Last week, Swedish performing rights organisation Stim held a livestreamed presentation about its perspective on AI in the music industry, ahead of its upcoming annual general meeting. In September last year, Stim proudly declared in a press release that it was launching “the world’s first AI license for music […] developed in close dialogue with rights holders and industry stakeholders”.
On the one hand, I appreciate that Stim is trying to be proactive in this changing technological landscape, and I understand the need for a license like this. On the other – and I realise that this probably comes from a deep feeling of unease about the growing proliferation of AI slop – it also feels kind of like negotiating with a hostile, bad-faith counterpart.
Last August – incidentally the month before Stim introduced its AI license – I published my first essay about artificial intelligence in music, generative AI tools and large language models. It is a subject that engages me and many others, not only creative professionals, and it has become my most-read blog post to date.
During Stim’s presentation last week, a lecturer from a Norwegian university discussed the current surge of AI in the context of 20th century technological shifts that seemed to threaten musicians and the music industry, but ultimately did not: music recording, radio broadcasts, amplified instruments, synthesisers and drum machines, and computer-based music creation.
The lecturer’s ultimate argument was that rights organisations simply need to establish ground rules for companies and platforms to follow. My immediate (and lingering) thought was (and is) that it is already too late for that. Never mind that some more recent AI start-ups might be interested in signing license agreements with rights holders; the biggest and baddest actors have already illegally trained their models on copyrighted music.
Even if these companies were to sign license agreements after the fact, would they agree to retroactively – and proportionally – reimburse the creators whose works they have illegitimately trained their large language models on? The fact that they broke the law to begin with, and now argue that they had the right to do so all along, makes me doubt the companies’ intentions.
Small Victories?
In November 2025, Stim’s German counterpart GEMA announced a court victory against OpenAI, the creator of ChatGPT. The court found that ChatGPT contained and could reproduce copyrighted works obtained by OpenAI without license. (Also, as a side note, GEMA refers to an AI license they launched a full year earlier than Stim did theirs, lending doubt to Stim’s claim to having the “world’s first” AI license for music…)
In that November press release, GEMA also mentioned an ongoing lawsuit against generative AI company SUNO Inc., creator of the music generator Suno AI. German law firm Härting detailed the lawsuit in their own press release back in January 2025. GEMA, the law firm wrote, had demonstrated the ability to reproduce popular songs such as Alphaville’s “Forever Young” and Boney M.’s “Daddy Cool” with Suno AI, therefore alleging that the tool had been illegally trained on those and other songs protected by GEMA.
This March, Härting published an update on the SUNO case. Apparently, SUNO Inc. admitted in court that Suno AI had been trained on songs protected by GEMA, but the company alleges that it falls under “fair use”, a U.S. legal concept meant to allow for satire or academic purposes. GEMA hopes that the court will apply the same standards in this case as in the case against OpenAI, I assume leaning on the previous ruling as a kind of precedent. A ruling has been scheduled for Friday, June 12th.
In their own press release (in German), GEMA describes the hearing as the first time a European court heard arguments about audio use by AI companies; the earlier case against OpenAI centred around song lyrics. GEMA’s CEO was quoted as saying that “the generative AI companies’ business models are based on human creativity, such as that of our members”.
It’s Actually Good For You
The words of GEMA’s CEO echo those of numerous other creative professionals, including those 400-plus signatories of last year’s open letter to the British prime minister about the dangers of the Data Use and Access Bill. The bill, according to a spokesperson for the British government, “is about using data to grow the economy and improve people’s lives”. This nebulous argument for improving people’s lives seems often to be used to distract from what I believe to be the actual reason: tying the global economy to the mast of big data, desperate for the economic growth promised by AI companies.
I asked the Stim representatives giving the presentation last week how they would monitor companies’ compliance with their AI license and felt further disheartened when they said it would primarily be based on self-reporting from the companies. At least, Stim seemed to be aware of the inherent problem with self-reporting, and said they were working toward an EU-wide position of putting the onus on AI companies like SUNO Inc. and OpenAI to prove that they have not illegally trained their models on copyrighted materials, instead of placing the burden of proof on copyright holders or rights organisations.
