Songs from the digital noise: authors against artificial intelligence

When, at the start of the Second World War, Churchill threw his full support behind the work of the technological institute on the Island (Britain), which brought together several thousand scientists led by Alan Turing, the aim was to “break” the German Enigma, one of the two most powerful cipher machines. Thanks to such systems, German ground divisions, aviation and submarines achieved rapid conquest across Europe and beyond. Many regard this as one of the key reasons why the war ended in victory for the Allied forces. Alan Turing is considered a pioneer of computer technology, but he was also the first to ask: “Can machines think?” From then until 1956, which is officially regarded as the birth of AI as a scientific field, and up to the present day, research has only intensified.

Today, we live in a time when machines are replacing us, accompanied by a growing fear of the unknown: where is this leading, and what is the ultimate goal? Once, the goal was to stop a war, understand communication on the enemy side, and create a machine capable of doing so and providing an advantage. But the underlying idea lay in the philosophical question posed by Turing himself: can machines think? Today, it seems that the paradigm has shifted to another question: can machines perform?

I will not dwell on the basics of AI, the subjects we all read so much about today: logic, rules, symbols, mathematical reasoning and problem-solving. But it is a fact that the development of deep neural networks, powerful servers and data sets has enabled the meteoric rise of AI.

Of course, today we are pioneers in consuming this technology in an environment that remains an “El Dorado” in legal terms, because the legal framework does not perceive the future and therefore deals consequentially with the aftermath and damage control. On the other hand, this opens up broad ground for philosophical interpretation and for dialogue among those who are meant to teach us how to master this wild horse. TANJUG / Jadranka Ilić

In the fields of media and contemporary culture, which are the subject of this text, everything echoes with that old question: can machines perform?

At Kolarčeva zadužbina, we are at the centre of a performative ecosystem, surrounded by musical authors, creators and performers, digital content creators, writers, publishers, and film, theatre and visual artists, as well as everyone else around them. From that perspective, we see a picture that is highly complex and raises a great many questions. The actors are diverse, ranging from those, currently few in number, who quickly anticipate reality and move towards AI, learning it, using it, and even training their own models, to those at the opposite end, who sense its power and see it as a kind of tsunami approaching, without any clear idea of how to respond.

We all now know that anyone today can “make” a musical composition, a song, in one minute, even without any musical talent or knowledge. The world’s two biggest AI generators, Suno and Udio, operate as text-based models, much like ChatGPT. It is enough to enter a prompt into the programme — a 1970s-style rock ballad, with lyrics about the unrequited love of a lonely musician wandering the world, with guitar, drums, piano, bass guitar and a rough male vocal — and the song will be finished in one minute. However, that song was made by an AI generator, because the system has been trained to create a song from scratch out of digital noise. The model knows what the frequencies of different instruments and human voices will sound like. Yet that work was created by AI, not by a human being.

That is how, among others, the fake band “The Velvet Sundown” came into being, deceiving millions of listeners on Spotify. In addition to generating music, these systems also use so-called stream-boosting, that is, algorithms that artificially “play” such songs, pushing them to the top of music charts and generating revenue.

In the music industry, major news emerged a few weeks ago when the three biggest record labels, which account for more than 80 per cent of the global recorded music market — “Universal Music”, “Warner” and “Sony Music” — asked the world’s music charts to remove all AI-generated music. They believe it must be upheld that music is the product of human labour, and not of a machine that generates content entirely on its own. At the same time, they do not dispute that AI can be used as a production tool. This would avoid the possibility of AI music being included in the charts.

Recently, “Sony Music” filed a new lawsuit against the company “Udio”, which had unlawfully taken 30,000 compositions from YouTube to train its AI models, although this has so far been done with several million songs. Publishers described it as theft on an epic scale. In doing so, copyright was violated and thousands of creators were harmed without being asked, without their consent, and without any consideration of a legal framework in which human beings and their authored works would be at the centre of the system. They were joined by the Recording Industry Association of America (RIAA) and the International Federation of the Phonographic Industry (IFPI). Victoria Oakley, IFPI chief executive, said publicly: “As fans, we want to recognise the achievements of the artists who make the music we love, and it is right that charts continue to shine a light on music with human creativity at its core. When properly licensed and authorised, AI offers exciting opportunities to enhance the creative toolbox. These principles pave the way for such innovation, while ensuring we do not include tracks generated by systems that exploit artists and their music. AI is evolving rapidly and it is crucial that the music community takes the lead in ensuring that technology and human artistry grow together responsibly.”

