void Infectious Unease Radio / Subterranean Zone Radio Research current to August 2026
AI, Computer Music, Human Labour, Lost Recordings and the Price of Making a Song
Where Does the Tool End, Where Does the Creator Begin — and What Happens When the Machine Becomes Cheaper Than the Musician?
By Gordon Lawrence Taylor — Voice In The Void Infectious Unease Radio / Subterranean Zone Radio Research current to August 2026
Artificial intelligence has entered music from almost every direction at once. It can clean a damaged recording, separate a voice from a decades-old cassette, suggest a mix, generate an instrument, manufacture a singer, write a melody or produce an almost complete track from a few sentences. It can help an independent musician who cannot afford a studio, while at the same time potentially removing paid work from musicians, producers, engineers and composers. It can preserve something deeply human, or create a performance that never happened. After more than three decades working around independent and underground music, I do not think the useful question is simply whether AI is good or bad, or whether using it amounts to cheating. The far more interesting questions are what we are asking the machine to do, what remains human, who is being credited and paid, what is being preserved or replaced — and whether the listener knows the difference.
Suppose You Have $1,000 and a Song
Imagine an independent musician. They have written a song and perhaps A$1,000 available to turn that idea into a finished professional recording. What happens next?
A current Melbourne studio advertises a full eight-hour recording day for A$750, including an experienced engineer. Full-production mixing is listed at A$450 per song, while analogue mastering begins at A$80 per song. Before additional musicians, production, editing, travel, revisions or equipment hire are even considered, one professionally produced song can move beyond A$1,000 very quickly.
There is nothing unreasonable about those prices. They represent a professionally treated acoustic environment, microphones, preamps, monitors, instruments, maintenance, electricity, years of technical training and something much harder to put on an invoice: human judgement. The producer listens. The engineer listens. The musician performs. They change things, disagree, make mistakes and sometimes discover that the mistake is better than the original idea.
Now place another possibility beside it. As of August 2026, Suno advertises subscription plans offering thousands of monthly generation credits, with its higher tiers providing access to Suno Studio and the capacity to create extremely large quantities of material for a fraction of the cost of a traditional recording session.
These are not equivalent products. A generative-AI subscription does not place microphones around a real drum kit. It does not hear an undesirable resonance in the room and move the microphone. It does not watch a nervous vocalist and realise that what they need is reassurance rather than another technical instruction. It does not necessarily recognise that the technically imperfect third take contains something emotionally extraordinary.
But economics does not ask only whether two things are identical. Economics also asks: is the cheaper version good enough for the job?
That may prove to be one of the most important questions in the entire argument surrounding artificial intelligence and music. AI does not necessarily have to become a better musician than the greatest human composer, producer or engineer. For certain jobs it may only need to become fast enough, cheap enough, convincing enough — and good enough.
The question is therefore no longer merely, Can a machine make music? Clearly it can. The more difficult question is: what happens to the people who make music when parts of their labour can be generated for the price of software?
Before answering that, however, we first need to answer another increasingly difficult question. How do we even know when AI has been involved?
When Listening Was Not Enough
I was recently sent a piece of music. I am deliberately not identifying the artist, the title of the recording or any information that could identify the person involved. That is not the purpose of this case study. The purpose is the process of discovery.
When I first heard the track, Ankit of LazyboyProactive, a very good friend and somebody whose knowledge of music, production, electronics and musical equipment I respect enormously, suggested that artificial intelligence might have been involved. LazyboyProactive itself has a long electronic-music history, tracing its development back to 2006 and working across progressive electronics, electro, synth-pop, rock, analogue synthesisers, guitars, vocals and drums.
Ankit’s observation made me listen again, but listening was not proof. My ears could tell me whether the track worked. They could tell me about atmosphere, bass, arrangement, rhythm, sound design and dynamics. They could perhaps make me suspicious. But an ear is not a forensic instrument.
Electronic music has spent more than half a century producing sounds that appear mechanical, synthetic, impossible or inhuman. A programmed rhythm does not prove AI. A synthetic voice does not prove AI. A strange guitar does not prove AI. A perfect edit does not prove AI. An extremely polished mix certainly does not prove AI.
So rather than publicly saying, This sounds like AI, I wanted to know whether the supplied digital material contained anything more concrete. It did.
Looking Inside the File
Apart from listening, the investigation involved digital-file and metadata inspection. An audio file can contain considerably more than the sound that reaches our ears. Depending upon how it has been created, edited and exported, it may retain technical traces including creation dates, software references, encoding information, XMP metadata, export information, file names, stem designations and other pieces of production history.
The material examined in this case included separate WAV stems corresponding to elements such as drums, bass, guitar and synthesiser. Embedded within several of those files was Adobe XMP metadata. Within that metadata was a very direct phrase: “made with suno studio.”
The same reference appeared across four of the supplied stems. Creation information associated with the material dated from 1 May 2026. That was considerably more meaningful than simply saying, I think I can hear AI. It provided technical evidence that Suno Studio had been involved somewhere in the production chain of those files.
But then the case became more interesting. The material also contained evidence of conventional production and editing software, including references to FL Studio 11 and Adobe Premiere Pro 2026 on Macintosh.
That distinction is crucial. Finding FL Studio does not make a recording AI-generated. A person can compose every note, place every drum hit, record every instrument, manipulate every sample, automate every effect and construct an entire composition manually inside a digital audio workstation. Likewise, finding Adobe Premiere tells us something about an editing or export process. It does not mean Adobe composed the music.
The discovery therefore was not simply that a computer had been used. Of course a computer had been used. The interesting finding was that generative-AI technology and conventional digital-production software appeared within the same workflow.
That is a far more realistic picture of where music production may be heading. A generated stem can be exported into a DAW, chopped, layered, resampled, distorted, reversed, time-stretched, played through hardware, combined with a real vocalist, mixed by a human engineer and mastered by another human. By the time the record is finished, the relationship between human and machine may be extremely complicated.
The metadata did not prove that Suno created every musical element, nor did it establish what proportion of the finished artistic decisions belonged to the human creator. What it established much more narrowly was that Suno Studio had participated somewhere in the creation or processing history of several supplied stems.
That is exactly why the binary question Is this AI? is becoming inadequate. A much more useful question is: What did the AI actually do?
Metadata Is Evidence — But It Is Not Magic
Metadata can be extremely revealing, but it is not an infallible truth machine. It can be stripped, overwritten or lost through conversion. Audio can pass through several applications. A generated element can eventually be altered beyond recognition, while a completely human performance can pass through a machine-learning tool without that tool composing a single note.
This is why evidence has to be interpreted rather than merely found. But the case illustrates something important: the ears raised the question; the files provided evidence beyond the ears.
That difference matters, particularly because broadcasters, labels and listeners are going to encounter this problem more frequently. Governments are also beginning to treat metadata and technical labelling as part of the provenance problem. China’s rules for AI-generated synthetic content include both explicit labels and implicit technical information embedded in files, while Europe is moving toward machine-readable identification of certain AI-generated or manipulated content.
