Music and AI, where do we stand?

Every time a new technology is introduced in the field of art, opposing factions emerge, which can be traced back to the two categories proposed by Umberto Eco in his famous essay Apocalyptic and Integrated (1964). On one side, the totally opposed, and on the other, the supporters. One example above all could be photography: when it was introduced, some "apocalyptics" – including C. Baudelaire and Walter Benjamin – speculated that such a powerful technology, capable of perfectly reproducing an image identical to reality, would mean the end of painting and illustration. But as we know, that was not exactly the case, and photography became a new art form. Resistance to technology has always been particularly strong and hostile in the field of music; just think of how much effort electronic music had to make before being recognized as having the same dignity as "music played with instruments." More recently, the industry has also had to face numerous technological changes in its consumption and distribution methods, which have generated significant ethical and economic conflicts. Consider the transition from CDs to MP3s and the advent of Napster and other peer-to-peer sharing systems, which led to the proliferation of piracy. Or the more recent total dematerialization of music and the new agreements between record labels and streaming platforms. The way the public consumes music is something often ignored in discussions about AI, but it is just as important as the way music itself is created.

The First Steps of AI in Music Creation

The first experiments on the musical potential of generative AI were largely received as harmless amusements. Consider Daddy’s Car, a Beatles-style song from almost ten years ago (back in 2016) in which, for the first time, music was composed by the AI program Flow Machines. Many people probably never heard of it, as at the time these "songs" were still considered mere experiments bordering on pranks. Four years later, during the peak of the pandemic, the so-called jukebox samples generated by OpenAI emerged. These were essentially one-minute stylistic exercises that blended musical genres and artist styles, with titles like Folk Rock in the Style of Simon and Garfunkel. Due to the use of real artist names, these samples are no longer available on SoundCloud today.

Deepfake

Over the past five years, things have become significantly more serious. Among the cases perceived mostly negatively is the phenomenon of so-called deepfakes, which continue to proliferate on YouTube and TikTok. These are essentially songs that use artificial intelligence to recreate the voices of famous singers. The most striking case was Heart On My Sleeve, a fake song by Drake and The Weeknd released in 2023 by the anonymous producer Ghostwriter. The similarity of the voices and the fame of the two artists allowed the fake track to accumulate over 20 million plays before Universal managed to have it removed from all streaming platforms. Other notable experiments include Jay-Z reciting Shakespeare, Kanye West singing Somebody That I Used To Know, and even an entire unreleased Oasis album – Aisis, The Lost Tapes Vol. 1 – which was actually just an album by the band Breezer with the singer’s voice replaced by an AI-generated Liam Gallagher voice.

If the previous cases can be considered akin to counterfeit goods, like a fake Rolex or a simple imitation, there are other cases that we might see as more or less virtuous, or at least where the legal issue does not exist and the ethical debate is more nuanced. One example is Grimes: she started off lightly with the project AI Lullaby – which allowed her voice to be combined with ambient sounds to help children sleep – and later decided to make her voice available so that anyone could use it to create new AI-generated songs, thus developing a potentially infinite continuum. As extreme as it may be, no one can contest this choice from an ethical or legal standpoint, as in this case, it is a free expression of her personal artistic vision, and she alone holds the rights to her voice and music.

The most striking virtuous case to date is certainly the 2024 release of a new Beatles song – Now And Then – which even won a Grammy in the Best Rock Performance category. Compared to Daddy’s Car, we are literally on another level (for those interested, there is a mini-documentary summarizing the song’s history): originally, it was an old tape with a home recording by John Lennon, whose quality was too poor to be turned into a finished song. The remaining Beatles had already tried in the past but abandoned the attempt due to the poor audio quality. However, thanks to a new AI-based technology – the same used by Peter Jackson for the documentary Get Back – it was possible to isolate the audio and achieve a quality acceptable for publication. This operation could be considered akin to cultural heritage restoration. And if the final result is a great new song from the greatest band in history, the overall judgment can only be positive.

Pros and Cons of AI Music

A possible future degeneration of the practices discussed so far could be getting carried away and falling into poor taste by publishing millions of new songs by deceased singers generated by AI. Said like this, it might seem dystopian, but considering the public's interest in posthumous albums and record labels' tendency to scrape the bottom of the barrel to release new music from deceased artists, this issue shouldn't be underestimated. Just think of how many albums by Jeff Buckley were released after his death or the low-quality sketches included in the so-called solo album of Kurt Cobain. Sure, in these cases, at least the songs had historical value because there were tracks intentionally recorded by the artist, but if even that basis were missing, what value would the songs have? Hard to say. And how would we judge the situation if artificial intelligence were used to "help" singers who are still alive but had to stop singing for health reasons? Would it be a positive or negative thing if someone like Steven Tyler, who was forced to leave the stage, decided to train AI to reproduce his voice and create a new Aerosmith album today? Again, it's difficult to say, but at this point, we are probably not too far from finding out.

