AI is letting people create completely new songs and generate lyrics or even melodies in seconds. In recent days, this has sped up dramatically, and people keep saying that AI will kill music.
Technology has caused turning points in history. We’ve seen it happen countless times, and we are in one right now for music. Artists have always sought new ways to find inspiration and create great tunes. Technology is, and has always been, the close collaborator in this.
This is an extremely complex and controversial topic, as there are a lot of layers to AI Music. In the style of Taylor Swift is different from cloning her voice, which is different from using her music to train an AI. And those nuances significantly impact how people treat the finished product. To really understand it, you have to understand how the music industry already works and how it might be changing.
A Fundamental Benefit of Building AI in Music
Stakes to get AI in Music are very high. AI is opening up a new playground for creativity and the potential is incredibly exciting. But like with any new technology, we have to approach it responsibly. If we get this wrong, we could jeopardize the way musicians make money and produce art. If we get it right, we can leap ahead in how we express ourselves as humans and shrink the gap.
The Gap is the difference between the idea that’s in your head versus what you can realistically create. This idea might be an algorithm, a picture, a melody, a song or even a symphony.
Sometimes we try to make exactly what we imagined. But much often we shrink the idea and we frame the limits of what we can do with the tools and skills we have. However, creativity isn’t the same as technical skill. We sometimes confuse them because you’ve often needed one to show that you have the other. But lowering the amount of technical skill needed to express creativity has produced more great music throughout history.
Figure 1. The Gap concept
In the 18th century you needed to be a musical genius in order to create a composition. Gathering a whole orchestra to test ideas cost a lot. That is why composers like Mozart and Beethoven had to test them by themselves. So there weren’t that many of those in the world that could do this. We would have less great music today if that were still the case.
Some skeptics may argue that AI making music will hurt human creativity and musicians. But we should recognize that artificial intelligence amplifies human creativity, it doesn’t replace it.
A Debate about What AI Music Should and Shouldn’t Actually Do
Musicians disagree about what to use AI for. Some artists already use technology like GPT to generate lyrics or Google’s music LLM to create new melodies. However, many would say that they strongly disagree that AI should write lyrics. And many would support that it should be able to generate music.
Another controversial use of AI is voice cloning. How about listening to a song by Frank Sinatra singing Get Low by Lil Jon? Or Linkin Park singing the Pokemon song? Or how about Johnny Cash singing Barbie girl? These songs were all made possible using AI voice clones and simply called “voice cloning.”
Voice cloning is obviously controversial, but there are artists who come out in favor of it. For example, Grimes, who is a Canadian pop singer with 1.3M subscribers on YouTube Music. Her most known song Oblivion hitted almost 100M views on YouTube. On Spotify, you will notice two verified artists called “Grimes.” One is human Grimes, and the other is GrimesAI, a collaborator for anyone who uses her voice on Spotify.
Figure 2. Interface of Elf Tech, a tool which let anyone to clone Grimes’ voice and change their voice into hers
Grimes recently released a tool called Elf Tech which lets anyone change their voice into hers. So anyone can upload a song and clone her voice, and then a song may get published on Spotify. Grimes has promised publicly, that when she will release her new album, she will release a competitive “made by others” AI album at the same time. However, a lot of other artists are extremely not okay with people using their voice this way.
Copyright vs. AI: When Should Artists Get Paid?
For centuries, the music industry has built a copyright system to ensure artists receive payment for their art or the intellectual property (IP) they create. For decades, the industry has defined what counts as inspiration versus copying to determine whether artists get paid. Now with the rise of AI, the question is whether it can fit into that system or if something’s got to change.
Figure 3. Four major situations regulated by the copyright system in music
Most music in the world falls into Case A (see Figure 3). This is when the artist is getting inspired by someone and not paying them. You may hear from Billie Eilish telling that she had a lot of inspiration from the Beatles. Or Sam Smith admitting that his greatest influences were Whitney Huston and George Michael.
Copying someone and paying them for it is also totally fine, which is the Case B on Figure 3. This form of music creation has various shapes. Three most frequent forms are:
- Sampling. Doja Cat, an American rapper and singer with over 13 million subscribers on YouTube, recently published her new song “Paint the Town Red”, which plays on a back end a line from another song from the mid-60s called “Walk On By” by Dionne Warwick. In case of sampling Doja Cat needs to pay whoever owns the original composition of the song and the original recording.
- Interpolation. A musician might rerecord a line or a melody from someone else’s song and use it in their new one. A great example of it is the controversial yet wildly popular song “7 rings” by Ariana Grande. This song is one of the best-selling songs in digital history with sales of over 13.3 million copies worldwide. However, Ariana keeps only 10% of her songwriting royalties because “7 Rings” interpolates the melody of another iconic song “My Favorite Things” written by Richard Rodgers and Oscar Hammerstein for “The Sound of Music” musical back in 1959. She is not paying for the original recording since she is not using it in her song. However, Ariana pays 90% of royalties to the original composer of the song.
- Covering. An artist can rerecord an entire song originally created by someone else. For example, Dolly Parton wrote and recorded “Jolene” in 1973. Numerous artists have covered this song, with Miley Cyrus‘s cover achieving five times more views than the original recording on YouTube. In this case, which is quite similar to interpolation the artist will pay the owner of the original composition.
