It was clear from the outset that artificial intelligence would clash with existing copyright law. But while politicians and courts are still busy pondering the consequences, technology is already several steps ahead of the game.
What if AI could produce music before it was actually composed by humans – and sue artists in the future?
Cameron Gorham produces floating ambient music under the name Venus Theory and earns his living primarily as a sound designer. Recently, he discovered a glaring loophole in the YouTube system. In principle, the platform allows content to be protected from unauthorised use and, if necessary, infringements to be held accountable. All rights holders have to do is register their music on a service called "Content ID". So far, so good, so seemingly simple. In practice, however, Content ID gives rise to a number of conflicts. Some of these could soon prove disastrous.
Published 7:51 am on Thursday 23rd January 2025 by Beat Magazine
Registered fraud on YouTube
First of all, many people are completely unaware of the fact that they have to register their works on Content ID in order to enjoy protection. As a result, third parties can close the gap by registering pieces they haven't even written themselves, skimming off the revenue and suing the actual authors. You would think that this kind of fraud would be quite easy to uncover. This is not the case, however, since Content ID was never intended to work as an instrument of copyright protection, but rather, the other way round, to make sure the platform remains from copyright conflicts. This can be seen in the way YouTube deals with alleged infringements.
Firstly, the system performs constant scans to see who uses registered content on YouTube. Any use that is not explicitly authorised immediately results in the material being blocked. If you're affected, you can object to the ban and ask for the block to be lifted again. Paradoxically, however, it is up to the plaintiff to decide whether the block should be upheld or lifted – it isn't hard to see how this can very easily be abused. Also, as soon as a content creator has appealed and been rejected three times, their account is automatically suspended. This means that attempting to assert your rights can turn out to be costly.
The situation becomes downright dystopian once AI enters the equation. To explore the possibilities of manipulation, Cameron decided to use an application to produce a new song based on one of his own tracks not yet registered with Content ID. The prompt: generate a track that sounds as much as possible like a Venus Theory song without constituting outright plagiarism.
The results were shocking!
On the one hand, the AI track was remarkably good. This alone opens up countless possibilities for film producers or any commercial enterprises to copy the sound that Cameron has developed over many years of trial and error, countless productions and thousands of unpaid studio hours without having to pay him for it. In addition, endless variations of Venus Theory titles can now be created, which can then be registered under a different author via Content ID.
It is true that neither copyright law nor Content ID permit the registration of material generated by AI. But who is going to verify whether those registering the content are answering that question truthfully? This allows for a truly bizarre scenario: Using Venus Theory as a seed, artificial intelligence could create a vast amount of pieces in his style and register it with Content ID. Once Cameron then uploads new music, he himself could be accused of plagiarising these tracks – which his own creativity and past releases helped create.
Copyright itself is also under fire
In songwriting, the possibility of writing music in the style of existing artists raises serious questions. Musicians are already running out of melodies. Tech lawyer Damien Riehl has convincingly argued that the number of notes that can be meaningfully used in pop music is frighteningly small. Anyone who sits down at their instrument or laptop is therefore faced with the challenge of not repeating one of the sequences of notes already in use. It can come as no surprise that the amount of court cases dealing with plagiarism has risen exponentially over the past years. The few unused combinations that are still available to us will soon be exhausted, mainly due to the supernatural computing power of "creative algorithms".
The consequence: either we all start listening to twelve-tone music or, as Riehl proposes somewhat more plausibly albeit just as radically, we exclude melodies from copyright. The proposal makes sense, sounds plausible and is made by a man who is obviously a big music fan. But it does raise one question: what will be left of copyright once we strip from one of its main pillars?
The question may feel urgent, but it is hardly new. After all, copyright law has been in a permanent crisis since the mid-1960s at the latest. From compact cassettes to CDs and downloads – there has been a constant need for setting limits on the permissibility of copying music. As a result of this process of constant but increasingly rapid change, copyright has been expanded, clarified and in many cases also restricted. However, the situation has hardly calmed down. On the contrary, the pace of technical innovation has increased, and with it the need for a legal response that takes equal account of the interests of all parties involved. Today, the pace has reached breackneed speed and every regulation is already outdated by the time it is passed.
A quick look at online videos demonstrates the urgency of this issue. Fil Henley from the band Wings of Icarus has become the web's authority on voice analyses, demonstrating meticulously and with scientific precision that almost all stars and superstars are ultimately only moving their lips to a backing track in their "live shows". He is also interested in the ways in which some channels are playing with fake identities or manipulating reality. As an example, Henley recently presented various videos in which, at first glance, young female guitarists seem to be playing their instrument. However, all of the clips are showing AI-generated images. At entry-level quality, artefacts are still quite clearly recognisable, elements disappear and suddenly reappear, fingers and hair occasionally become blurry. Overall, however, the quality is remarkable. When it comes to the most technically advanced version, only trained observers will be able to recognise the deception.
Crimes of the future
In a way, you could argue that these techniques will make it possible to infringe the rights of creators in the future. This is because there are no precedents in existing law for many of the situations mentioned above and that there is therefore no way of dealing with them. The situation is further complicated by the fact that different countries not only have different understandings of copyright, but will also assess the same situation differently. For example, in most countries it is not legal, or only permissible within very narrow limits, to use the facial features, likeness or voice of a person without their permission. Canadian rapper Drake was faced with these facts when, using AI, he collaborated with Tupac Shakur and Snoop Dog on one of his tracks. Tupac's heirs indignantly refused to provide their consent and the song had to be withdrawn. However, in Australia, for example, restrictions are far less strict in this regard, which raises questions about how current law can be enforced internationally. In addition, there can be striking voice and facial similarities in humans as well – where should one draw the line?
"New technologies make it possible to infringe the rights of creatives in the future. This is because there are no precedents for many current problems in existing law."
Not every dark fantasy of the future will eventually materialise. But you don't have to conjure up Skynet to realise how far computer-assisted music production has already progressed. In a good-humoured video, producer Henry Clarke demonstrates how he is able to create a song that is remarkably listenable within a couple of hours: First, he asks ChatGPT to write lyrics and a chord progression. He then copies this information into the editor of the "Band in a Box" software. The app then creates a complete instrumental arrangement to match, which feels astonishingly organic. All Clarke has to do now is sing a melody and tweak the arrangement a little - although these aspects will soon no longer require any human involvement either. Entrepreneur, IT expert and musician Ravin Mehta has just founded the Made by Robots label, in which AI plays a central role, and emphasises that there is only one fair way to deal with artificial intelligence in the creative sector:
"For now the only way to be safe for commercial use is to either create the data yourself or to license the raw data." Purchasing a model that has already been trained is not an option, as this would inevitably involve the use of protected material."We have created a data pipeline that can take any kind of musical - or not so musical - data and train a model that can then be used to create something new," says Mehta, "Examples are the tech house tracks on the label but also the music made from data we got from sensors on plants that we used to train an AI model and then make music with it."
It will clearly require new ethics that are politically and legally supported to master this situation. A lot of time will pass before this happens, however - how much do musicians still have left?
Want more? Get more!
Subscribe to the digital edition of BEAT Magazine via Plugins-Samples.com and get more gear, in-depth workshops, reviews and 11 GB exclusive plugins and new sounds with every monthly issue!