Although publishers had already begun one form of prevention by introducing hidden digital code, a watermark inaudible to the human ear, into musical works in order to prevent scanning and use for training AI models, they have now taken an organised step further. One idea was that such content, generically labelled “AI slop”, should be removed from all charts. This was followed by the idea of isolating AI-generated content in “walled gardens”, which “Universal Music” proposed to the company “Udio”. Under that model, users would be able to play with the catalogues of stars, generate and remix material, and do other things, but with the restriction that such content could not be taken out of or downloaded beyond the platform.

The key demands put forward by publishers and organisations from the phonographic industry are as follows:

- AI systems should not be allowed to use protected compositions without an appropriate legal basis or licence when training their models;

- Authors and performers, as rights holders, alone are entitled to decide on the use of their work for AI and to receive appropriate financial compensation if they agree;

- AI companies must be transparent about which data were used to train their models;

- The issue of AI-generated voice or the imitation of artists is particularly sensitive. IFPI advocates protecting the voice, likeness, identity and artistic style of performers from unauthorised commercial imitation;

- A ban is not being advocated, because AI should be a partner rather than a replacement for creators, but within a model in which technological progress and the rights of creators advance together.

This trend of using material to train AI models is present everywhere, including in the creative industries, in music as well as film and television, but also in the gaming industry, advertising and elsewhere.

Will humans still be able to think?

Here is an example of where AI is present in the film industry. It is involved in screenplay development and dramaturgy, storyboarding, concept art, casting, digital locations, shot generation, face replacement or alteration, digital doubles, the de-ageing of actors, dubbing, translation, automatic lip-sync, editing, sound design and visual effects. From a vast database of film material, AI can create what is effectively a new film. We all still remember the major strike by screenwriters represented by the Writers Guild of America (WGA), actors and other performers represented by the Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) in the United States in 2023, which lasted almost six months. They were aware that, from footage of an actor, AI could create a credible model, that is, a replica that could be used in the future.

The strike led to certain agreements and standards, but what were actors actually most afraid of? Any film in production may have as much as 100 hours of recorded material, while only around two hours are actually used, on average. After a certain number of years, a new narrative or screenplay could be created, and replicas of actors could be placed in an entirely different context, speaking a completely different textual narrative. This created considerable fear among actors, but also among others in the film production chain.

In film, television production and advertising alike, AI can now independently create a human voice, intonation, lip movement and synchronisation. In production terms, that means a television programme, for example, can instantly acquire as many new versions as there are spoken languages on the planet. In the gaming industry, its application is at an exceptionally high and sophisticated level, making it possible to create new worlds entirely personalised to the player.

Across different categories in the creative industries, the presence of AI is very high in the film and music industries, television and photography, while in advertising, gaming and design its level of use is extremely high. Even in architecture, the level of use is high.

All of this has contributed to a dramatic acceleration and expansion of production, making work easier for content creators. In the final reckoning, however, authors may not have gained much benefit, because the overall level of production has increased.

What today is a product of humans, and what of machines?

When it comes to cultural institutions such as Kolarčeva zadužbina, we actively monitor developments in our ecosystem. UNESCO, for example, published the first global document on ethical standards relating to AI as early as 2021, and in 2025 a report by the independent expert group CULTAI, formed by UNESCO to address the relationship between culture and AI, was published. Following these trends, Kolarčeva zadužbina is working on its first document, the AI and culture policy of Kolarčeva zadužbina, whose primary aim is protection, covering the domain of AI policy, rules, contracts and archive digitisation; then opening space for experiments, through an AI lab, artistic projects and AI education; and finally positioning, through the Kolarac Forum on AI and culture, international partners and European projects.

Given the overall picture of AI’s development, it is clear that we must anticipate its role, significance and control within a system that the global community must create. In that way, we will have a chance to understand where all this is leading.

And at that point, one might pose a counter-question, rather than an answer, to the famous Alan Turing: will humans still be able to think?