What happened inside one anonymous music file therefore points toward a much larger future. The file itself may increasingly tell part of the story of how the music was made.
Hundreds of Recordings Can Arrive in a Week
For me, this is not simply an academic problem. Through Infectious Unease Radio and Subterranean Zone Radio, I receive music continuously from individual musicians, bands, producers, promotional companies and independent record labels. Across a week, that can mean hundreds of tracks and associated submissions passing through my hands.
Industrial, gothic, darkwave, post-punk, shoegaze, experimental electronics, ambient, dub, roots, future garage, post-dubstep, jungle, drum & bass, breakbeat and music that does not fit neatly into any established category all arrive for consideration.
For decades there was a basic working assumption when a recording arrived: the credited artist had substantially created the work. That assumption can no longer automatically be made.
Some music arriving today may have used machine learning only to remove noise. Another record may have been AI-assisted during mastering. Another may contain AI-assisted stem separation. Another may contain one generated atmospheric texture. Another may contain generated lyrics. Another may contain an artificial backing singer. Another may contain a synthetic lead voice. Another may be almost completely prompt-generated.
Yet every one can arrive with exactly the same description: new music submission.
That creates a new problem of provenance.
Imagine Reviewing a Performance That Never Happened
Suppose I receive a recording containing what appears to be an extraordinary vocal performance and write that the vocalist’s emotional restraint during the final verse is one of the strongest elements of the recording. But suppose there was no vocalist. The voice was generated.
Or I praise the drummer’s extraordinary ghost-note work, but there was no drummer. Perhaps there was not even a person manually programming those drums.
Now reverse the situation. Perhaps the drummer was entirely real and AI was used only afterwards to separate the drum performance from a difficult source recording.
These are completely different production histories. The resulting sound may still be excellent, but if I am discussing musicianship, performance and authorship, the difference matters. Without some degree of provenance, music criticism could begin describing fictional labour.
We could find ourselves praising performances that never occurred. That does not automatically make the resulting sound artistically worthless. It simply means we need to know what we are actually reviewing.
“AI Music” Is Too Broad to Mean Very Much
This is why the phrase AI music increasingly bothers me. It collapses several fundamentally different processes into one expression.
At one end is technical AI: noise reduction, source separation, restoration, intelligent corrective processing, mastering assistance and other systems that primarily work upon existing human-created material.
Then there is AI-assisted production, in which human beings remain responsible for the central composition and performance but machine-learning systems help with parts of the process.
Further along comes generative assistance, where AI creates particular expressive elements — perhaps a texture, chord sequence, percussion layer, lyric, melody or backing voice — which are incorporated into a substantially human-directed piece.
Then there is genuine human-machine co-creation, where both parties materially determine the finished expression.
Beyond that sits substantially generative music, where the system creates major elements including melody, harmony, voice, instrumentation, arrangement or structure. At the furthest end, a human’s role may be primarily prompting, selecting and requesting revisions to an almost completely generated recording.
And then there is another category entirely: synthetic impersonation, where technology creates a recognisable performance associated with a real person who did not actually perform it.
The international recording industry itself has begun moving toward more precise language. In July 2026, IFPI and a broad coalition of music organisations supported terminology distinguishing AI-Generated from AI-Assisted sound recordings.
That is a much more useful direction, because AI helped clean my vocal is not the same statement as AI created my vocal.
Computer Music Is Not AI Music
This distinction is essential, especially for anybody involved with electronic music.
All generative-AI music is computer music. Not all computer music is generative AI.
Computer-assisted and algorithmic composition existed long before modern text-to-music systems. A producer running Ableton Live, FL Studio, Logic, Cubase, Pro Tools, Reason, Renoise or another DAW is not automatically creating AI music. Neither is somebody programming a synthesiser, using MIDI, sequencing, cutting samples or designing a procedural or algorithmic composition system.
The important question is not whether computation was involved. It is: who made the musical decisions?
A conventional DAW might wait for me to say, Put this snare here. A sampler waits for me to choose which fragment I want. A synthesiser waits for me to determine the notes or programming that activate it. A sequencer executes events I have specified.
A contemporary generative system can instead be asked:Give me a dark, atmospheric, 170 BPM jungle track with broken drums, dub bass, cinematic pads and a melancholic female vocal.
The system can then determine an enormous number of details: notes, rhythm, sound, singer, phrasing, chords, instrumentation, transitions and arrangement. The human has still made a creative decision. They have described an intention. But many of the detailed expressive decisions have moved elsewhere.
Traditional digital tools overwhelmingly ask: What do you want me to do?
Generative AI can increasingly be asked: Work out what should be done.
The Machine Has Been in the Studio for a Very Long Time
None of this began in 2023. Recorded music has always changed when new technology entered the room.
Electrical recording changed performance. Magnetic tape allowed sound to be physically rearranged. Multitrack recording dismantled the idea that a record had to document one continuous performance. Synthesisers made entirely new timbres possible. Drum machines mechanised rhythm. Sequencers automated musical events. MIDI separated musical instructions from the machine producing the sound. Sampling turned recorded history itself into raw material. Digital audio workstations allowed one person to undertake work previously distributed through entire studios.
Algorithmic composition predates today’s generative-AI boom by decades.
The historical progression therefore is not human music → suddenly AI. It is closer to acoustic instruments → electrical recording → magnetic tape → synthesis → sequencing → sampling → MIDI → digital editing → algorithmic systems → machine learning → generative AI.
The machine has been in the studio for generations. The real question is what we ask the machine to do.
Dub Already Taught Us That a Studio Can Be an Instrument
Anyone claiming that authentic music must involve untouched instrumental performance has to explain dub.
Dub transformed the studio from a place where music was merely recorded into an environment where the recording itself could be performed. Vocals disappeared. Bass and drums dominated. Delay became rhythm. Reverb became architecture. Fragments entered and vanished. The mixing desk became an instrument.
Technology was no longer outside the music. It was part of the musical language. The producer and engineer could become performers of the recording itself.
Industrial Music Never Feared the Machine
Industrial and experimental music pushed the question still further. Tape loops, found sound, noise, oscillators, distortion, mechanical repetition, drum machines, sampling, cut-ups and processed speech repeatedly challenged the assumption that a musical instrument needed strings, skins or keys.
The machine could be instrument, subject, environment and metaphor simultaneously.
So an argument that says machines make music fake would erase much of the culture that produced the music I have spent decades supporting. But that does not automatically make every use of generative AI equivalent to a synthesiser.
Jungle Shows Us Why the Difference Matters
Jungle and drum & bass make the question beautifully complicated.
A drummer performs a break. It is recorded. Somebody samples the recording years later. The break is accelerated, cut into tiny pieces, re-pitched, reversed, time-stretched and rearranged. The resulting rhythm may contain combinations that the original drummer never physically performed.
