It’s Friday. You wake up and open your music streaming platform of choice to listen to new releases this week. They come in an automated playlist generated based on your listening habits and favourite music.
You put it on shuffle and, after a couple of songs, one plays where you don’t recognize the voice, but sounds generic and, let’s be frank, not great.
When you check the details you recognize the band or artist’s name, but the album cover screams ‘generated by ChatGPT’ and the song was, in fact, also generated by artificial intelligence.
First came AI-generated bands like The Velvet Sundown, trying to pass themselves off as original and authentic music, despite being generated out of existing references. Now, AI-generated songs are sneaking in the profiles of human artists, more or less known, like Erreway, and often deceased, like Angela Lansbury.
As part of an investigation into generative artificial intelligence, The Atlantic examined some of the most popular AI music generators, like Suno and Udio.
They found large music datasets, with more than 21 million recordings, used to train these music generators without any consent from musicians, producers or songwriters, raising serious questions about copyright, licensing and transparency that remain unanswered.
But because things can always get worse, an evolution of this nightmare is all the AI-generated songs sneaking in the discography of existing artists.
The problem is not only that someone used AI to generate a song. It is that this song can appear to belong to someone who never recorded it, sometimes within the catalogue where legitimate work is hosted.
This music is often not clearly labelled as AI-generated, but since they are above albums and singles you know and love, the algorithm take them as official and thinks you will love them, too.
How many songs are too many?
B.A.I. (before artificial intelligence), Variety reported in 2022 that 100,000 new songs are uploaded to music platforms daily.
A.A.I. (after artificial intelligence), I found reports that 50,000 fully AI-generated tracks are uploaded to Deezer daily and, even more alarming, 97% of their users can’t tell the difference.
Music streaming services rely on a combination of algorithms, metadata, listening behaviour and editorial curation to help navigate this catalogue.
Your listening history tells the platform what you like, other people’s listening behaviour helps it predict what you might like next, metadata helps organize the catalogue and algorithms then use all of that information to recommend music, generate playlists and decide what appears in front of you next.
This system can work remarkably well when the underlying information can be trusted, but AI changed that assumption.
If the number of songs being uploaded keeps growing, and some are generated by AI and uploaded under legitimate artists, it becomes a problem of reliability for these platforms, different from simply having too much music.
The internet always created more content than we could consume. More than 400 hours per minute were uploaded to YouTube in 2019.
Search engines helped us find websites, social media feeds helped us discover artists and streaming platforms helped us navigate enormous catalogues of music, movies and television.
Algorithms became increasingly important because there was simply too much to sort through manually.
But generative AI takes that abundance to another level where the amount of content competing for attention can grow much faster than our ability to consume it.
And that is where human curation becomes important.
The more content AI generates, the more we may need actual humans to tell us what is actually worth our attention.
Could the Radio Star kill AI?
Before streaming platforms and algorithms, radio stations and music magazines played an important role in helping audiences navigate what new music was available.
Radio DJs could decide which new songs deserved airtime, introduce listeners to artists they had never heard before and provide context around music that might have been lost among everything else being released.
Music magazines did something similar, with writers and editors deciding which albums to review, which artists deserved a feature and which new releases were worth telling their readers about.
And don’t forget about MTV, which in its original form made a whole new business model out of curating music videos 24/7.
Their decisions were influenced by an industry of record labels, commercial interests or cultural biases, and getting music heard often depended on gaining access to a small number of gatekeepers.
But their role was not simply to fill space or keep people entertained. They helped audiences navigate an increasingly crowded music landscape by making choices about what was worth listening to.
In fact, we have always relied on other people to help us make these decisions. We ask friends what music they are listening to, read reviews before listening to an artist, follow journalists whose recommendations we trust, listen to DJs who introduce us to artists we have never heard before and buy albums or tickets to a concert because someone whose taste we know recommends them.
The internet gave us new ways of doing this, introducing algorithms that could perform some of the same filtering on a larger scale.
But algorithms and humans are not filtering for the same thing.
An algorithm can look at my listening history and determine that I am likely to enjoy a particular song. It can identify that people who listen to one artist often listen to another, or that a particular song is becoming popular among people with similar listening habits.
