AI makes music. But for what purpose?

Powerful tools, fragile illusions, and a disturbing perspective on the future of copyright that no one seems to notice.

Generative AI tools for music

Suno, Share, Wondera: names that until recently were the exclusive preserve of tech enthusiasts have now entered the everyday vocabulary of anyone with even a passing interest in music. Generative artificial intelligence applied to music is a concrete, accessible, and often surprising reality. But expectations, tensions, and paradoxes are building around these tools that are worth examining carefully and without prejudice, without avoiding considering the future implications, however unpleasant.

Moods and expectations

I happened to frequent some online communities dedicated to these tools, and I took the opportunity to try to understand the atmosphere among users of these new tools and their detractors. After a few weeks of perusing the forums, I was able to draw some interesting observations: reading the posts and listening to the songs published by users, at least three very distinct user profiles emerge.

There are those who use the platform purely for fun, without pretensions, as a new and fascinating game to play in their free time, and this is perhaps the healthiest and most honest use that can be made of it.

There are those who, with carefully crafted lyrics and sophisticated sound choices, seem to follow a credible artistic path for which they have already laid the foundations before the arrival of the IA and which therefore uses the tool as a bridge between idea and realization, bypassing the expensive and lengthy production processes.

And then there's a huge amount of sloppy production, lyrics straight from elementary school essays, copied and re-copied genres, individual users posting dozens of songs a day; many don't even make the effort to write the prompt themselves, preferring to have it prepared by other platforms like chat. gpt, Claude o Gemini.
In short, a lack of imagination and a lot of presumption, highlighted above all by a recurring question which, contrary to what one might expect, is not "how can I improve this song?" but rather "how can I increase my listenership?", "how can I monetize it?", "which platform is best for me to distribute on?"

It's the paradox of the music slot machine: a tool designed to break down creative barriers is primarily used as a shortcut to success and effortless earnings. It's a shame that these platforms, at least currently, offer very little control over the final result. Once the song is generated, tweaking the errors and the structure of the composition is a long, frustrating, and often fruitless process. If it turns out well, it's a stroke of luck; otherwise, it's better to hit the "generate" button again and try again, just like a slot machine. And as with any slot machine, the illusion of an easy win is the real selling point.
The problem isn't the instrument. It's forgetting that throughout the history of music, success has always been the sum of talent, skill, years of work, and a good dose of luck, even when there were great songwriters, great producers, and major record labels behind it. Reducing all this to a text prompt is naive, to say the least.

Adding to the interesting picture are the reactions of many "traditional" musicians who, in these discussion forums, seek to highlight the limitations of AI-generated music. The responses they receive are almost all similar: "Now that everyone can make music, do you feel threatened?", "Your time is up," "You're no longer the only ones who can make music." This narrative of revenge betrays something unresolved: it's not enthusiasm for a new creative instrument, but rather a prior frustration with a musical world perceived as elitist and unattainable, as if being a musician were an innate privilege. But being a musician isn't the prerogative of a lucky few; it's study, sacrifice, practice, years spent making mistakes, and moments of discouragement and the desire to give up, just like any other profession that requires training and commitment. In short, it's certainly not easy, but it's within the reach of all mere mortals.

Legal disputes and the sustainability of proliferation

In the meantime, the sector has already had to deal with the legal reality. In June 2024Sony Music, Universal Music Group, and Warner Music Group have sued both Suno and Udio, accusing them of massive copyright infringement for training their models on copyrighted recordings without authorization.
The lawsuits were later resolved in settlements: Universal Music Group reached a settlement with Udio in October 2025, followed by Warner Music Group, which simultaneously also signed a settlement with Suno in November 2025.

The terms of the Suno agreements stipulate that downloads will remain available to paying users, subject to a monthly limit, and introduce new experiences that will allow users to create content using the voices, compositions, and images of Warner artists who choose to participate. In short, the major labels have stopped fighting and have chosen to sit down at the table, which isn't necessarily good news for those hoping for a curb on proliferation.
And while all this is being discussed, the real problem is quietly building elsewhere. In the 90s, according to data cited by Wikipedia In the album era, approximately 35.000 albums were released annually worldwide, a number filtered and selected by the industry. In 2016, with the advent of digital, the figure was already around 158 new songs per day. Today, according to statements by the CEOs of some major music labels, Billboard, are published approximately 100.000 songs per day on all streaming platforms. Deezer he estimated that in 2025 he will receive approximately 50.000 AI-generated songs every day, accounting for 34% of all uploads.