Was the sampler cheating? Music history largely decided that it was not. The sampler became an instrument. Editing became musicianship. The decision-making became composition.
But this example also exposes the difference between sampling and contemporary generative AI. The jungle producer may spend hours making hundreds of detailed decisions about individual milliseconds of audio. If somebody instead types Make me a dark 1994 atmospheric jungle break and the machine supplies a finished rhythmic solution, much more of the decision-making has been delegated.
The sonic language may sound related. The creative process is not identical.
The Beatles — When AI Helps Recover Rather Than Invent
One of the most useful case studies is The Beatles’ “Now and Then.”
Machine-learning-assisted MAL technology developed through Peter Jackson’s WingNut Films work was used to separate John Lennon’s existing recorded voice from other elements in an old demo.
The machine did not invent a new Lennon vocal. It helped expose a human performance that already existed.
That distinction is fundamental: restoration recovers; generation invents.
The same broad family of computational techniques may participate in both, but culturally, ethically and musically they are not the same act.
And this opens another side of artificial intelligence that deserves much more attention.
When the Machine Saves the Recording
Much of the argument surrounding AI concerns what machines can create. But what about what they can save?
What happens when the master tape has disappeared? What happens when the only surviving recording has been played hundreds of times? What happens when the cassette has deteriorated through heat, humidity, age or poor storage? What happens when a reel-to-reel master suffers from binder breakdown?
Magnetic tape is not immortal. Physical and chemical deterioration can damage the carrier and interfere with playback. Sticky-shed syndrome, for example, can cause deteriorated binder to leave residue on tape heads and guides and make safe playback increasingly difficult.
Sometimes there is no perfect master waiting somewhere else.
The damaged tape is the archive.
Human Restoration Is an Art of Its Own
It is important not to discuss AI restoration as though human restoration engineers were simply waiting for software to make them redundant.
Serious audio restoration can begin before the recording has even been played. A specialist may inspect the physical tape, identify its condition, determine whether it can be played safely, select the correct machine, clean and calibrate equipment, establish the proper speed, align the tape path, correct head azimuth, monitor tension, listen for mechanical problems and decide whether the carrier requires specialist treatment.
International archival guidance recommends careful replay alignment and correction of azimuth because information lost through poor physical playback cannot simply be recreated afterwards through digital processing.
That matters. If the original signal is extracted badly, no sophisticated AI can magically restore information that was never captured properly in the first place.
Sometimes the most important person in an AI-assisted restoration remains the person carefully threading an old tape through the correct machine.
Sometimes There May Be Only One Good Playback Left
Badly deteriorated magnetic media can require specialist intervention before transfer. Sticky-shed deterioration can cause audible squealing, residue accumulation and reduction in audio quality. Certain vulnerable tapes may require carefully controlled treatment before a preservation transfer can safely occur.
Imagine the responsibility. There may be no duplicate, no safety copy, no multitrack and no alternative source. The restoration engineer may effectively be deciding how to retrieve the only surviving physical record of a performance.
They must decide how much noise can be removed before the music itself disappears; whether distortion is damage or part of the original recording; whether a click is a fault or part of the instrument; and whether cleaning a tape makes it clearer or simply less authentic.
These are human judgements.
Restoration Is Not the Same as Modernisation
A good restoration does not necessarily attempt to make a 1978 cassette sound as though it was recorded yesterday. Sometimes the goal is much more modest and much more difficult: make the original performance as accessible as possible without destroying what it was.
Tape hiss may be annoying. Remove too much and you may also remove cymbal decay, breath, ambience and guitar texture. Remove every click and you risk damaging musical transients. Push noise reduction too far and voices begin acquiring watery digital artefacts.
The machine can analyse. The machine can propose. The human restoration engineer decides where to stop.
Now AI Can Enter the Restoration Room
Contemporary restoration software can use machine learning and advanced spectral processing to isolate dialogue, music and effects, separate stems, reduce noise and repair damaged material.
Professional systems such as iZotope RX increasingly combine machine-learning-assisted analysis with detailed manual control.
This is a profoundly different use of artificial intelligence. Here the machine is not being asked, Compose something. It is being asked:
Help me hear what is already there.
But Restoration Can Cross Into Reconstruction
Suppose AI removes hiss from a damaged vocal. That is restoration. Suppose it separates that existing vocal from a piano. Still restoration or recovery.
Suppose several seconds of the original vocal are physically missing and a generative system creates the words or notes it predicts the singer would have performed. Now something different has happened.
That missing performance was not recovered. It was constructed.
The result may be musically convincing, useful or even beautiful, but it should be described accurately.
Reconstruction is not restoration.
Once again, the question is not simply whether AI existed somewhere in the process. It is what AI was authorised to do.
Then There Is the Person With a Box of Old Cassettes
Not every important recording belongs to a major label or a national archive. Some of the most valuable recordings in existence are sitting in wardrobes, garages, sheds and shoeboxes.
The only surviving recording of somebody’s first band. A rehearsal made when everyone was seventeen. A punk band that existed for three months and never entered a professional studio. A community-radio broadcast. An early rave recording. A dub session. A home-recorded electronic experiment. A cassette label that disappeared forty years ago. A song performed by somebody who later died. A parent playing an instrument. A friend speaking between songs. A mixtape made by somebody you once loved.
Commercially, the cassette may be worth almost nothing. To its owner it may be priceless.
This introduces another category of value that the commercial music industry does not measure particularly well: sentimental, psychological and symbolic value — memory itself.
A Mixtape Can Be More Than the Songs on It
Think about an old cassette somebody made specifically for another person. There are the songs, but there may also be handwritten track choices, radio announcements, a badly timed pause button, a few seconds of room conversation, tape hiss, a click, a voice in the background, a drop-out before the chorus or a cassette mechanism stopping.
Technically, these things may be defects. Emotionally, they may be the entire reason the object matters.
Suppose AI can remove every trace of tape hiss. Should it? Suppose it can remove the person speaking quietly in the background. What if that person is no longer alive?
At that point audio restoration becomes partly a question of memory ethics.
Sometimes damage has become part of the remembered object. A useful approach may be to preserve an archival transfer as faithfully as possible and create a second restored listening version.
The original survives. The restoration becomes another interpretation rather than a replacement.
The Garage May Contain Cultural History
This matters enormously for underground music.
History is not always stored in famous studios. Sometimes history is sitting on a cassette in somebody’s garage.
Independent punk. Industrial tapes. DIY electronic projects. Early jungle and rave recordings. Small sound-system sessions. Rehearsal-room tapes. Community radio. Bands that never had a record contract.
Much of underground culture survives because somebody simply never threw the tape away.
AI-assisted restoration could potentially help make portions of that enormous informal archive audible again. In this context, AI is not necessarily threatening underground history.
It may help preserve it.
