Artificial intelligence (AI) has brought many possibilities and challenges to the music industry. AI is used by streaming services to generate personalised playlists for consumers, arguably positioning themselves as cultural intermediaries by deciding which music we listen to. AI is used by labels and A&R teams to not only find the industry’s next big talent, but also to analyse whether a song will be a chart-topping hit; tools like HITLAB and Musiio can analyse up to 80,000 minutes of music in less than 60 seconds, processing data that provide insights into which songs should be released and which artists should be invested in. DIY and independent songwriters and artists can now travel down the cheaper road to release by using AI to mix and master their music, dispensing with the need for trained and talented producers. As with many aspects of our lives, AI has slowly been inflitrating the music industry, perhaps for good, perhaps not… the jury is still out.
Terrifyingly for some, AI can also now write the music you listen to and just anybody and everybody can take advantage of this developing technology. Perhaps you are an experienced songwriter who wants to experiment with AI as a tool and a creative partner. Maybe you find yourself with writers block and need some help crafting a new melody. Perhaps you write a mean hook but you can’t play the drums with your two left feet and you need some help writing a decent beat. Or maybe you have no musical training or experience whatsoever, but you fancy releasing a song anyway and seeing if you can make it in the music industry… how hard could it be?? Whatever position you find yourself in, there is an AI out there ready to help.
AI music generation has been hovering in the wings for more years than many realise, but it’s really in the past year or two that the technology has been good enough to produce music that could pass the Turing Test. However, as with any new technology, AI-generated music raises important legal and ethical questions, particularly in the realm of copyright law. There remains the question of whether AI-generated music or lyrics can or should qualify for copyright protection; the human requirement of many copyright frameworks and the concept of authorship and ownership raise the complicated question of legal personhood, which which if offered to AI could open Pandora’s box! (We’ll save this for another time). Equally as complex an issue is copyright infringement; there is potential for an AI system to infringe upon copyright protected music in its training data (read more about AI training processes here), and to infringe upon copyright protected music in its output. While AI music and the legal and ethical questions that come with it are bewildering to many and a minefield for the music industry to navigate, this is not completely unchartered territory! To understand the challenges of AI-generated music, perhaps we can draw lessons from the music industry’s handling of other landmark watershed moments involving technology, such as sampling and streaming.
Sampling
Sampling has become a ubiquitous practice in modern music production, but its origins and evolution are fascinating and complex. We can attribute the advent of sampling to French composer Pierre Schaeffer who was splicing together bits of old recorded material in the 1940s; this type of music became known as musique concrète. However, sampling as we know it today is the practice of taking a portion of a sound recording and/or a song and using it in a new composition. Music can be sampled in two different ways: autosonic quotation involves using a portion of a sound recording, anywhere from one second of sound to an entire riff or chorus line; allosonic quotation uses material from an existing song in a new song by way of re-recording or performing the quotation, also referred to as interpolation.
Sampling has been a significant practice of the music industry since the late 1960s and early 1970s, when artists and producers began experimenting with new ways of creating and manipulating sound. Early samplers were often rudimentary, consisting of simple tape recorders and record players, however, they paved the way for the more sophisticated sampling technology that would emerge in the 1980s and 1990s. Samplers such as the Akai MPC, the E-mu Emulator III and various machines by Korg, Roland and Casio facilitated the popularisation of sampling in hip hop and electronic music in the late 20th century; the 21st century and digitalisation witnessed the rise of digitial audio workstations (DAWs) such as Pro Tools and Ableton which can now do the same job in a faster and more streamlined manner, with added potential for creativity. AI has since found its way into the world of sampling with DAW plugins such as stem separation tools, sample finder tools and there are also platforms that will create new samples for you to use.
