Training By Way of Theft

 

Artificial intelligence has been around for decades. Computers containing machine-learning algorithms is nothing new. What’s different about today is that we’re starting to see a ‘revolution’ of sorts, where this technology is entering many facets of our lives. Computers are only getting stronger, and more software engineers are entering the workforce. Whether it be in the workplace, in a car, on a canvas, or in a studio, AI is becoming more and more capable of doing our own tasks. Almost anyone who uses social media has heard of an AI cover of a popular song. If you simply look up ‘AI cover’ on YouTube, you’ll get results such as ‘Frank Sinatra – Smells Like Teen Spirit’ or ‘Lana Del Rey – Skyfall’. AI-generated music has amassed popularity in the last decade, recently entering major streaming platforms and catching the mainstream eye. A big concern among artists and developers alike is the issue of copyright. While the development of music AI offers certain benefits, it presents an overwhelming threat to not just singers and songwriters, but large organizations such as record labels and copyright agencies too. The industry is challenged, and is consequently faced with redefining the meaning of intellectual property. This technology can’t intrude upon or replace human creativity, and must be regulated.

In recent years, AI-generated music has seen a huge spark of interest. A couple of decades ago there wasn’t even an industry that supported this type of technology, especially in relation to anything musical. It’s important to understand why AI started to be used in music. “Modern AI research began in the 1950s, inspired by the hypothesis that the learning process could be effectively replicated with computers.” (Sunray 185). AI has had 7 decades to develop, along with technology becoming more powerful, the demand for software engineers and programmers becoming even higher, and the music industry growing to newer heights. Much like the way a human would learn through trial and error, computers could be programmed to do the very same thing, just on a larger scale. However, the idea that AI could produce its own music all together is only a recent idea. A huge ‘boom’ of interest in this technology has only been noted since 2015 (Drott). 

How does a computer even make its own music? How is it possible that covers on YouTube sound the slightest bit accurate? Much like how AI works in other fields, the foundations of music creation are the examination and replication of data. In context of “When trained on enough raw audio, the model can generate music closely resembling the unique qualities of whichever artist(s) and/or genre(s) the user selects.” (Sunray 185) From an outside perspective, that doesn’t seem so complicated. It’s understandable that a computer, once trained on these singular principles, is able to generate tons of content. Additionally, there’s an emphasis on the amount of data that should be fed into the programs. “…machine learning is ‘more like farming, which lets nature do most of the work’” (Drott). It’s a general principle that the more data that’s used, the better the output becomes. Examples like AI vocal covers mentioned previously show how highly complex this software already is. Programmers have done most of the hard work for the industry. Music AI programs are made in a way that any average joe can use it to create tons of unique content without putting much effort in. 

Music AI has had plenty of time to develop, and it’s even more evident when you enter the realm of social media. As mentioned in the introduction, YouTube has countless songs and other media containing AI-generated content, which is just the tip of the iceberg. Any search on apps such as Facebook, Instagram, or Snapchat will show the plethora of content that’s been created using this technology. There’s never been more outlets, for example, when a user wants to manipulate the voice of a singer or popular figure. In some scenarios, users do not have to gather any data for AIs to use. They’ll do it themselves. “Often, it is alleged, AI models plunder the databases without permission. Those responsible for the source material complain that their work is hoovered up without consent, credit or compensation” (A battle royal). There is no regard for the sources of data, which questions whether the use of it is fair or not. This technology could be viewed as reckless, since it has no care for authors’ rights. 

Once AI-generated content reaches the market, it’s deemed as unlawful and infringing upon the rights of artists and record labels. When tampered content involving UMG (Universal Music Group) artists entered top streaming platforms, the corporation didn’t hesitate to discredit the use of the technology. “The company is asking streaming companies to cut off access to their music catalogue for developers using it to train AI technology.” (Nicolaou). The principle that machines train with copyrighted material, specifically without permission, violates human intellectual property. Companies such as UMG have lost out on market share due to the availability of AI-generated music on streaming platforms. On the other hand, some could argue that the machines have their own IP.

The concern stretches even further. If all of this data is on the web, free for AI programs to gather and train with for whatever purpose, then these corporations and artists will continue to lack privacy. AI programs could potentially claim their own IP over their creations, whether it uses copyrighted material or not, consequently opening the door for huge lawsuits and adversity from the music industry itself. This issue heavily disincentivizes the use of AI in music-generation, as it takes away from the benefits that it could offer if further developed (The Economist). The technology has been around music for a while already, whether it’s included in Spotify playlist recommendations, transcribing audio samples, or generating chord progressions. Recently, we’ve even seen examples such as music being created for soundtracks, advertisements, or even video games (Sturm). In addition, not many professionals discuss the benefits of training AI with your own material. Thomas Rabe, CEO of Bertelsmann, mentioned, “If it’s your content, for which you own the copyright, and then you use it to train the software, you can in theory generate content like never before” (Pitel, Laura, and Olaf Storbeck). This could increase the possibilities for musicians making their own content. Even though AI’s use in music has shown tons of benefits, the legal, political, and moral conflicts surrounding its music-generation makes it hard to not shut the door on the idea all together. 