AI Could Also Democratise Personal Restoration
Professional archival restoration is skilled work, and skilled work costs money. Many ordinary people cannot afford to send every family cassette, rehearsal tape or home recording to a specialist restoration studio.
Increasingly accessible software can provide basic noise reduction, repair and separation tools that previously required far more specialised equipment and expertise.
That opens an interesting consumer possibility. Millions of people may eventually be able to recover recordings that would otherwise remain unheard.
Some will achieve useful results themselves. Others may discover that the most valuable or physically fragile tapes still require a professional.
So AI could paradoxically reduce some routine restoration labour while increasing public awareness of audio preservation and creating new demand for specialists.
The future may not be human restorer versus AI.
It may be an experienced human restorer using AI as another tool.
The Human Voice Creates Another Boundary
Restoration brings us naturally to the issue of synthetic voice.
The voice is not simply another synthesiser patch. A voice communicates identity, accent, age, language, geography, body, personality and history.
AI can now participate in voice modelling, which means the key issue may increasingly be consent.
Holly Herndon’s Holly+ project offers a useful counter-example to unauthorised cloning. It deliberately explores an authorised digital version of Herndon’s voice and wider questions about governance, identity and the use of synthetic performance.
The important question therefore is not merely whether the voice was artificial. It is:
Did the person whose voice became the model agree?
An authorised synthetic voice and an unauthorised imitation may use related technology. Ethically, they are worlds apart.
Style Is Not the Same as Identity
Music has always developed through influence. No musician owns every distorted industrial bass tone, every minor-key post-punk harmony, every dub delay, every chopped break at 170 BPM or every shoegaze reverb.
Genres exist because musicians learn, imitate, mutate and transform shared vocabularies.
But there is an important distinction between make something that belongs to this broad musical tradition and make listeners believe this specific human being performed something they never performed.
That is where questions of voice, identity, consent and deception become particularly serious.
Copyright Is Beginning to Ask the Same Question as Music
The United States Copyright Office has approached generative AI through a principle extremely relevant to musicians:
Where is the human authorship?
Its 2025 analysis concluded that using AI as part of a creative process does not automatically prevent copyright protection. Human-authored material, sufficiently creative selection and arrangement and meaningful human modification may remain protectable.
But prompting alone generally does not provide enough human control over the generated expressive detail to make the user author of everything the system produced.
That does not settle every philosophical question about creativity, but it creates a useful practical distinction.
The law is not simply asking whether AI was present. It is asking:
What did the person actually author?
Now Return to the $1,000 Song
This brings us back to economics.
A traditionally recorded song can support an entire chain of labour: songwriter, singer, guitarist, bassist, drummer, keyboard player, programmer, producer, recording engineer, assistant engineer, editor, mix engineer, mastering engineer, session musicians and the studio itself.
A generative system can increasingly be asked to produce lyrics, melody, harmony, instrumentation, voice, arrangement, mood, tempo and production characteristics.
One subscription therefore begins occupying economic territory that was previously distributed among multiple human occupations.
That does not mean it performs all those jobs in the same way. It means the customer may decide that it supplies enough of the desired outcome that the human beings are no longer hired.
Economically, acceptable can matter more than perfect.
The Producer Is Not Just Somebody Who Makes Sound
If human production is to defend its value, the argument cannot simply be that humans sound better. That may become increasingly difficult to demonstrate universally.
The stronger argument is that a producer does far more than create sound.
A producer might say: The first verse is too long. That technically perfect vocal is emotionally dead. Use the rough take. Stop playing so much. Delete the synth. The song is trying too hard. Don’t fix that mistake — that’s the most human part of the recording.
The producer interprets the person. They notice insecurity, excitement, fatigue, fear and overconfidence. They can change the psychological atmosphere of the room.
The relationship between artist and producer can become part of the finished recording. That is difficult to quantify economically, but it is real.
The Engineer Is Not an EQ Preset
The same is true of the recording engineer.
Engineering involves acoustics, microphone choice, placement, phase, gain structure, room behaviour, signal flow, monitoring, bleed, equipment faults and instrument tuning.
A vocalist takes half a step backwards and the sound changes. A drummer receives a different headphone mix and plays differently. An amplifier begins behaving unpredictably and somebody decides to keep it.
AI can automate increasing amounts of technical processing. But recording actual people in physical space remains a physical and interpersonal event.
That may become increasingly valuable as synthetic production grows.
The Most Vulnerable Music May Be the Music Nobody Cares Who Made
The first major economic pressure may not fall upon internationally famous artists. It may fall upon music whose customer primarily requires a function.
Advertising beds, corporate videos, generic social-media music, low-budget game cues, podcast backgrounds, library music, temporary film scoring, demo arrangements and generic mood music are obvious examples.
Imagine a client who simply needs two minutes of uplifting electronic music without vocals. Historically, that could mean a composer, producer, library or licensing service. Today it can also mean generating twenty possibilities almost immediately.
The human composer does not have to lose because they are less talented. They may lose because they are more expensive.
This is one reason the phrase good enough matters so much.
Creator Organisations Are Worried About Exactly This Substitution
CISAC commissioned economic modelling which projects that, under its assumptions, 24% of music creators’ revenues could be at risk by 2028, with music libraries among the areas especially exposed to generative competition.
These are forecasts, not established future outcomes. They should be treated as economic scenarios rather than prophecy.
APRA AMCOS’s Australia/New Zealand research reveals a similarly complicated picture. It found 38% of respondents already using AI in their work, while 54% agreed that AI could assist human creativity. At the same time, 82% were concerned that AI could threaten creators’ ability to make a living.
These findings destroy the simplistic idea that musicians divide neatly into pro-AI and anti-AI camps.
A musician can use source separation and oppose unauthorised training. A producer can use intelligent mastering while rejecting synthetic impersonation. An experimental composer can work creatively with generative systems and still demand that musicians whose work feeds commercial models are compensated.
There is no contradiction in those positions.
Lower Cost Can Also Open Doors
Professional production is expensive. That has always excluded people.
Someone may have extraordinary ideas but no money for a studio. Another musician may live far from professional recording facilities. Someone may not know session players. Someone with a disability may find traditional instruments physically inaccessible. A songwriter may want to hear a rough orchestration before investing money in professional recording.
AI can lower those barriers. It can help people sketch ideas, prototype arrangements, explore sounds and experiment before spending money.
That can represent genuine democratisation.
So we must be capable of holding two apparently conflicting truths at once:
Cheaper creation can give more people a voice.
Cheaper creation can reduce the value of people who previously earned their living providing that labour.
Both can be true.
Technology Has Displaced Musical Labour Before
This is not the first time technology has changed who gets hired.
Drum machines affected certain forms of session drumming. Synthesisers could replace instrumental parts. Sampling changed how recordings were constructed. Home studios reduced some demand for commercial facilities. DAWs transformed editing. Software instruments allowed one producer to create arrangements that once required many performers.