Awareness of the problem offers me little comfort, though, in a world with light-touch regulation like the aforementioned British Data Use and Access Bill. While the European Union’s AI Act may be more encompassing and seemingly more restrictive, it still offers many exceptions to its prohibitions that, to me at least, seem ripe for exploitation in the name of (say it with me) improving people’s lives.
Both Sides
Back in February, videographer and science communicator Hank Green opened a monologue titled This is Going to be Very Messy by saying: “Artificial intelligence is a little bit perplexing, because you can not like it for a lot of different reasons.” He goes on to discuss a number of his concerns about AI, hoping to ultimately find out “the thing I am most worried about”.
The first concern that Green mentions is “an internet of slop”. He describes how while he already encounters obviously AI-generated video content, he has probably also seen videos without being able to recognise them as fake. He argues that we are making the internet worse every day by filling it with ever more AI-generated content. An open question posed by Green early on goes to the core of what many are worried about:
What is the economic incentive to create new content if it’s just being ingested by the AI machine and turned into summaries that are not monetised for the people who originally wrote that stuff?
Hank Green (28 February 2026) This is Going to be Very Messy
In one of the comments to the video, a viewer describes how they found meaning and fulfilment in their life through art, in spite of living with multiple chronic health conditions, and how sharing their art with others helped them feel less alone. However, realising that they’d had their art non-consensually used to train AI tools was a devastating emotional blow: “After my writing was scraped my heart was a little broken and I didn’t feel like I could keep posting. I feel betrayed if I’m honest.”
As an artist my biggest concern is a selfish one. I don’t like my art being stolen and used to create a soulless amalgamation of every other artist they’ve ever stolen from. I’ve had my writing illegally scraped and used as AI training data and I no longer post any art online to share with the world even though I want to. I don’t mind giving people my art for free, I enjoy sharing, but it being stolen to use in training machines trying to replace me is different to a human making a print of my painting or downloading/printing out my writing so they have a personal copy.
YouTube user @violetskies14’s comment on This is Going to be Very Messy
Creative professionals fear losing our livelihoods and control over our works and it is mostly our voices that are heard. But we are not the only ones being exploited by the “AI machine”, to quote Hank Green. This is an amateur, in the wonderful literal sense of the word, a practitioner of something not for profit but for pleasure, who felt their purpose for making art was ruined.
In a reply to that comment, another viewer described themselves also as suffering from some kind of disability, but with “basically the exact opposite” experience of AI. It had instead given them “the opportunity to develop artistic skills and be productive and find meaning”.
The pace of artistic development that I have accomplished in the past month since I discovered it is not something I could have imagined 2 months ago. I moved through learning basic tool use, to needing professional grade tools within 2 weeks and a month out I’ve developed a product that people enjoy and am able to feel a sense of accomplishment and contribution to society again, something I haven’t experienced in over a decade. I am sorry that you are having the opposite experience and I hope you come to appreciate the technology and that it enriches your life as it has mine.
YouTube user @jayjayhooksch1’s comment on This is Going to be Very Messy
I am a fairly restrictive commenter on YouTube, but this was one case where I could not keep myself from joining the discussion:
I feel inclined on the one hand toward appreciating what generative AI tools can do for disabled people such as yourself, for instance, but on the other hand – as a professional creative artist myself who stands to lose work and credibility to generative AI – I cannot square that with the fact that these tools were developed based on others’ creations, illegitimately. And while your story is a sympathetic one, it is also in my opinion kind of a slap in the face to the OP to write basically that “hey, at least people like me get to feel good using the result of what made you feel terrible”.
YouTube user @DavidSaulesco’s comment on This is Going to be Very Messy
The optimistic disabled AI user replied to me shortly thereafter, writing that they intended their original comment as “a message of hope that included sympathy and well wishes”. To their credit, they also admitted that they were “simply ignorant” and described that they felt great pleasure in being productive, creating something that was perceived of as valuable. I can absolutely sympathise with that, as I wrote to them, but I take issue with how they conflate the generative AI tools with the humans who use them. Here’s how they argued their position with another viewer:
It’s like saying that because someone is an artist, and they are capable of drawing anything, and they sell their services, that they are not allowed to watch an episode of Pokemon because that is training them to make pokemon images for their business. It’s ridiculous. If you put an image on the Internet, intelligence will view it and learn from it, that’s common sense. Doesn’t matter if that intelligence is artificial or otherwise.