Sampling and copyright law
During the early days of sampling, copyright law did not specifically address the use of samples in new compositions and songs and sampling was used freely without permission or payment to the original copyright owners. However, as sampling became more common, record labels and copyright owners began to take notice. The rising popularity of sampling raised important legal and ethical questions about copyright law and intellectual property and a wave of copyright infringement lawsuits ensued. With lack of precendent in this area and no case law to be guided by, those accused of copyright infringement by sampling would rely on the doctrine of fair use as a defence in US courts. Fair use in US law allows for use of copyright protected material without permission for a limited and ‘transformative’ purpose, such as parody, criticism, comment, news reporting, teaching, scholarship, or research. To compare, UK law would rely on the copyright limitation fair dealing, a pre-emptive limitation to copyright protections rather than a defence. Fair dealing allows for the use of copyright protected material without prior permission from the copyright owner in certain circumstances, similar to those mentioned above. However, fair dealing is more limited in scope than fair use; an exception that applies in the US may be regarded as infringement in the UK.
The first major copyright case against sampling in the music industry was Grand Upright Music, Ltd. v. Warner Bros. Records Inc., decided by the United States District Court for the Southern District of New York in 1991. The case involved a song called ‘Alone Again‘ by rapper Biz Markie, which included a sample of the song ‘Alone Again (Naturally)‘ by singer-songwriter Gilbert O’Sullivan. O’Sullivan sued Warner Bros. Records, the distributor of Biz Markie’s album, for copyright infringement. The court ultimately ruled in favour of O’Sullivan, finding that the use of the sample without permission constituted infringement.
The ruling in Grand Upright Music, Ltd v. Warner Bros. Records Inc. was significant because it established that the use of even a small portion of a copyrighted work could be considered infringement, and that permission must be obtained from the copyright owner before using a sample. The case ultimately led to the development of industry-wide standards and practices for sampling as record labels began requiring artists to obtain permission and pay for the use of samples. In the case of an allosonic quotation or interpolation, a licence for the use of the music and lyrics of a song is required; in the case of an autosonic quotation or direct sample of a sound recording, an additional licence for the use of the sound recording is required on top of the licence for the use of music and lyrics. This means that to sample a sound recording of a song is more expensive than to simply re-record or cover the song quoted.
Case law continued to determine in what circumstances sampling could be considered copyright infringement. For example, many believe that you can use a sample of up to two or three seconds without permission, as per the de minimis doctrine in US and UK law, which holds that the use of a small amount of copyrighted material may be permissible if it is insignificant or trivial. It should be noted that UK copyright law states that infringement must involve the whole or any substantial part of the work, which raises the question of how substantiality can be measured. UK courts and US courts have tended to take a rather subjective view of it in relation to copyright infringement, relying more on the quality and not quantity of copying, decided on a case-by-case basis.
The case of Bridgeport Music, Inc. v. Dimension Films dispelled the myth that a certain number of seconds of a work can be legally sampled. The hip-hop group N.W.A. used a two-second guitar riff from Funkadelic’s ‘Get Off Your Ass and Jam‘ in their song ‘100 Miles and Runnin’‘. The lawsuit was brought by Bridgeport Music, which owned the rights to the Funkadelic recording, and alleged that the sample constituted copyright infringement. The case ultimately went to the Sixth Circuit Court of Appeals, which ruled in favour of Bridgeport Music, finding that the use of the sample without permission constituted infringement. Quality over quantity; even a two second sample is substantial enough to be considered copyright infringement.
What is considered to be insignifcant and trival use is subjective and the quality of a sample used is as important, if not more so, than the amount of the sample used. As well as deciding whether the amount of the work used is reasonable and appropriate, courts must consider whether the use of the work affects the market for the original work and/or whether it could act as a substitute for the original work. How will the original copyright owner be affected (usually monetarily) by the use of this sample?