But those same technologies also produced new jobs, genres and cultures: electronic producers, programmers, plugin developers, sound designers, bedroom studios, new engineers, new labels and entire forms of music that previously could not exist.
Generative AI may repeat part of that pattern. Its difference may lie in its breadth. It can operate across composition, arrangement, instrumentation, voice, lyrics, sound design and production simultaneously.
Labels Are Already Having to Choose
Record labels now face the same provenance question broadcasters face.
What exactly are they signing? Who wrote it? Who performed it? Can it be protected? Who owns the training material? Is there an unauthorised voice hidden inside it? Can the label safely exploit the recording?
LabelRadar has already introduced disclosure categories asking whether music was created mostly with generative AI, partly with generative AI or without generative AI. Importantly, its policy explicitly excludes ordinary mixing and mastering tools, audio cleanup, stem separation and noise reduction from its definition of generative-AI music.
The information is then used to route submissions according to label preferences.
That is exactly the kind of distinction the industry needs.
Some Labels Have Drawn a Harder Line
Some independent labels simply do not want generative material.
Globetrotting Music, for example, currently asks prospective artists to submit material containing no AI-generated content, alongside restrictions concerning samples, cover melodies and certain royalty-free material.
That is a legitimate curatorial position.
A label has never been merely a mechanism for distribution. A strong label represents taste, community, aesthetic identity and trust.
Some labels may decide that predominantly human authorship is part of what their name means. Others will not. That diversity of policy is likely to increase.
We may eventually see human-only labels, hybrid labels, AI-assisted labels and labels built specifically around generative artists.
Bandcamp Has Made Human Creation Part of Its Identity
Bandcamp adopted one of the strongest platform positions in January 2026.
It states that music and audio generated wholly or substantially by AI are not permitted, and that AI tools cannot be used to impersonate artists or styles.
Bandcamp explicitly frames the policy around maintaining confidence that music found on the platform was created by human beings.
Notice the wording: wholly or in substantial part.
That matters. It is not the same as claiming that any recording touched by machine learning must be rejected.
This distinction will become increasingly important as machine-learning functions disappear invisibly into ordinary production software.
Deezer Has Chosen a Different Road
Deezer has not simply banned all generated music. Instead it has built detection, labelling and recommendation policies around it.
In July 2026, Deezer reported receiving approximately 90,000 fully AI-generated tracks every day, exceeding 50% of new daily deliveries at peak levels in June. Yet those recordings represented only about 1–3% of actual listening.
Deezer also reported that up to 85% of streams associated with fully AI-generated tracks during 2025 were fraudulent, and it removes detected AI music from algorithmic recommendations and editorial playlists.
Those numbers reveal something important.
The immediate problem may not be that everybody prefers AI music. The problem may be that the supply of music becomes almost infinite.
A human being sleeps, gets sick, loses inspiration and takes six months to finish an album. A generative system can produce another version, then another, then another.
When music itself becomes effectively limitless, the scarce commodity becomes attention.
Independent musicians were already fighting for attention before generative AI arrived. Now that competition can potentially be industrialised.
Most People Cannot Reliably Tell by Ear
Deezer’s international Ipsos research involved 9,000 people across eight countries. Deezer reported that 97% failed to identify every AI-versus-human example correctly in its blind listening test.
That does not prove that AI music is artistically equal to human music, nor does it mean trained engineers never hear clues.
It means something narrower:
“I can always tell” is not a serious industry policy.
And this is another reason false accusations are dangerous. A strange vocal does not prove AI. A polished master does not prove AI. An artist releasing large quantities of music does not prove AI. Electronic music sounding synthetic certainly does not prove AI.
Evidence matters.
Spotify Is Taking Yet Another Approach
Spotify has emphasised artist control while targeting impersonation, spam and deception rather than treating every AI use as illegitimate.
Its current policy prohibits unauthorised vocal impersonation while allowing artists to authorise use of their own voices. Spotify also supports more granular AI disclosures in music credits and has argued that AI involvement should be treated as a spectrum rather than a simple AI/not-AI binary.
This gives us three very different platform philosophies.
Bandcamp says substantially generative music does not belong there. Deezer allows it but identifies and restricts its recommendation. Spotify focuses on consent, spam, impersonation and disclosure.
There is no single industry answer.
Major Labels Are Moving From Courtrooms to Licensing Tables
The major-label position is equally revealing.
The early phase of the current generative-music conflict involved litigation and allegations of unauthorised use of copyrighted recordings. But by late 2025, some of those relationships were shifting toward licensing.
Universal Music Group and Udio announced a settlement and plans for a licensed AI music-creation platform. Warner Music Group has said its AI partnerships should involve licensed models, appropriate economic value and artist/songwriter choice over uses of name, image, likeness and voice.
Separate licensing agreements involving Universal, Sony, Warner and AI-music companies have pointed in a similar direction.
This may tell us where much of the commercial industry is heading.
The argument becomes less Should AI exist? and more:
Who licences it? Who gets paid? Who can opt in? Who controls the voice? Who controls the catalogue?
Different Countries Are Running Different Experiments
There is no single international law governing AI music. Instead, the world is effectively testing several philosophies at once.
🇦🇺 Australia
Australia’s Copyright and Artificial Intelligence Reference Group continues examining issues including AI inputs, transparency, licensing and enforcement. Importantly, the Australian Government has stated that it is not considering a broad text-and-data-mining exception to Australian copyright law.
That does not solve every AI question, but it means Australia is not simply abandoning existing copyright controls in the name of training AI.
🇪🇺 European Union
The European Union has moved strongly toward transparency. Article 50 of the EU AI Act became applicable from 2 August 2026 and includes obligations aimed at making certain AI-generated or manipulated content machine-detectable and appropriately disclosed.
The principle is significant: if synthetic content becomes normal, its origin should become more visible.
🇬🇧 United Kingdom
Britain remains in a much more contested policy phase. The UK Government’s March 2026 report examined multiple approaches to copyrighted material used in AI development, alongside transparency, enforcement, computer-generated works and digital replicas.
The UK has not produced a simple answer that makes the conflict disappear. The debate remains one of balancing a major creative sector against ambitions in AI development.
🇺🇸 United States
The United States Copyright Office has placed human authorship at the centre of copyrightability. Its current position allows protection for sufficiently human-authored elements in AI-assisted works while rejecting the idea that prompts alone necessarily constitute authorship of machine-determined expression.
Again the question becomes: Where did the human make the expressive decision?
🇯🇵 Japan
Japan has developed its own detailed interpretation of AI and copyright. The Agency for Cultural Affairs’ General Understanding on AI and Copyright in Japan examines the application of existing copyright law to AI development and output and stresses that its guidance represents an interpretation of current law rather than a binding answer for every specific technology.
Japan demonstrates why claims that AI training is simply legal or illegal worldwide are misleading. The legal position depends upon jurisdiction and circumstances.
🇨🇳 China
China has taken a notably technical approach to synthetic-content identification. Its rules distinguish explicit labels, visible or audible to users, and implicit labels embedded technically within generated files.