YouTube user @jayjayhooksch1’s comment on This is Going to be Very Messy
A third viewer joined the discussion later, coming to the optimistic disabled AI user’s defence but with the same intellectual fallacy:
I learned from other artists. I practiced other styles to attain my style. AI is doing the same. Keep enjoying what you do. The people arguing against you are arguing against the thing that has made humans successful. We copy the actions of other humans. Reproducing someone else’s art is one thing. Using someone’s style is another. Enjoy being able to create what you want without having to feel guilty. AI is a tool. We are tool users.
YouTube user @CozyCognite’s comment on This is Going to be Very Messy
In a roundabout way, I argue this last comment could be viewed as an argument against generative AI tools. They describe themselves as having practiced other styles and learned from other artists, then argue that “AI is doing the same” only to, in my opinion, contradict themselves toward the end, stating (correctly) that “AI is a tool”. If it is a tool it is not sentient, and therefore it cannot do the same.
A User can Also Be a Total Tool
The term “training” is a convenient, easily understood shorthand for the process of refining the output capability of generative AI tools. However, I believe it subtly reinforces a perception of these tools as almost sentient, that they have an equal or similar creative ability to a human. In my opinion, AI “training” is definitionally not the same as a human practicing or honing a skill precisely because the AI “training” is performed by an automated system, ultimately just a (very advanced) computer programme, and not a human person.
These tools are often marketed as lowering the threshold for their users to edit or create things, sometimes described as “democratising” creative expression. That idea is similar to the sentiment expressed by the optimistic disabled AI user, as well as their defender. But it is rooted, I think, in a profound misunderstanding of the creative process itself.
Andreas Møller, co-founder of Danish AI-powered website builder Nordcraft, put it well in a blog post last June. I strongly encourage you to go read his post in its entirety, but I will summarise his main point here. Møller’s issue is specifically with using the term “democratisation” to market these products.
In this context, democratization seems to mean going from a (usually free) tool that requires study and learning to using a paid tool that does most of the work for you. Going from difficult but free to simple but controlled by a single company is almost the exact opposite of what the word democratization means.
Andreas Møller (24 June 2025) Stop Saying Democratize When You Mean Dumb Down
Furthermore, Møller points out that this kind of marketing implies more or less heavy-handedly that, before these tools came along, people were deliberately being kept away from drawing, writing, composing or designing by some abstract, undefined group. “It implies that having to learn things is a form of oppression,” Møller writes. This is an insidious kind of populism that plays to the same dichotomy of elites versus commoners as has poisoned the well of political discourse.
Møller writes from the perspective of building websites, but I think his arguments are applicable in this field as well. If you own a computer, you could just as easily download free and open-source software like Ardour, Audacity or MuseScore Studio as use a generative AI tool like Suno AI. There are tons of resources available to learn how to use each programme, as well as about how to create music in general.
The Path of Least Resistance
Two recent articles on Ars Technica describe how AI is plaguing U.S. higher education.
The prestigious, extremely competitive Princeton University may be home to some of the brightest, sharpest minds out there, but it is not immune to cheating. In fact, Nate Anderson writes: “Cheating is easier at Princeton than at some other places because the school does not allow its professors to proctor exams.” This relies on an “honour code”, basically a gentlemen’s agreement, along with an unwillingness of fellow students to rat out their peers. A 2025 senior student survey mentioned in the article found almost half of all seniors had witnessed cheating and chosen not to report it.
Another Ars writer, Scott K. Johnson, described in April the use of AI as “the most demoralising problem I’ve faced as a college instructor”. Since the appearance of generative AI tools like ChatGPT, Johnson has had to take time away from teaching to “moonlight as a detective and prosecutor because students without the motivation to do the work don’t have to skip it anymore”.
Johnson’s article is long and depressing but a worthwhile read. He concludes by stating he hasn’t encountered any students who believe they actually learn when they let AI tools do their work for them, but use them anyway. Corroborating Nate Anderson’s description of how many students already take cheating with AI for granted, Johnson quotes a recent conversation he overheard:
A few months ago, I overheard some college students talking about their classes. One was complaining about an assignment they needed to do that night, and another incredulously asked why they wouldn’t just have ChatGPT do it. The first replied, ’This is my major, I actually need to learn stuff in this class. I use AI for my other classes.’