The consequences are already visible and quantifiable. According to the Luminate 2024 reportOf the 202 million songs available on streaming platforms, 175,5 million don't reach the minimum threshold of 1.000 annual streams required by Spotify to generate royalties: 86,88% of all music on the platform doesn't generate a cent. Millions of files occupy servers, consume bandwidth, and require management: at zero cost to those who upload them, at a real cost to the platforms and, indirectly, to the entire music ecosystem.

Doubts about copyright

But there is an even more subtle aspect, which almost no one has yet stopped to reflect on: the problem of the saturation of musical creative space.
The possible combinations of notes, rhythms, and harmonies are mathematically finite and not infinite as one might think. This was demonstrated in concrete terms by the programmer and copyright lawyer already in 2020. Damian Riehl with a simple yet disruptive reasoning: "if melodies are essentially mathematical combinations, and the combinations are finite, sooner or later someone will have already written them all." Together with the musician and programmer Noah Rubin he transformed this thought into a real and functional tool by creating an algorithm capable of generating 300.000 melodies per second to be catalogued and released into the public domain.

His logic was defensive: in the copyright system, whoever first deposits a melody becomes its owner, and anyone who uses it in the future risks a lawsuit for infringement. By anticipating everyone and making the entire possible melodic space available to everyone, Riehl sought to preemptively neutralize this threat, reclaiming that territory from anyone who might want to use it as a legal weapon.

Now let's project this scenario into the era of generative AI. Millions of users press a button every day and get complete songs; melody, harmony, lyrics, and arrangements, all in a matter of seconds. Most of these songs are uploaded to platforms, and a portion are even registered for copyright protection. Not because there's any real artistic value to protect, but because the process is automatic, or nearly so, and costs little or nothing.
The paradox that's quietly unfolding is this: a human composer who writes an original melody tomorrow, the fruit of his own creativity and artistic journey, might be denied the right to register that piece because that sequence of notes has already been recorded. Recorded by whom? By a Suno user who, six months earlier, had automatically generated thousands of songs, which he'd never actually listened to, and likely doesn't even remember their existence.
The burden of proof, in such a dispute, would fall entirely on the human composer: to demonstrate that their creation is independent, original, and not derivative. A Kafkaesque legal battle against a machine-generated piece, filed by someone who doesn't even know they made it.
What AI risks doing, in short, is not only flooding platforms with junk music but colonizing the future of human creativity, preemptively occupying a space that no one has yet explored.

Self-referential entertainment

There's one final scenario, perhaps the most radical, worth at least sketching out. Generative AI applied to music won't stop at the proliferation of productions lacking artistic value. Once the fortune-seeking binge wears off, it could degenerate into a self-referential mechanism where everyone creates their own music, completely eliminating the need for an artist as an intermediary between an emotion and its sonic translation. 

The consequences would be more profound than any copyright issue: music has always functioned as a bridge between those who create it and those who listen to it, between different cultures, between generations. One of the most formative experiences in a listener's life is being surprised by something unexpected, a previously unexplored genre, an artist far removed from their usual tastes, and recognizing something of themselves in it; it's a way of discovering that someone on the other side of the world has experienced the same thing as you and has described it better than you could have. 

A system that knows you too well can't offer you this. It only gives you back what you already are: each locked in their own perfect sound bubble, devoid of the thrill and risk of discovery. 

The music entertainment business as we know it would be destroyed. But the greatest damage would be anthropological.

Conclusions

The point isn't to be against AI in music. The point is to ask ourselves what we really want to do with these tools. Offer the ability to produce a song even to those with interesting ideas and a pressing need for expression, yet lacking familiarity with musical instruments and/or production software? Surrender to a despotic future, delegating the choice of what to listen to to algorithms, locking ourselves in our own sonic bubble and forever renouncing the joy of discovery? Or, as seems to be already happening before our eyes, try to win at the slot machine, further simplifying a paradigm already in place in the world of traditional production, where identical songs are churned out according to the so-called "success" canons? 

A concept perfectly summarized by Madame in the song “Mai più” included in the recently released album:

“The conveyor belt songs, The Horror Factory, hoping for a hit.”

NOTES

The approach of deserves a separate discussion Moises, which differs substantially from Suno and Udio. Instead of generating complete songs from a text prompt, Moises focuses on generating and manipulating individual tracks—a bass line, a drum kit, a guitar—to integrate into an existing musical project. It's a tool designed for those who already know how to play an instrument: musicians who want to speed up their workflow, add elements they haven't mastered, or experiment with new sonic directions without distorting their creative process. An approach that's more collaborative than substitutive, and arguably more honest to what music should be. 

Pasquale Tarricone