AI & The Music Industry
Improvements in AI have the potential to dramatically alter the music industry
At its core, a song is about capturing an emotion, an ephemeral moment loosely connected and conveyed with words and music.
At 33, during my lifetime I’ve witnessed the music industry change drastically — from CD’s, Napster, Limewire, YouTube, iTunes, BitTorrent, vinyl, to the rise and ubiquity of streaming. I won’t go into the digital disruption of the industry be cause it’s been covered at length numerous times, but after being dragged kicking and screaming into the 21st century, the industry has finally transformed.
The music business has come a long way since the existential doom of the late 2000’s, when physical product was in free fall, torrenting was rampant, and before streaming had thrown the industry a life boat. According to a 2020 research report by Goldman Sachs, global music industry revenues (recorded, live, publishing) are projected to grow from $77 billion in 2019 to $142 billion in 2030. The industry has undergone a renaissance under streaming (issues with royalty payments to artists and songwriters notwithstanding) and has enjoyed the strongest growth in recent memory. But there could be another tectonic shift on the way…
In 1963, at age 15, Ray Kurtzweil wrote his first computer program. He created pattern-recognition software that analyzed the works of classical composers, and then synthesized its own songs in similar styles. In 1965, he was invited to appear on the CBS television program I've Got a Secret, where he performed a piano piece that was composed by a computer he had built.
Over 50 years later, music created by computers using advanced artificial intelligence that passes the Turing Test is becoming a real possibility, and in my opinion — inevitable. We no longer need the man behind the curtain or for Kurtzwell to play the piano.
The most striking recent example that reminded me of how close we are was Jukebox from OpenAI. For those unfamiliar, OpenAI is an artificial general intelligence (AGI) research company based in San Francisco and headed by Sam Altman, and easily one of the most exciting companies and research organizations going — I highly recommend following their projects and blog.
From the OpenAI blog:
We’re introducing Jukebox, a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles. We’re releasing the model weights and code, along with a tool to explore the generated samples.
I’m nowhere near qualified enough to give a detailed explanation on how their neural net was trained, check out the project link on the OpenAI website for a detailed breakdown, but here’s an example of an AI generated track in the style of David Bowie:
There are multiple moments in that song where I bob my head along — it’s not that far from an Ariel Pink b-side (don’t hate me). Check out the other examples and how the tracks evolved with the methodology of the research.
Here’s an instrumental (piano) only sample:
The use of AI and machine learning in the music industry isn’t new, but it’s prevalence is continually growing. Companies like LANDR use AI/machine learning to offer mastering services at scale and have recently branched into beats and samples. Amper aims to provide an alternative to stock and production music libraries by using AI to create tracks according to parameters such as mood, tempo, genre, and instrumentation.
From my vantage point, the production music businesses will be the first area of the music industry to feel the impact. I’m not sure how viable those businesses look when you can create an unlimited number of tracks in a variety of styles, genres, tempos, and instrumentation — with virtually no cost.
Of course, it’s one thing to use AI to create formulaic production music, and another altogether to create genuine art — music that not only passes the Turing Test but connects with the human experience and emotions, but is this really such a huge leap? Studying patterns, melody, music theory, song structure, lyrics, and experimenting with variations — that’s the nature of songwriting and music creation. Is it unrealistic that AI trained on millions or billions of songs could synthesize the process?
“For a songwriter, you don't really go to songwriting school; you learn by listening to tunes. And you try to understand them and take them apart, and see what they're made of, and wonder if you can make one, too.” - Tom Waits
Throughout the modern era, culture was always changing. Things fell in and out of vogue, and human civilization essentially rode cultural waves together. But the internet completely changed this dynamic, it enabled a sort of cultural quantum mechanics where everything is happening at the same time. It enabled individuals to take pieces of culture throughout history and live them simultaneously — the concept of a shared cultural experience based solely on when or where people lived seems completely outdated.
I’ve always wondered how this dynamic would play out with music. What would the evolution of music look like? What would it feel like? What would appear exciting and new to a world that has all of human culture, all of history at its fingertips? More and more, I think AI will play a major role in the evolution of music.
Grimes is perhaps a little more cynical (not to mention her views on AI and communism) :
“I feel like we’re in the end of art, human art. Once there’s actually AGI, they’re gonna be so much better at making art than us.”
Maybe we’re just fooling ourselves, maybe music that resonates with another person is a wholly human domain that a computer could never genuinely match — at best reaching an artificial flavor of the real thing. Perhaps AGI will forever remain a mirage, always just out of reach and around the corner, evading the best and brightest from OpenAI to Deepmind — but then again, maybe not.
Thanks for reading, if you liked this post please subscribe and follow me on Twitter!
Disclaimer: The companies, stocks, and investments discussed in this newsletter are not intended to be construed as buy recommendations and should not be interpreted as investment or financial advice. This newsletter is for informational and entertainment purposes only, and should not be relied upon as a basis for investment decisions. All statements made regarding companies or securities are strictly beliefs and points of view held by myself or contributing guests and are not endorsements or recommendations to buy, sell, or hold any security. The writer may maintain positions in the securities discussed in this newsletter. I do not guarantee the accuracy or completeness of any information and such information may be subject to change without notice. The information provided in this newsletter does not, and is not intended to, constitute financial advice. All images and media are used in accordance with fair use, under the principles of commentary and criticism.