Scott K. Johnson (13 April 2026) To teach in the time of ChatGPT is to know pain
In a similar way, if you use a generative AI tool to render a musical idea to audio, you didn’t really create the music. The idea was yours, sure, but you had no real control over how the idea, the “why”, materialised itself into a “what”, and you didn’t have real agency or ownership of the process, the “how”.
The Right Tools
The tools we use also shape what we create with them.
(Too) many still point to a lack of key changes in popular music as a kind of smoking gun proving that contemporary music is worse and dumber than it used to be. I think that kind of statement requires simplifying musical analysis to the point of meaninglessness. However, musician and data analyst Chris Dalla Riva describes in a very interesting post on Tedium from November 2022 how writing songs with computers instead of acoustic instruments changed how the songs were written.
On a piano or guitar, for example, certain chord progressions or key signatures are easier to play than others. (In fact, if you know what to listen for, it can be possible to identify if a song was written by a guitarist or a pianist!) However, as Dalla Riva writes, with a computer the songwriter is no longer “constrained” by their instrument: “the computer is key-agnostic. If I record a song in the key of C major […] and then decide I don’t like that key […] I can just use my software to shift it into that different key.”
Additionally, the so-called piano roll layout common in music software lends itself to a different kind of writing, argues music professor Joe Bennett, quoted by Dalla Riva in the article. It encourages a stronger focus on vertical elements, such as instrumental layering and production, rather than linear, horizontal elements such as chord sequences, melodic transformation – and key shifts.
Pointing only to the prevalence (or lack thereof) of key changes suggests, in my opinion, a blunt or incomplete understanding of how music is written. To be fair, it can absolutely be a sign of bland, poorly made music, but it certainly doesn’t have to be.
Andreas Møller concluded his post on the Nordcraft blog with a couple of notable statements: “When the tools we use no longer require any skills, we no longer matter”, and “having to invest time in learning new skills is not oppression; it is what brings you freedom.”
Using music software like Dorico or Reaper instead of a stack of manuscript paper or a multitrack recorder automates some tasks, simplifies others, and lets me do many things on my own that would otherwise have required a team and/or a substantial amount of equipment. However, I still need to both learn how to use these tools and understand how to fit different pieces together into a piece of music. This, again, highlights what I believe is the fundamental difference between composing music with a computer and generating music with an AI tool.
Music software, as opposed to generative AI, also lends me a combination of control and freedom that enables me to recreate the ideas in my head as accurately as I am able to. This of course doesn’t mean that I can’t imitate other, extant music. Following the idea posited by the disabled AI user’s comment above, I could absolutely write a parody of the Pokémon TV theme song. Doing so on my own, I would only suffer the limitations of my own skills, rather than the thick layer of abstraction that is the training data assimilated by the AI tool, which is then disseminated back to me via prompt.
Simply put, I believe that using generative AI is cheating. It is a shortcut to the reward without putting in the work. It brings to mind painting a chessboard with checkered paint, like Santa’s little helper in Disney’s 1932 animated short film Santa’s Workshop.
The Wrong Job
Earlier this year, the Swedish public service broadcaster SVT was lambasted over using generative AI to make at least eight children’s programmes. These range from Hjärnsläpp (“Brain Drain”), where two streamers imagine things like clouds made of candy or covering the entire planet in gold that are visualised with AI animations, to En zombies bekännelser (“Confessions of a Zombie”), a story about a nine-year-old zombie boy posting videos online about his life that is entirely animated using generative AI.
Editor-in-Chief of SVT’s children’s programmes Petter Bragée defended the former as “grand thought experiments impossible to visualise without this new technology” and the latter as “AI-animated humour showing Max’s journey toward viral fame […] a series about social rejection [and not] AI slop”.
Critics include journalist Björn Werner, who describes the programmes as “brain rot” and “among the worst I’ve seen on a screen”; fellow journalist Victor Malm, who overall casts a wider net in his criticism but also mentions the “generic stories, breakneck edits and [unmistakably] artificial animations”; and Josefine Engström, chairwoman of Svenska Tecknare, the association of Swedish illustrators and graphic designers, who calls SVT out as a taxpayer-funded, public service broadcaster using generative AI instead of employing human animators. The intended audience, the children, also seems overall unimpressed, to say the least.
