AI Music Is Here. The Real Fight Is Over Who Gets Erased

AI-generated music used to feel like a novelty. Strange, clever, occasionally creepy, but still easy to keep at arm’s length. That phase is over. In April 2026, Deezer said fully AI-generated tracks made up 44% of all new music uploaded to its platform , nearly 75,000 tracks a day . And as of May 2026, Spotify and Universal Music Group have announced a licensing deal for AI-generated covers and remixes from participating artists. So AI music is not coming. It is already here , uploading, remixing, licensing itself, and quietly filling the pipes. For musicians, that is unsettling. I do not think there is much point pretending otherwise. If songs can be generated cheaply, endlessly, and at that speed, then human-made music is entering a very strange new competition. But the problem is not simply that AI can make music. The more interesting question is what happens around the music. Who made it? Who shaped it? Who gets named? Who gets paid? Who quietly disappears? That is where the real argument begins. AI can help people practise, write, produce, restore recordings, and get past the blank page. It can also generate synthetic tracks at an industrial scale, imitate artists, blur authorship, and make the human part of music harder to see. So the question is not “is AI music good or bad?” That is too easy. The better question is: when AI enters music, who stays visible, and who gets pushed out of the frame? The human role is the difference This is where a lot of the debate gets annoying. People talk about AI in music as if it is one object: one movement, one threat, one miracle, one thing you are either excited about or morally opposed to. But the more useful question is not what the technology can do. It is what happens to the human role once the technology is involved . A tool that helps someone practise piano is not the same as a tool that floods streaming platforms with anonymous synthetic songs. A stem separation tool is not the same as a fake band with fake press photos. A songwriting assistant is not the same as a model accused of training on copyrighted recordings without permission. The technology overlaps. The human role does not. I am not anti-AI. I use AI. I find it useful, strange, occasionally brilliant, and occasionally like being trapped in a lift with someone from marketing who keeps saying “content at scale”. But I am a musician, so I am suspicious of anything that treats music as output first and human expression second. Music is not just audio. It is taste, memory, friction, bad takes, strange accidents, tiny mistakes, half-working cables, rehearsal rooms, ridiculous arguments, and the bit of the song that only exists because someone played it wrong once and everyone decided to keep it. AI can help with that world. It can also sand it down until there is nothing left but playlist-friendly beige. When AI helps people learn music The useful side of AI in music is not always the glamorous bit. Not fake bands, synthetic singers or “write me a hit song in the style of everyone who ever got underpaid by a record label”. It is often the smaller stuff that helps someone practise without giving up . Anyone who has tried to learn an instrument knows the moment. You can play one part. You can sort of play another. Then you put them together and suddenly both hands behave like they have never met. That is where AI could actually help. Not by magically turning anyone into a musician. Any app promising that should be taken outside and made to practise scales. The useful version is more practical: guide me through the song, notice where I’m stuck, and let me ask for help before I abandon the whole thing . There are already plenty of apps using recognition and feedback in music learning. I looked at familiar names like Simply and Yousician, but what stuck out to me was the newer wave of tools that put something closer to an AI tutor inside the practice process . ROLI’s AI Music Coach is one example, although it sits inside ROLI’s own hardware ecosystem. Artie, an AI piano teacher app made by ArtMaster, was another example of that tutor-style approach, and the one I could test with a normal keyboard setup. What interested me when I tried it was not the grand “AI will replace teachers” idea, because I do not think it will. It was the smaller thing: the learner can ask questions, get guidance and keep practising when they would normally stop. None of this replaces a good teacher. But most beginners are not sitting next to a teacher every time they practise. They are alone with an instrument, a phone, and the growing suspicion that maybe they are not “a music person” after all. If AI can help there, not by replacing the learner but by keeping them in the process , that feels useful. When AI stays in the assistant role AI can also help musicians make things , which is where I am much less suspicious of it. Songwriting is not always inspiration arriving from the clouds. Sometimes it is staring at the same four chords for forty minutes wondering whether you have written something moving or just accidentally reinvented a Coldplay B-side. Sometimes you need a different chord, a lyric angle, a bassline idea, a drum pattern, a texture, a reference, a way out. This is where the assistant role makes sense. Tools like AudioShake , Hookpad , iZotope Ozone and ChatGPT are doing very different jobs, but they all fit into that assistant role. One can separate stems. Another can help with harmony or rough song ideas. Another can get a demo closer to mastered. Another can help with lyrics, structure or the blank-page