Today, sampling remains an important part of many genres of music, and the legal and ethical issues surrounding sampling continue to evolve. The music industry adapted to the practice of sampling, and now rightsholders offer licensing options to artists who require permission to use samples; both for the use of music and lyrics and for the use of sound recordings. The rise of digital sampling led not only to the creation of licensing solutions but also to questions about the originality and creativity of music created using samples, just as AI music has more recently. It’s perhaps pertinent at this point to note that when an AI system is trained with existing music, we are considering that music to be high quality in nature and copied/used in its entirety, therefore the de minimis doctrine may only be used if an insignificant or trival amount of the training data ended up in the AI’s output. The distinction between fair use (US) and fair dealing (UK) is too complex a topic to detail here, but it is unlikely that the use of copyright protected works as AI training data would find itself within the realms of fair dealing in the UK. Some however believe that it could be considered fair use in the US. (More on this another time).
Is artificial intelligence simply an evolution of technology which should be mirrored by the evolution of legal and ethical issues that technology presents, or should the music industry learn from the historic case of sampling to inform how the use of AI should be regulated in the future? At the very least, should the music industry be thinking ahead to creating a licensing environment for the use of copyright protected music to train AI systems now rather than in retrospect as in previous experiences?
Streaming and copyright law
Today, we understand streaming to be the distribution of music through online platforms funded by advertising revenue and subscription fees paid for my music consumers. However, the concept of streaming began as the digitisation of music allowed recordings to be ripped and copied from digital files stored on a CD and later shared online without the permission of copyright owners; this, of course, constitutes copyright infringement. This practice presented a new set of challenges for the music industry in the late 1990s into the 2000s, leading to widespread piracy fostered by platforms like Napster, Limewire and The Pirate Bay which allowed users to fileshare music online. The ease with which music could be shared and downloaded illegally led to significant losses for the industry, as consumers turned away from legal channels of music consumption in favour of free alternatives. Piracy and streaming transformed the way we consume and listen to music and subsequently the way we write, record and release music.
Digital online piracy caused significant damage to the music industry, as sales of physical media declined and digital sales through distributers like Apple’s iTunes failed to pick up the slack. The music industry wasn’t a stranger to technological development of course. Sony as a technology company, now one of the ‘big three’ music labels and music publishers, saw the potential of releasing new music players and new music formats, and the music industry as a whole piggy backed on the evolution of technology, realising the financial reward of rereleasing old records. Many music fans who had previously bought albums on vinyl later repurchased those albums on cassette, then on CD, some on MiniDisc (though that fad didn’t last long), and later these albums were downloaded digitally to MP3 players. Suddenly the music industry was able to increase the profitability of music they previously believed would find their way into obscurity – financially at least.
So, digital online piracy was a big problem. The intention of this post is not to dive into how piracy and streaming changed the way contracts were approached between businesses and artists, not is it to explore how live music became a major money spinner as a consequence, but it’s important to at least acknowledge that the fact that music could be accessed for free radically changed the business model of the music industry. The music industry’s knee-jerk reaction, however, was to pursue legal action against online piracy and the industry continue to be very quick to litigate where infringement of copyright is concerned.
The legal battle between the music industry and Napster is one of the most significant copyright cases in music history: A&M Records, Inc. v. Napster Inc.. Napster was a peer-to-peer filesharing service that quickly gained popularity, with millions of users downloading and sharing music illegally online. In 2000, the Recording Industry Association of America (RIAA) filed a lawsuit against Napster, alleging that the service was facilitating copyright infringement on a massive scale by enabling users to share copyrighted material without the permission of the copyright holders. The case went to trial and in 2001, a federal judge ruled in favour of the RIAA, issuing an injunction that required Napster to shut down its service. The judge found that Napster was facilitating copyright infringement, that the company was aware of the infringing activity taking place on its platform and that the company was therefore liable.
The Napster case had a significant impact on the music industry and on copyright law. It established that peer-to-peer file-sharing services could be held liable for copyright infringement, and set a precedent for future cases involving similar services. In response to the Napster case and with the realisation that physical media was no longer a money-spinner, the music industry began to explore and embrace alternative models for digital distribution, including legal streaming services such as Pandora, Spotify, Apple Music, Amazon Music, Tidal and YouTube. The Napster case also helped to shape copyright law, as courts grappled with the challenges of enforcing copyright in an increasingly digital world.