Editor-in-Chief Petter Bragée deflects the criticism by claiming that Disney “have gone much further” than SVT and even stating plainly that SVT has had “fairly little Swedish animation previously”. I have to admit that it is mildly enraging that he basically claims ‘we don’t employ Swedish animators’ as a defence for using generative AI tools instead.
Also, when Bragée defends the quality of these programmes, he refers not to the AI-generated parts but to “the child actors performing in their own voices” and “the script written by human writers”, especially the screenwriter for Gudarnas berg (“Mountain of the Gods”) with a Master’s degree “connected to Nordic mythology” (sic!), whatever that means. (Having watched the first episode of Gudarnas berg, I hope they hire a trained screenwriter instead, next time.)
Incidentally, Gudarnas berg also features music generated with the very same Suno AI that is currently being sued by GEMA!
When pressed by one of SVT’s own reporters about why they don’t clearly label all AI-generated content, Petter Bragée refers to the broadcaster’s general guideline about not misleading the viewers, saying “no one believes it is real”. But the problem isn’t just the risk of misleading the viewers, it is also a question of honestly and truthfully labelling how the content has been made.
I suspect the real reason why SVT isn’t being more candid is simply because they know full well how unpopular it would be.
I Need Something Real
Last November, French music streaming service Deezer presented the results of an international survey of 9 000 people “exploring attitudes and perceptions around AI and music”, conducted by market research firm Ipsos. The survey, Deezer states, “shows overwhelming support” for labelling AI-generated music. Deezer has taken a clear stance against AI music, including aggressively detecting and tagging AI-generated content.
According to the survey, 97% of respondents failed to determine whether three music tracks were fully AI-generated or not. A little more than half of them, 52%, claimed not being able to tell the difference made them feel uncomfortable. In his video monologue from February this year, Hank Green talked about exactly this – the anxiety of not being able to tell whether something was artificial or not. Clearly, he is not alone.
Half of respondents, 51%, thought both that AI will play a significant part in music creation in the coming decade and that generative AI will lead to the creation of more low-quality music. Almost two thirds, 64%, of respondents answered that they believe AI could lead to a loss of creativity in music production.
Besides half of respondents feeling uncomfortable not being able to tell AI-generated music apart from man-made creations, 40% would avoid listening to AI-generated music if they came across it and 45% would like to not see AI-generated music at all.
Consistently, at least two thirds of respondents think it is unethical to non-consensually train generative AI tools on copyrighted material (73%), that AI-generated music threatens the livelihood of artists (70%), that AI-generated music should earn lower payouts than man-made music (69%), and even that training generative AI tools on copyrighted material shouldn’t be allowed at all (65%, so almost two thirds).
This can also be compared to the early demise of OpenAI’s video generation tool, Sora in March of this year. Less than six months after launching it as a stand-alone app, it was “unceremoniously killed off”, as described by Futurism writer Victor Tangermann.
Besides apparently burning through unfathomable amounts of OpenAI’s money due to its resource-heavy structure, its popularity plummeted after a burst of interest upon launch. Ahead of the app’s launch last year Sam Altman, OpenAI’s embattled CEO, hailed Sora as the harbinger of a “Cambrian explosion” of creativity and boasted that “along with it, the quality of art and entertainment can drastically increase”.
Of course, instead of an increased quality in art and entertainment, we got photorealistic videos of people shoplifting, copyright-infringing parodies of cartoons, disrespectful videos of deceased celebrities, and some very divisive children’s shows produced by or for the publicly funded Swedish public service broadcast company. (To be fair, I do not know whether they were made using Sora or some other AI tool, but I think my point stands regardless.)
In SVT’s official blog post from February this year where Editor-in-Chief Petter Bragée defended these shows, he included answers to some specific questions, including how SVT makes sure AI services aren’t trained on copyrighted material. His response is that SVT “employs external providers that claim their services are not non-consensually trained on copyrighted material”, but also that they “cannot completely verify the statements of all providers”.
Not only do I find Bragée’s responses extremely dissatisfactory, but I believe they will age like milk. Remember that SUNO Inc. admitted in court the very next month after his post that they had in fact trained Suno AI on copyrighted music without permission. Whether or not GEMA ends up winning that court case, but especially if they do, I hope many of us will be holding SVT’s feet to the fire.