problem. None of that makes the tool the artist. It gives you material, options, shortcuts or a way back into the work. But the musician still has to choose. And that is the bit people keep underestimating. Taste matters. Judgement matters. Knowing what to delete matters. Knowing when the rough version is better matters. Knowing that the bass is too loud even though you love it matters. AI can give you more material. It cannot tell you who you are , which is annoying, because most musicians would quite like to know. When AI starts pretending to be culture The problem is not just that AI can generate songs. It is that it can generate them endlessly, cheaply and with enough surrounding imagery and backstory to make them look like part of a real music culture. One case that shows why this gets uncomfortable is The Velvet Sundown , an AI-generated band that reportedly gained more than a million Spotify streams before many listeners realised there were no real musicians behind it. That is different from someone generating a joke track or a quick demo. This was not just an AI song. It was a band, an identity, a story, a thing listeners could attach themselves to. We are used to pop music being constructed. We are used to studio projects, fictional bands, anonymous producers and carefully polished artist images. But there is usually still a person somewhere: a singer, a writer, a producer, a performer, someone who can be interviewed, misunderstood, blamed, adored or photographed looking tired outside a hotel. With AI acts, that relationship starts to collapse. Who are you listening to? Who gets paid? Who gets credited? Who is accountable? And should listeners be told when the artist they are hearing does not exist in the normal sense? I think they should. Not because AI music should be banned, but because AI songs and AI artists are not quite the same problem . AI songs raise creative questions. AI artists raise trust questions. Listeners deserve to know the difference between a human artist using tools and a synthetic act manufactured by prompts, models and marketing. Pretending those are the same is not innovation. It is just bad labelling. Copyright, consent and the awkward question of payment The legal fight around AI music is not a boring side issue. It is the centre of the whole thing. In June 2024, the RIAA announced copyright infringement lawsuits against Suno and Udio, alleging that the companies copied and exploited copyrighted sound recordings without permission to train their music-generation services. That is the argument underneath everything. AI companies talk about creativity, access and new tools. Artists and rights holders ask a simpler question: did you train your system on our work without asking? That answer matters. If AI models are built on decades of recorded music, songwriting, production, performance and style, then the people who made that music cannot simply be treated as raw material. Human creativity is not a free buffet for companies that want to sell the leftovers back as the future. This is why the Spotify and Universal deal matters. It suggests one possible route: AI music built around permission, participation, credit and payment. Spotify describes the deal as a way for artists and songwriters to share directly in the value created by AI-generated licensed covers and remixes on the platform. That does not solve everything. It does not tell us what happens to independent artists. It does not fix style imitation. It does not answer whether listeners will care. It does not magically make AI music fair. But at least it points to the right principle: if someone’s voice, songs, recordings or style help power the system, they should have a say. So where do the humans fit? I do not think AI is going to kill music . People said recording would kill live performance. Synthesizers would kill musicianship. Drum machines would kill drummers. Sampling would kill originality. Auto-Tune would kill singing. Streaming would kill albums. Some of those fears were overblown. Some were partly right. Some were just musicians being musicians, which means suspicious, dramatic and occasionally correct. But music kept mutating. Musicians are good at stealing tools from the future and making them emotional. AI will be no different. Use it where it helps. Use it to practise, learn, sketch ideas, build demos, understand chords, or write boring admin emails so you have more time to do the actual music. But do not outsource your taste. That is the bit that matters. If everyone can generate fifty tracks before breakfast, the valuable skill becomes knowing which one, if any, deserves to survive lunch. If everyone can make polished audio, rough edges become more interesting. If machines can imitate style, lived experience matters more, not less. The problem with AI music is not that it uses machines. Music has always used machines. Pianos are machines. Tape machines are machines. Synths, samplers, drum machines, DAWs and Auto-Tune are machines. The problem is when the machine stops being an instrument and becomes a substitute for the person. So maybe the question is not whether AI belongs in music. It already does. The question is what kind of musical future we are building with it. One where more people can learn, create, practise, produce and experiment? Or one where music becomes endless synthetic wallpaper, endlessly generated, endlessly recommended, and increasingly detached from the awkward, brilliant, inconsistent humans who made us care about it in the first place? That is the fight. Not humans versus machines. Musicianship versus infinite content. \