The measures cover generated text, images, audio, video and virtual content, and official guidance describes metadata containing information about generated-content status and service-provider identity.
This has a remarkable connection with the anonymous case study earlier in this article.
Metadata itself can become part of provenance.
🇨🇦 Canada
Canada’s federal consultation on copyright and generative AI has concentrated on broad questions including copyrighted material used for AI training, authorship and ownership of generated material, and liability when generated output infringes copyright.
Canada, like Britain, illustrates a legal system attempting to catch up while the technology is already commercially deployed.
🇳🇿 Aotearoa New Zealand
New Zealand offers another interesting contrast. Government cultural research published in 2026 found high adoption of generative AI among creators already using digital tools, including use for exploring ideas, generating work and assisting dissemination.
Rather than one global settlement, we therefore have multiple regulatory experiments occurring simultaneously.
Indigenous Culture Makes the Question Deeper Than Copyright
There is another issue that should not disappear beneath commercial arguments about royalty payments.
APRA AMCOS reports that 89% of Aboriginal and Torres Strait Islander music creators surveyed believed AI could increase cultural appropriation.
That raises questions copyright does not always answer adequately. A musical form may contain ceremonial significance, language, community ownership, traditional knowledge or cultural restrictions. A generative system can reproduce sonic characteristics without understanding any of those relationships.
The question is no longer simply, Who owns this recording?
It may be:
Who has the cultural authority to use this material?
AI did not invent cultural appropriation. But industrial-scale generation can accelerate it enormously.
Perhaps Music Will Split Into Two Economies
One possible future is a separation between two broad music economies.
One may contain enormous quantities of inexpensive, functional, synthetic music: background music, mood playlists, advertising, corporate content and generic soundtrack work.
Beside it may develop another economy centred increasingly on human provenance.
A record may eventually advertise:
Human composed. Human performed. Human produced. Human mixed. Human mastered. No substantial generative AI.
This may sound excessive now, but Bandcamp’s current policy already turns human creation into part of its cultural identity.
The strange consequence of synthetic perfection may be that human imperfection becomes more valuable: the breath before the vocal, the slightly late snare, amplifier hum, accidental feedback, the flawed note everyone decided to keep — evidence that somebody was actually there.
An AI Can Generate Dub — But Did It Live Dub?
This is where the debate becomes cultural rather than purely technological.
A generative system can learn sonic characteristics associated with dub: deep bass, spring reverb, tape-style delay, dropouts and offbeat rhythm.
But dub is also Jamaica, sound-system competition, studios, migration, engineering experimentation, Rastafari, economics, community and history.
Jungle is more than an Amen break. Industrial music is more than distortion. Goth is more than minor chords. Shoegaze is more than reverb. Future garage is more than shuffled drums and rain samples.
Genres are not merely statistical clusters of sonic features.
They are histories expressed through sound.
A model can learn the features. Whether it can possess the history is another question.
Human Music Has Never Been Completely Pure Either
We should also be careful about romanticising human creation.
Human musicians imitate. We use presets. We learn established chord progressions. We borrow production techniques. We absorb scenes. A jungle musician learns by listening to jungle. A dub producer inherits dub vocabulary. A gothic band learns what gothic music sounds like.
That fact is often used to compare AI training with human learning.
There is some conceptual resemblance, but the difference in scale remains enormous.
A human cannot ingest millions of recordings in an industrial computing system and generate thousands of commercial alternatives continuously.
Scale changes cultural and economic consequences.
What Does “No AI” Actually Mean?
There is another difficult question for strict label policies.
Suppose somebody declares:
No AI has been used anywhere in this recording.
What counts? Machine-learning noise reduction? Stem separation? An intelligent EQ? Automatic mastering? Restoration? A plug-in whose internal architecture the musician does not even know contains machine-learning technology?
This is why an absolute prohibition on anything containing AI technology may become increasingly difficult to define.
A clearer artistic boundary could be:
No substantial generative-AI authorship.
That focuses on who created meaningful expressive material rather than whether a hidden algorithm appeared somewhere inside the production software.
LabelRadar’s policy makes essentially this practical distinction by separating generative composition from mixing, mastering, cleanup, stem separation and noise reduction.
Perhaps Recordings Need Something Like an Ingredient Label
Imagine if music releases simply told us:
Composition: Human Lyrics: Human Lead vocal: Human performance Drums: Human programmed Instrumentation: Human + software instruments Generative AI: Atmospheric texture in introduction Mix: Human engineer Master: AI-assisted, human approved Synthetic voice: None
Or:
Concept: Human Lyrics: Human Composition: Predominantly generative AI Voice: Synthetic Arrangement: Human edited Mix: AI-assisted Disclosure: Generative production
Now the audience actually knows something.
Most importantly:
Neither label tells us whether the record is good.
That remains a musical judgement.
Transparency is not censorship.
It is information.
What I Would Like Labels to Tell Broadcasters
If a label sends music to me for radio or review, I do not need a twenty-page technical declaration. But increasingly I would value a straightforward answer to a few questions.
Was generative AI used in the composition? Was a synthetic lead voice used? Were major instrumental parts generated? Were the lyrics generated? Was AI limited to mastering, restoration, separation or technical processing? Was an identifiable human voice modelled? Was that use authorised?
That is enough to contextualise the work properly.
I Do Not Want to Become the AI Police
This is just as important.
I receive too much music to conduct a forensic investigation on every recording. Nor should broadcasters be expected to operate as technological detectives.
My job is to listen, research, contextualise, review, broadcast and support independent music.
False accusations could seriously damage real artists.
So two principles have to coexist:
Disclose substantial generative involvement.
Do not accuse musicians without evidence.
The anonymous case study in this article is useful precisely because it moved beyond subjective listening. Ankit raised the possibility. My ears made me curious. The file metadata then supplied technical evidence.
That is very different from making a public accusation because something sounds strange.
A Possible Independent Radio Position
For my own purposes as a broadcaster and reviewer, I would draw the line this way.
Technical AI assistance — restoration, source separation, noise reduction, corrective processing and related tasks — does not by itself transform a substantially human recording into “AI music.”
Generative AI assistance — where AI creates meaningful expressive elements such as lyrics, melody, instrumentation or vocals — should be disclosed.
Substantially generative music should be identified clearly when AI has created most of the central audible musical material.
Synthetic identity should require meaningful authorisation if a real person’s voice or identity is represented.
And evidence should come before accusation.
That is not anti-AI. It is simply pro-information.
So Is AI Cheating?
Sometimes. Sometimes absolutely not.
The word cheating requires a rule.
If a songwriting competition requires original human composition and somebody secretly submits prompt-generated music, that is cheating.
If somebody says I sang this and no such performance occurred, that is deception.
If somebody manufactures an unauthorised synthetic performance designed to make listeners believe a real artist participated, the issue becomes considerably more serious.