The Issue of the Public and Functional Music

It seems that a large portion of the general public today is no longer very interested in knowing who the artist they are listening to is but is more concerned with the overall mood being conveyed. This is why playlists generated directly by platform editors or algorithms are proliferating, particularly those that offer so-called "functional music," that is, music to listen to while performing certain activities like running, studying, or sleeping, or those that evoke a certain mood like relaxation or melancholy. From here, agreements have emerged between two major record labels - Warner and Universal – and the company Endel, which is practically a leader in the field of functional music creation aimed at reducing stress through AI. On Endel’s Spotify profile, there are entire albums with self-explanatory titles such as The Most Beautiful Calm or Antarctican Dream Machine (Relax Soundscape).

The Issue of Ghost Artists

While AI-produced music by Endel is openly declared, the same cannot be said for music produced by so-called ghost artists on Spotify. The issue was recently brought to light by journalist Liz Pelly in a newly published book titled Mood Machine - The Rise of Spotify and the Costs of the Perfect Playlist. The investigation revealed that Spotify uses external companies - still linked to the platform - commissioning them to create instrumental tracks that are then placed in the most listened-to playlists under fake artist names to generate royalties, which are later redistributed to the platform itself. At the moment, there is little certainty about how these songs are created or who their authors are, but the possibility that AI plays a role cannot be ruled out. After all, the main company behind these fake artists – Epidemic Sound – has already openly stated that it allows its composers to use AI.

The Regulatory Framework – Why Are Artists Protesting?

In light of what has been discussed so far, are artists right to protest against AI interference? It depends on what their concerns are, but it is true that there is still a lot of work to be done at the regulatory level. In recent days, a proposed law by the UK government, aimed at allowing AI training with copyrighted cultural products, has sparked significant debate. The backlash led to the creation of a collective of a thousand musicians—including Kate Bush and Damon Albarn of Blur—who released a silent protest album, a record symbolically composed of twelve empty tracks, whose titles, when put together, form the sentence: The British Government Must Not Legalise Music Theft To Benefit AI Companies. In the U.S., the first state to pass a law protecting musicians from AI-related abuses was Tennessee, where last year the so-called Elvis Act (Ensuring Likeness Voice and Image Security) came into effect. The issue is that, being a state law rather than a federal one, it has no validity beyond the state's borders. At the federal level, the closest thing to protection against AI abuse is an advertising law that prohibits the use of someone's voice in advertisements without their consent. As for the European Union, the gradual introduction of the AI Act began this year, but the road is still long, and its impact and effectiveness remain to be seen.

Expert Opinions on AI Music

According to Francesca Lagioia—a professor of Legal Informatics, AI and Law, and Ethics for AI at the University of Bologna—art created by artificial intelligence can never replace or be compared to that created by humans because AI's creative capacity is only «combinatorial», meaning it is merely capable of reassembling pre-existing ideas but cannot generate a truly new idea. This is reassuring, though it opens up a world of philosophical speculation about what actually constitutes a new idea. When shown an AI-generated animation, the master of Japanese animation, Hayao Miyazaki, expressed disgust—not out of principle, but because the creation resembled a parody of a person with motor disabilities. In his view, only a machine with no understanding of human suffering could conceive such an aberration. While pure, Miyazaki's perspective may seem a bit too innocent to our cynical Western eyes to be entirely convincing. A more concrete and reassuring view comes from music critic Simon Reynolds, who insightfully pointed out something seemingly obvious: AI «does not listen» to music—it has neither ears nor a body—but all music affects us through our sensory and kinesthetic faculties. AI cannot fully understand music because it lacks our physical and neural experience.

Finally, the most complex and compelling response—precisely because it is more visceral and human—came from Nick Cave a few years ago when replying to a fan who attempted to write a song “in the style of Nick Cave” using AI: 

«Writing a good song is not imitation, replication, or pastiche; it is the opposite. It is an act of self-annihilation that destroys everything one has previously tried to create. These are dangerous departures, heart-stopping leaps that propel the artist beyond the limits of what they recognize as their known self. This is part of the authentic creative struggle that precedes the invention of a truly valuable piece of work; it is the breathless confrontation with one's own vulnerability, one's own danger, one's own insignificance, contrasted with a sense of sudden and shocking discovery; it is the redemptive artistic act that moves the listener's heart, where the listener recognizes in the inner workings of the song their own blood, their own struggle, their own suffering. This is what we humble human beings can offer, which AI can only imitate—the transcendent journey of the artist constantly grappling with their own flaws. This is where human genius lies, deeply rooted within, yet reaching beyond, those limitations.»