In the streaming era, where music is distributed over the internet and stored on servers owned and operated by streaming platforms, it can be more difficult to control the unauthorised distribution and reproduction of copyright protected music.To address this issue, the music industry developed a number of new legal and technological solutions and copyright law has evolved to include new provisions specifically aimed at streaming services. For example, the US Digital Millennium Copyright Act (DCMA) includes a safe habour provision that protects streaming services from liability for copyright infringement, provided that they comply with certain requirements, such as promptly removing infringing material upon notice from copyright owners.
In UK law, several amendments have been made to the Copyright, Designs and Patents Act, 1988 (CDPA) to reflect the challenges and opportunities presented by the rise of digital technologies and streaming services. Before Brexit, the UK implemented the EU’s Copyright and Information Society Directive (InfoSoc) 2001, which amended the CDPA to provide for the exclusive right of copyright holders to make their works available to the public, including through digital transmission. This provision recognises the significance of streaming services and the need to protect copyright holders’ rights in the digital age. InfoSoc and the DCMA also recognised the use of digital rights management (DRM) and technical protection measures (TPM). DRM technologies govern the use, modification and distribution of copyright protected works and of systems that enforce these policies, such as encryption and licensing agreements.
There have been several high profile legal cases regarding the circumnavigation of DRM and TPM within the music industry and in other creative industries – such as film and gaming – and the legal framework surrounding DRM and TPMs in the music industry is an ongoing area of discussion and debate for a number of reasons: technology is constantly evolving and changing, and with each new development come new questions and challenges regarding copyright law and enforcement; there is often a tension between the desire to protect intellectual property and the desire to foster innovation and creativity. While DRM and TPM can be an important tools for protecting copyrights and preventing infringement, they can also be seen as overly restrictive or inhibitive to legitimate uses of copyrighted material. Finding the right balance between protection and openness is an ongoing challenge for the legal system.
Streaming has also highlighted the issue of secondary copyright infringement. This applies to those who indirectly infringe copyright by facilitating copyright infringement by others. There are several theories by which a party can be liable for secondary infringement: (1) contributory infringement, where one has a connection to infringing activity contributes to it; (2) vicarious liability, where one has the right and ability to control infringing activity but fails to do so while benefitting financially; (3) active inducement, where a party distributes a device intended for use in infringing copyright; and (4) willful blindness, where one is aware that infringing activities are taking place but chooses to overlook. In 2018, BMG Rights Management, a music publisher, sued Cox Communications, a cable and internet service provider, for contributory copyright infringement and vicarious copyright infringement: BMG Rights Management v. Cox Communications. BMG alleged that Cox failed to terminate the accounts of users who repeatedly engaged in copyright infringement by sharing and downloading music through peer-to-peer file-sharing networks. The court found that Cox was liable for contributory copyright infringement because they consistently reactivated user accounts following a termination, despite having knowledge of infringing activity. The court also found that Cox was vicariously liable for copyright infringement because it had the right and ability to control the infringing activity and received a direct financial benefit from it.
As streaming continues to evolve and new technologies emerge, it is likely that there will be further cases that will help to shape and clarify the legal framework governing streaming and copyright law. As AI music generation systems rely on existing music in their training, copyright infringement is a real threat. It is possible that the music industry may need to develop and enforce DRM and TPM to prevent copyright protected music being used by AI companies without permission and without the payment of a licence fee. Perhaps it is also possible that we may see some litigation like the cases mentioned above where AI platforms are sued for secondary copyright infringement: (1) AI companies could be held liable for secondary infringement if they are aware that their AI models are trained on copyright protected music without permission from rightsholders but overlook the matter (contributory infringement, vicarious liability and willful blindness); (2) if platforms powered by AI companies merely facilitate users to infringe on copyright, they could also be liable for secondary infringement (active inducement). Stem separation tools are at risk of this kind of litigation and this is an issue that has already been recognised by the RIAA.