But using machine learning to remove noise? That is difficult to describe sensibly as cheating.
Source separation? Again, hardly the same thing.
Using an AI-generated sound as raw material and spending hours destroying and rebuilding it? That enters a much more complicated creative area.
Generating an entire composition and openly stating that it was generated? A listener may dislike the method, but there has not necessarily been deception.
So perhaps cheating is the wrong organising principle.
The Better Question Is Delegation
Instead of asking Did you use AI?, perhaps we should ask:
What did you delegate to AI?
Did you delegate noise reduction, mastering, stem separation, a texture, a bass line, a drum pattern, harmony, lyrics, melody, the vocalist, the arrangement or the entire song?
The differences between those answers are enormous.
And Then Comes the Question of Ownership
Imagine a track in which the human wrote the lyrics, AI generated a melody, the human selected a voice, AI performed it, the human rearranged the second verse, AI created the instrumentation and a human mix engineer then spent two days radically transforming the result.
Whose record is it?
There may be no single philosophically perfect answer.
But there is one minimum requirement that seems increasingly reasonable:
Tell us what happened.
Then listeners, critics, broadcasters, labels and artists can decide what significance to attach to it.
The Tools Mean Nothing Without the Artist
There is one thought I keep returning to when considering artificial intelligence, studio technology and the future of music, and it comes from my late father, Lawrence Alexander Taylor, born in 1925.
My father belonged to a generation that witnessed astonishing technological change, yet he was always looking further ahead. He often spoke about future tools, future machines and technologies before many of them existed in anything like the form we recognise today. He understood that recording equipment would become more powerful, more accessible and capable of doing things that would once have seemed impossible.
But he was equally clear about something else. Technology alone could never make somebody an outstanding artist.
“You can have all the technology in the world. You can have all the tools, all the money, the finest studio and the best equipment. But none of it means anything if you do not have what is within you — the creativity to become an outstanding artist, to make a difference, and perhaps to create a new sound.”
— Lawrence Alexander Taylor, 1925–2007
Coming from somebody born in 1925, those words have always stayed with me.
He was fascinated by where technology might lead, but he never confused the power of the tool with the power of the artist using it.
That distinction feels extraordinarily relevant now.
Artificial intelligence can make production faster. It can reduce costs. It can restore damaged recordings, separate instruments, correct faults, suggest arrangements, generate sounds, create voices and produce complete songs.
But none of those abilities guarantees that the result will matter.
You can have the most expensive recording studio in the world and produce something completely forgettable. You can own every synthesiser, microphone, computer and plug-in available and still have nothing original to say.
Equally, somebody with almost nothing can create a sound that changes a culture.
Music history is full of people who transformed inexpensive, primitive, damaged or supposedly limited technology into something genuinely new: tape machines, basic synthesisers, early samplers, cheap drum machines, home computers, second-hand equipment, small studios, bedrooms, garages and improvised sound systems.
The equipment became important because somebody had the imagination to use it differently.
The tool matters. Knowledge matters. Access matters. Technology matters.
But the creative impulse still has to come from somewhere.
Perhaps my father’s words become even more important in the age of artificial intelligence because we are approaching a period in which almost anybody may eventually have access to production capabilities that once required an extremely expensive professional studio.
If extraordinary technology becomes ordinary, then the real distinction may return to one of the oldest questions in art:
What does the person using it have within them?
AI may give somebody extraordinary tools.
It cannot automatically give them an extraordinary artistic vision.
And perhaps that is exactly what my father understood decades before these technologies existed.
The Irony — AI May Make Human Music More Valuable
For decades digital technology made recorded music increasingly abundant. Copies became practically infinite. Streaming turned millions of recordings into an instantly accessible library.
Generative AI now introduces something qualitatively different: the possibility that new music itself becomes almost infinitely producible.
If that happens, what remains scarce?
Not the MP3. Not the WAV file. Perhaps not even the song.
What remains scarce may be attention, trust, identity, community, history, reputation — and a real human story.
Perhaps the unexpected consequence of synthetic music will be to remind us why human-made music mattered in the first place.
And this is where my father’s observation returns.
The tools may change beyond recognition. The studio may become software. The software may become intelligent. The machine may become capable of extraordinary things.
But the tool is still not the artist.
Conclusion — What Part of the Music Was Still Yours?
The argument surrounding artificial intelligence and music is frequently reduced to two slogans. One side says AI is cheating. The other says AI is just another tool.
Neither statement is good enough.
AI can be a restoration device, a mastering assistant, a source-separation tool, an accessibility technology, a compositional aid, a sound generator, a collaborator, a vocalist, a composer, an impersonator, a mass-production system, a preservation tool and a substitute for human labour.
Those uses cannot sensibly be treated as the same thing.
Electronic and underground music should understand this better than almost anyone. Dub demonstrated that a mixing desk could become an instrument. Industrial music placed machines inside its artistic language. Jungle turned sampled fragments into rhythmic structures that no conventional drummer had physically performed. Drum & bass grew alongside samplers, sequencers and increasingly powerful computers. Ambient, darkwave, future garage, post-dubstep and experimental electronic music repeatedly blurred the boundary between instrument, studio, processing and composition.
So rejecting something simply because a machine participated would erase much of our own history.
But that history does not require us to pretend generative AI is merely another guitar pedal.
Something has changed.
Traditional machines overwhelmingly expanded our ability to execute musical decisions. Generative systems increasingly have the capacity to make musical decisions.
That boundary deserves serious attention.
My own experience brought the issue down from theory. A recording arrived. Ankit of LazyboyProactive, a close friend whose musical and technical judgement I respect, suggested that AI might have been involved. Listening alone could not establish that, so the digital material was examined.
Multiple instrumental stems contained metadata referring to “made with suno studio.” The same material contained traces of conventional digital production software.
That small case contained the entire modern debate.
It was not simply human versus machine. It was a production chain in which human and machine processes appeared to coexist.
And that question now follows me into the hundreds of pieces of music, artists, bands and labels that pass through my world.
Some recordings may be completely human-created. Some may use AI only to master or repair a human performance. Some may include one generated element. Some may be genuine human-machine collaborations. Some may be almost entirely synthetic.
I cannot responsibly determine all of that from sound alone. Nor can audiences. Nor, increasingly, can labels rely upon assumptions.
At the same time, AI may be doing something extraordinarily valuable. It may help rescue a John Lennon vocal that already existed. It may help a restoration engineer separate music buried beneath noise. It may make the only surviving recording of a forgotten band audible again. It may allow somebody to recover a cassette of a parent, friend or partner whose voice they have not heard clearly for decades.
That recording may have no commercial value whatsoever. Its emotional value may nevertheless be impossible to calculate.
Artificial intelligence can therefore simultaneously raise questions of employment and preservation, exploitation and accessibility, authorship and restoration, impersonation and memory.