But algorithms are poorly equipped to identify whether a song is authentic, actually belongs to the artist whose name appears on the metadata or if there is a reason I should care about it beyond the fact that it resembles something I have already listened to.
But, as artificial intelligence makes the amount of available content grow disproportionately, a human curator can make those judgments.
A music fan who follows a particular artist or genre can recognize when something does not make sense. A music journalist can listen to a new release and provide context about where it fits within an artist’s work. A curator can decide that something is worth recommending not simply because an algorithm has identified a pattern, but because they have actually listened to it and believe it deserves attention.
If you want to listen to some of the 100% AI-free playlists I made, click here or find me on Spotify and Deezer.
Dust off the vinyl player
It’s not a coincidence that these ideas come at a time of renewed interest in physical formats. As streaming has made almost every song instantly available and algorithms have increasingly shaped what we discover, buying a vinyl record, a CD or even a cassette can feel like a return to a more curated way of experiencing music.
You choose an album rather than being continuously presented with something else to listen to, and the physical object itself provides a different kind of context around the music.
The return of physical music and the renewed importance of human curation offer an alternative to an environment where algorithms are constantly deciding what comes next.
From gatekeepers to curators
If the platform cannot guarantee that every recommendation is authentic, then having people whose job is to listen, verify, contextualize and recommend becomes more valuable.
Traditional gatekeepers, before the rise of the Internet, decided who got published, promoted and found an audience, often excluding those they deemed not commercially or culturally valuable.
New platforms online changed that by allowing anyone to publish and distribute their work without permission.
That democratization is worth protecting.
But removing gatekeepers does not mean removing the need for curation.
The difference is that curation does not have to mean deciding who is allowed to participate, just helping people navigate what is already there.
Making it possible for anyone to put music online does not mean treating everything as equally deserving of our attention.
This distinction matters because there is a tendency to see curation as something that belongs to old media, but a Reddit community, the YouTube channel that consistently recommends good albums, newsletters that filter through new releases or a social media personality sharing songs from around the world, these are all forms of curation.
Digital platforms have made curation invisible, prioritizing automated recommendation systems, and the wave of AI-generated content we live under makes that more important to question.
Platforms need to curate, too
Streaming services have a responsibility. Spotify switched to AI-curated playlists while axing thousands of jobs.
That is no coincidence.
These platforms need to understand the value of human staff, even if it’s just to protect their bottom-line before AI makes their products unusable.
If a song is uploaded under the name of an established artist, the listener should not have to figure out on their own whether it is legitimate. Platforms should have systems to organize artist profiles, manage metadata, recommend songs and identify copyright issues. Verification and labelling should be part of that same infrastructure and prioritized as the amount of AI-generated content increases.
Curation should also not be about finding the most popular or engaging music. An algorithm can become very good at identifying what will generate another listen, without necessarily considering whether that content is authentic, meaningful or even particularly good.
But humans can introduce a different set of priorities. A curator can decide that an obscure album deserves to be heard even if it has not generated much engagement.
Those decisions are difficult to reduce to an algorithmic prediction about what someone is likely to click on.
Scarcity in abundance
The next stage of digital media will not be about generating more content, we have plenty of it, but figuring out how to navigate all of it.
Finding something good becomes harder when there is so much more to choose from. And finding something authentic becomes tougher when the platforms we rely on cannot always tell the difference.
We shouldn’t go back to a system where a small number of institutions decide what everyone gets to listen to, but we need to recognize that abundance creates its own kind of scarcity.
When there is more content than anyone can possibly consume, the scarce resource becomes attention, and the people who help us decide where to spend it become more important.
We still need algorithms to help us navigate abundance. They can surface things we would never have found on our own, make enormous catalogues usable and introduce us to content based on patterns that humans could never process. But we also need to put more value on the people who can make sense of what those algorithmic systems surface.
The easier it becomes to make something, the less valuable is having more of it. What becomes harder, and therefore more valuable, is discovering something that deserves our attention.
We built algorithmic systems to help us find more music. Now we need human systems to help us decide what is actually worth finding.