AI-generated music and copyright law
AI-generated music raises new legal and ethical questions for copyright law, just as sampling, digital piracy and streaming continue to, particularly in regards to ownership, originality and copyright infringement. Determining ownership of AI-generated music is a complex issue that continues to be researched and discussed in academic circles and in music industry. Though the description of such requirements differ among jurisdictions, copyright law also requires that a work be original and creative. Some, the US and the EU included, require that a work be created by a human; the UK however does not. There is a lack of harmonisation internationally which further complicates the case of AI-generated music and copyright law, and while AI systems can create music that may be indistinguishable from human-created music, the question of whether such music meets the originality and creativity requirements of copyright remains unclear.
However, to address these challenges, perhaps the music industry can draw lessons from its experiences with sampling and streaming. As with issues of sampling, industry-wide standards and practices can be developed for the creation and distribution of AI-generated music, and in the creation of a licensing environment which would facilitate the clearance and licensing of copyright protected music for AI training. The truth about training materials used by AI companies remains behind a heavily chained and guarded black box, but we do know that machine learning systems need big data to be trained on and that this data needs to be high in quantity and quality; the larger and better the input data, the better the output product. Some AI companies do work with ethics in mind by hiring in-house composers to create datasets, by using open source datasets, and by relying on music in the public domain as training data, but perhaps there is a case of what we don’t know won’t hurt us? Many AI companies are not transparent in their uses of existing music and will not release details of their training data.
The music industry is currently under threat of losing the right of reproduction to AI companies. The EU Directive on Copyright in the Digital Single Market, 2019 (CDSM), Article 4, introduces a copyright exception which allows for the use of copyright protected material as training data so long as the copyright owner hasn’t (in a machine readable format) opted out of the exception. Having chosen not to implement the CDSM following its withdrawal from the EU, UK policymakers were close to providing the same exception but without allowing copyright owners the option to opt out, meaning that music creators and owners of sound recordings in the UK would have had no say in whether their music was mined as data for the training of AI music systems; this decision is currently on hold and the future of UK law is currently unknown and still in discussion. If we were to learn a lesson from sampling, surely the answer it to create a licensing environment that would benefit the AI industry, but provide rightsholders with the option to either offer their music as training data and be financially compensated for that use, or to prevent their music being mined by AI companies.
Legal and technological solutions can be developed to address the unique challenges of AI-generated music. For example, blockchain technology could be used to track ownership and licensing of human-created music as AI training data and to track ownership and distribution of AI-generated music, promoting transparency not only within the music industry but also with consumers. Taking lessons from experiences with streaming, DRM and TPS could continue to be developed to protect copyright protected music from the threat of AI crawling and AI training. The music industry could also come down harder on platforms that appear to facilitate the infringement of copyright protected materials by others. Perhaps AI itself could even be used to detect and prevent copyright infringement.
There are many challenges here and while the previous few paragraphs may to some appear like simple solutions and to others complete nonsense, the truth of the matter is, as the UK and the US are common law jurisdictions, historically legal issues around new technologies have only been decided once there is case law. If we can take anything concrete from experiences with sampling and streaming, somebody needs to sue somebody else for us to be able to decide how the issues of AI-generated music should be dealt with. We can research, we can discuss, we can test theories and hypotheses, we can look to the past to better plan the future, but the truth of the matter is, ultimately, a court will decide.
However, we may see some litigation sooner rather than later, potentially through RIAA complaints against various AI music companies, but more likely by example from other creative industries. Getty Images is the process of suing Stable Diffusion for scraping its content (copyright infringement). Class action has also been filed by creators against Stability AI, Midjourney and DeviantArt for the same reason. While the inputs and outputs in these cases are images, the processes are the same as those used in AI music. The outcome of these cases should set a precedent for whether AI music companies can use copyright protected works in the future and may inform the music industry on how to better regulated the threat of AI music in the future. My hunch is still on licensing and DRM though! Let’s learn from the past and get ahead of the curve for a change!
What are your thoughts? Are there alternative solutions? Leave a comment and join the discussion!