There’s a moral dilemma with having a computer take over parts of songwriting or music production. What’s original about having an AI make a song for you, especially if you claim that it’s your own? “As in so many areas of labour where AI and technology leads to more efficient production lines and increased profit but human redundancy and deskilling, can the same happen in music?” (Sturm). In many other fields, we’ve seen how automated processes in the workplace are affected. Technology is supposed to get rid of the mundane. What’s left for us are the more complex, thought-provoking tasks. While this could be viewed as an improvement for music production, it’s difficult to equate with concrete forms of labor. The manufacturing of goods is straightforward, and arguably requires no ounce of creativity or inventiveness. Writing a song is a form of art. It doesn’t get consumed the same way other goods would. If we let computers do most of the hard work, it could take away from the process. There will be less human elements to the music, and it’ll be just another good that came out of a factory with robot workers.

AI and music need to have a clear boundary. This undoubtedly goes for other forms of art too. As much as we can use algorithms as tools, the technology isn’t here to simply replace us. Only recently have we started to see the legal, political, and economical downsides to AI music-generation. Of course this isn’t going to stop anyone from using the technology, but it will definitely change the way we view the relationship between music and AI, and how we further implement it into the industry. Between artists’ IP being threatened, corporations losing revenue to computer-generated songs, and privacy being lost, these issues will not go unnoticed. We can’t let AI have its own identity. Though it can make its own works of art, and technically claim its own IP rights, we are inevitably responsible for the potential damage that it could cause on the art of music. The technology was created by humans, and must abide by our orders, unless we are ok with a world where AI does what it wants and makes its own rules. The creators of these technologies are slowly but surely realizing the immense power of them, in an almost omni potential sense. As time goes on, there will definitely be an increasing demand for this technology whether it’ll have a good or bad effect. While it’s easy to focus on all the good parts about AI-generated music, it’s been proven that using the technology and letting it dominate could easily backfire.

 

Works Cited

 

Sunray, Eric. “SOUNDS OF SCIENCE: COPYRIGHT INFRINGEMENT IN AI MUSIC GENERATOR OUTPUTS.” Catholic University Journal of Law and Technology, vol. 29, no. 2, 2021, p. 185, https://advance.lexis.com/document/?pdmfid=1516831&crid=a31c936d-01a2-4172-85f8-ba37c07838ed&pddocfullpath=%2Fshared%2Fdocument%2Fanalytical-materials%2Furn%3AcontentItem%3A63TY-4NB1-F2F4-G05F-00000-00&pdcontentcomponentid=144192&pdteaserkey=sr0&pditab=allpods&ecomp=tmnyk&earg=sr0&prid=a43be8a3-1bfd-45c7-8895-aa919186fd6f#. 

Nicolaou, Anna. “Streaming services urged to clamp down on AI-generated music.” FT.com, 2023. ProQuest, https://libdatabase.newpaltz.edu/login?url=https://www.proquest.com/trade-journals/streaming-services-urged-clamp-down-on-ai/docview/2799410181/se-2.

Pitel, Laura, and Olaf Storbeck. “AI is an opportunity for creative industries, says Bertelsmann boss.” FT.com, 2023. ProQuest, https://libdatabase.newpaltz.edu/login?url=https://www.proquest.com/trade-journals/ai-is-opportunity-creative-industries-says/docview/2818664318/se-2.

Drott, Eric. “Copyright, compensation, and commons in the music AI industry.” Creative Industries Journal, vol. 14, no. 2, 2020, pp. 190–207, https://web.p.ebscohost.com/ehost/pdfviewer/pdfviewer?vid=0&sid=7a75085d-2f4c-4d8a-a69d-8b6849874352%40redis. 

Sturm, Bob L. T., et al. “Artificial Intelligence and Music: Open Questions of Copyright Law and Engineering Praxis.” Arts, vol. 8, no. 3, 2019. ProQuest, https://libdatabase.newpaltz.edu/login?url=https://www.proquest.com/scholarly-journals/artificial-intelligence-music-open-questions/docview/2328395614/se-2.

“A Battle Royal Is Brewing over Copyright and Ai.” The Economist, The Economist Newspaper, 15 Mar. 2023, www.economist.com/business/2023/03/15/a-battle-royal-is-brewing-over-copyright-and-ai.