Governments around the world are responding differently. Australia is maintaining its existing copyright framework rather than introducing a broad text-and-data-mining exception. Europe is imposing new transparency obligations. The United States is emphasising human authorship. Japan is interpreting AI through its own copyright structure. China is building explicit and metadata-based labelling into its approach to generated content. Canada continues working through training, authorship and liability. New Zealand shows high creative adoption alongside continuing cultural and economic questions.
Platforms and labels are dividing as well. Bandcamp has drawn a human-first boundary. LabelRadar demands disclosure and allows labels to make their own decisions. Some independent labels reject generative content outright. Deezer detects and labels fully generated material while confronting an extraordinary volume of synthetic uploads. Spotify concentrates on consent, impersonation, spam and detailed disclosure. Major recording companies are increasingly building licensed AI relationships rather than simply attempting to make the technology disappear.
And underneath all of these arguments lies money.
A traditional record can provide work for musicians, producers, recording engineers, mixers, mastering engineers, studios and session performers. Those people have equipment costs, training, experience, families, rent, mortgages and decades invested in learning how to listen.
A machine does not have to become more culturally meaningful than they are in order to compete. It merely has to become sufficiently useful to somebody who otherwise would have paid them.
That is why perhaps the most dangerous phrase in this entire debate is not artificial intelligence.
It is:
GOOD ENOUGH.
At the same time, the same technology can help somebody who could never afford a studio hear their own musical imagination realised. It can help recover a damaged tape. It can reconnect somebody with a memory. It can provide new instruments to people whose circumstances previously excluded them.
This is why the subject cannot be reduced to a moral panic.
AI can democratise. AI can exploit. AI can assist. AI can replace. AI can restore. AI can imitate. AI can create.
Several of these things can be true at the same time.
So I no longer believe the most important question is:
Is AI cheating?
The more difficult and useful questions are:
Who made the musical decisions? What did the machine actually generate? What did the human actually create? Was the machine recovering something — or inventing something? Whose work helped build the system? Who consented? Who was credited? Who was paid? Who may no longer have a job? And did the listener know what they were hearing?
Then comes the final question.
Perhaps the one that will define this period in music history.
When machines can produce endless technically convincing recordings, when another song can be generated almost instantly, when listeners cannot always distinguish the process by ear, when AI can sit beside the DAW, the sampler, the producer and the engineer, and when that same technology can either manufacture a performance that never happened or rescue a performance we thought was lost forever:
WHAT PART OF THE MUSIC WAS STILL YOURS?
My late father, Lawrence Alexander Taylor, understood the essential point long before anybody was discussing generative AI.
The tools may become extraordinary.
But extraordinary tools do not automatically create an extraordinary artist.
And perhaps beyond even that lies one final question:
WHEN THE RECORD IS FINISHED — DO WE STILL CARE WHO, OR WHAT, MADE IT?
The answer may determine far more than the future of artificial intelligence.
It may determine what we eventually decide music itself is worth.
Selected Research and References
Australian Government — Copyright and Artificial Intelligence Reference Group — Australian work examining copyright, AI inputs, licensing, transparency and enforcement.
Australian Attorney-General — AI and Text/Data Mining — Australian Government position confirming that a broad text-and-data-mining copyright exception is not presently being pursued.
APRA AMCOS — AI and Music Report — Australia/New Zealand creator research covering AI adoption, economic forecasts, creator livelihoods and Aboriginal and Torres Strait Islander cultural concerns.
European Commission — EU AI Act, Article 50 — Transparency requirements concerning certain AI-generated and manipulated material, including machine-readable identification.
UK Government — Copyright and Artificial Intelligence Report, March 2026 — Policy analysis covering AI training, licensing, transparency, enforcement, computer-generated works and digital replicas.
United States Copyright Office — Copyright and Artificial Intelligence — Human authorship, copyrightability, AI-generated expression and generative-AI training.
Japan Agency for Cultural Affairs — General Understanding on AI and Copyright in Japan — Japanese interpretation of existing copyright law as it relates to AI development and generated output.
Cyberspace Administration of China — AI-Generated Synthetic Content Labelling Measures — Chinese rules concerning explicit and implicit labelling of generated text, images, audio, video and other content.
Government of Canada — Copyright in the Age of Generative Artificial Intelligence — Consultation and stakeholder findings covering training, authorship, ownership, infringement and liability.
New Zealand Ministry for Culture and Heritage / Creative New Zealand — Research and guidance concerning generative-AI adoption, intellectual property, transparency and cultural responsibility.
The Beatles — “Now and Then” — Official documentation concerning MAL-assisted separation and recovery of John Lennon’s existing vocal performance.
Library of Congress — Magnetic Tape Preservation / Sticky-Shed Research — Scientific and preservation research concerning deteriorating magnetic audio carriers.
International Association of Sound and Audiovisual Archives — Professional guidance concerning analogue tape playback, alignment, azimuth and preservation transfer.
iZotope RX — Professional Audio Restoration — Machine-learning-assisted restoration, noise repair, source separation and spectral-processing tools.
Holly Herndon / Mat Dryhurst — Holly+ and related projects — Research and artistic experimentation concerning authorised synthetic voice, consent and data governance.
Bandcamp — Keeping Bandcamp Human — January 2026 policy prohibiting music generated wholly or substantially by AI and prohibiting AI impersonation.
LabelRadar — Generative AI Submission Policy — Submission disclosure framework distinguishing generative material from technical processes such as mastering, cleanup and stem separation.
Globetrotting Music — Artist Submission Policy — Example of an independent label explicitly declining AI-generated submissions.
Deezer — AI Music Detection, Upload Volume and Streaming Fraud — Statistics concerning fully generated uploads, listening share, AI detection, recommendation policies and fraudulent streaming.
Spotify — AI Protections for Artists, Songwriters and Producers — Policies concerning synthetic impersonation, spam, consent and AI-related disclosure.
IFPI and international music organisations — AI-Assisted / AI-Generated Labelling — 2026 terminology and chart principles distinguishing different levels of generative involvement.
Universal Music Group / Udio — Licensed generative-AI platform agreement following settlement of copyright litigation.
Warner Music Group — AI Principles and Licensing Partnerships — Licensed models, compensation and artist/songwriter choice concerning identity and voice.
KLAY / Universal / Sony / Warner — Examples of cross-major licensing arrangements involving generative-AI music systems.
CISAC / PMP Strategy — Global Economic Study — Scenario modelling concerning the potential impact of generative AI on creator revenues.
Moreish Studios, Melbourne — Public recording, mixing and mastering rates used in the economic comparison.
Suno — Subscription Pricing and Suno Studio — Current subscription model and generation capacity used for the economic comparison.
LazyboyProactive — Official Project History — Background on Ankit and the long-running electronic music project LazyboyProactive.
Gordon Lawrence Taylor — Voice In The Void Infectious Unease Radio / Subterranean Zone Radio
Supporting independent underground music, artists and labels across borders, languages, scenes and cultures.