What Grimes is doing with her Voice cloning which I described above, also perfectly fits in Case B too (see Figure 3). People are making new songs and recordings, but they are using her voice and likeness. So they’ve agreed that they should pay her.
In all cases described above, some original owner is getting paid, and it doesn’t matter how the track was done. If someone is using AI to do any of these things, they have to pay.
The entire copyright system is built to protect the artist from a situation illustrated in Case C of Figure 3. Copyright protects the original song, recording (also called “the master”) and separately the composition of the original song itself.
But, as it often happens with legal matters, it’s less a list of rules and more a list of lawsuits. In 2015, for example, Pharrell Williams and Robin Thicke had to pay $5 million after losing a lawsuit. The lawsuit claimed that their song Blurred Lines copied the composition of Marvin Gaye’s Got to Give It Up. On the other hand, Ed Sheeran recently won a lawsuit over a chord progression in his song “Thinking Out Loud,” which the heirs of Ed Townsend claimed he copied from a song he co-wrote.
Sometimes it’s just messy. Like Olivia Rodrigo retroactively gave Taylor Swift songwriting credits on some of her songs, and therefore a lot of money in royalties because other people accused her songs of being too similar.
But what about an artist’s vibe, such as creating something “in the style” of Taylor Swift? What if someone makes a song that sounds like two artists collaborating, but it’s neither their composition nor their original recording? This is where all hell breaks loose. Impersonation isn’t allowed, but parodying is. Copying specific producer tags is prohibited, but you can replicate phrases others have used. Using another song’s melody is off-limits, but you can employ popular chord progressions.
Should Artists Get Paid if an AI Trains on their Music?
As seen from the examples above, right now the way the industry solves the payment issues is based entirely on the output, not the input. Artists receive payment only if the end result sounds enough like them, even if their music wasn’t used in the creation. Conversely, you could download someone else’s entire song and avoid payment if you change it enough to make it unrecognizable.
This fundamental approach has been a cornerstone of the entire copyright system. On the one hand, every musician learned on the back of every other musician in the past. The critical dispute happening in the music industry now is whether we should change it because of AI. If we do that, it would mean filling a box that so far has never been filled: paying artists for what the industry calls “inspiration” (see Case D in Figure 3).
The copyright system is just incredibly complicated and there’s a fight every time people disagree. This holds true whether the work involves AI or not. But there is a question that’s specific to AI, which is: is that inspiration, or is that copying?
AI tools are not sampling or interpolating or covering. They are not taking pieces of the music in their data sets and collaging them. They are analyzing patterns and creating entirely new works based on them. At this stage there is no consensus in the industry whether this is an inspiration or copying.
One of the Most Important Legal Questions of the Century
The AI tools we’re talking about require thousands of songs to learn. These training data sets are just enormous. One specific example can be AudioCraft, a suite of large generative music models released by Meta in August 2023. Just one AudioCraft model, MusicGen, in order to enable its algorithm to generate patterns of sounds in response to prompts analyzed patterns in some 400,000 recordings with a collective duration of almost 28 months. Stable Audio, a model released by Stability AI in September 2023, trained on about 800,000 tracks. Stability AI, a London-based firm known for creating the famous Stable Diffusion image generation model, developed it.
The problem with AI tools is that the companies that release them often aren’t exactly forthcoming about what sources they have been using. There are lots of discussions going on right now, including the arguments that AI firms should “document and disclose” their sources.
There are several ongoing lawsuits worth keeping an eye on across different art forms. In September 2023, a major lawsuit targeted OpenAI’s ChatGPT, alleging that the company illegally copied copyrighted works of authors to train the AI. More recently, major record companies—Universal Music Group, Warner Music Group, and Sony Music Entertainment—sued AI music generators Suno and Udio for “mass infringement” of copyright.
On the other hand, copyright holders do not want to sound like enemies of progress. Those responsible for the source material acknowledge the fact that the more access to training sets there is, the better AI will be. And without such access there may be no AI at all. In other words, the industry might die in its infancy.
For those of us who work in the tech industry on the edge of AI development there is nothing left but to closely watch what is happening in this IP battle with AI and eagerly wait what will be the answer to one of the most important legal questions of the century: “Will copyright law allow robots to learn?”
About the Author
Yuri Berchenko is a global expert in the digital and entertainment industry with 15 years of product management and business development experience in Big Tech.
His leadership in the technological sector includes innovative and powerful product partnerships efforts in video and music streaming space, including launching value-adding products with such companies as Samsung, Vodafone, Orange, PayPal and Revolut.
Moreover, as one of our speakers at the second annual Data Innovation Summit MEA in Dubai, Yuri presented an early look at AI experiments which may transform the future of digital entertainment. Gain in-depth insights as he shares strategies for responsible AI innovation through partnerships!
For the newest insights in the world of data and AI, subscribe to Hyperight Premium. Stay ahead of the curve with exclusive content that will deepen your understanding of the evolving data landscape.
Disclaimer. The opinions expressed in this publication are solely those of the authors. They neither reflect nor purport to reflect the opinions or views of YouTube, Google, Alphabet Inc., their respective parent companies or affiliates or the companies with which the authors are affiliated and may have been previously disseminated by them. The authors opinions are based upon information they consider reliable, but neither YouTube nor Google, nor Alphabet Inc., nor their affiliates or parent companies, nor the companies with which the authors are affiliated warrant its completeness or accuracy and it should not be relied upon as such.
Add comment