Thoughts on recent AI music related news
As someone who has been involved in generative music for most of my career, I wanted to weigh in on some recent news related to AI music. Specifically, I want to share my views on some of the uses and benefits of using these tools, while also addressing some recent concerns.
My Work and Approach
I am a composer, and started learning to program as a means of expanding my creative toolbox. At first I mostly wrote scripts to try out various process based ideas I had, as it is easier to be objective about whether the result is good or not when you can try out variations within seconds rather than hours of working things out by hand. Later I began developing tools for using computers in live performances, including editing the code producing the audio and video live on stage.
Eventually my personal and academic work on generative music led to me joining a startup called Amper Music in 2015 (which was acquired by Shutterstock in 2020), where I became Chief Scientist and was the architect behind their AI music composition system. Our focus was empowering non musicians, typically video creators who would otherwise use stock music, to 1) get the music they are looking for faster, 2) avoid spending time and effort editing the audio to match the video, and simplify the licensing for using the music. With Amper, you at a minimum specify how long the music should be (optionally also saying there major changes should happen), the genre, and the mood that the music should convey, and in a matter of seconds you have a finished piece of audio. We also offered the ability to change almost anything about the result while keeping everything else as close to unchanged as possible.
Specifically, the music that Amper was focused on was what we referred to as "functional music": something that is valued for how it is used, as opposed to "artistic music": something that is valued for its artistry on its own merits. As Amper was most focused on being a stock music alternative, it was primarily background music: no singing or lyrics. Artists have used Amper to get inspiration, or even generate some or all of the backing tracks for them to write melodies on top of, such as Taryn Southern. Even though the product was not tailored specifically to artists, creative people found ways to use it to expand their owncreativity.
Recent Goings On
Over a year since the iconic electronic music duo Daft Punk broke up, an interview came out citing that one of the major factors of their decision to disband was due to advances in AI. I have been a huge fan of Daft Punk, and even though the news of their break up hit me hard, I was even more upset that AI, a field that I actively contribute to, was part of the cause.
Shortly after that, two instances of AI generated vocals popped up. The first was an artist rapping with an AI version of Jay Z (here is a video by MKBHD discussing his reaction to the track). Then there was the fictitious collaboration between Drake and The Weekend. In both cases, many artists and listeners have come out with strong condemnation.
My Response
With respect to the fake vocals, I would not classify this as "AI music" so much as I would say that deepfakes can now sing (or rap) convincingly. I would caution against associating the creation of synthetic vocals with AI music as a general statement. These are some of the most convincing instances of deepfakes in music, but there have been deepfakes of speakers and video for a long time, but that has not led to general condemnation of all of speech synthesis and video generation. There are plenty of perfectly reasonable uses for this kind of technology (Hatsune Miku being a classic example). What should be condemned, like with all deepfakes, is the unauthorized use of someone's identity. I hope that everyone can separate this particular use from their opinions on AI music as a whole.
As for Daft Punk, while I understand and respect their reasoning, I would like to offer my more optimistic outlook on the future of music in the age of AI. Like I said at the start, I started working on generative music as a means of extending my own expression, not as a way of replacing it. Everything that I worked on professionally has been geared towards augmenting creativity, not replacing it. I frequently say that AI will allow us to automate parts of the artistic process that we do no want to spend our time on anyway, giving us more time to focus on the truly creative work that we ultimately want to be doing.
AI will make some tasks more efficient, it will make others obsolete, while at the same time creating new opportunities, just as all major technological advancements have in the past. What it will never do is fundamentally replace creativity. As soon as AI is able to faithfully reproduce a genre, people will come up with a new genre. If anything, this will create a game of whack-a-mole that could inject more novelty into our art.
I often cite this cartoon regarding one person using AI to write an email from a single sentence, and someone else using AI to summarize the email back to the original sentence and ask: in that world, do we even value the full email? It is not so much a question of whether it should be possible or whether people should do it, but more about whether we as a society should continue to value long form writing if a summary would suffice? This does not mean the death of writing, however it does mean that we need to reevaluate what about writing we find valuable, and what parts of that do we believe are only valuable coming from a human.