Music & AI: the revolution will be automated
Sir Brian May, legendary lead guitarist of the even more legendary band Queen, recently shared his thoughts on the future of music with the rise of AI. More specifically, he conveyed his fears that this (fairly) new technology would have such an impact on the industry that there might not be any — human — musicians left in the end. His actual quote is: “I Think We Might Look Back on 2023 as the Last Year When Humans Really Dominated the Music Scene”.
The reason May’s opinion is interesting — besides the fact that Queen is my favorite band of all time — is that the man happens to be a scientist on top of being an artist. Not only did he build the famous “Red Special” guitar he’s been using since the 1970’s with Freddie Mercury and the gang, but he also completed a PhD in astrophysics, a feat rarely achieved by guitar heroes. Or anyone else for that matter.
In other words, an actual scientist with bona fide rockstar status is telling us that AI is a crucial threat to music and its future, at least as far as humans are concerned, so quite fundamentally. Is that the case? Is music eventually doomed with the rise of this new phenomenon? Is there any hope?
In short, a) not really, b) not sure, c) yes indeed. Let’s elaborate.
First off, what exactly is AI? Depending on the context, the technology offers somewhat creative, predictive results based on a mountain of data that was fed into an algorithm. Put more simply, the “intelligence” of an AI model relies on 2 key ingredients: the quality of the algorithm it is based on, and the quality of the data it is fed — over time. Many AI models still give fairly mediocre results, although the average level has been rising sharply in recent years (see: the ChatGPT jump).
Second, how is AI applied to the music industry? There are many different answers to that, but the main use case today is that of virtual creations, i.e. songs being sung by artists who do not exist. Or, rather, artists who do exist but did not sing that particular song — unknown AI artists can also be found, although they tend to attract less attention. There you have it: Freddie Mercury singing an Adele song, John Lennon performing “Karma Police”, Elvis and Eminem brought together for a duet… the sky is the limit.
Or is it? Actually, that sky remains relatively low: in order for any music AI to perform well, i.e. give a result that is truly enjoyable to the human ear — which remains the benchmark in any scenario — the only real option is for the algorithm to actually play off an existing song. Put simply, Freddie Mercury can sing “Someone Like You” because it already exists. The recent controversy that rose around a “new” duet by Drake and The Weeknd, which caused the internet or at least Spotify to break, was based on the same principle: an actual singer performed the song, which then got transformed by an AI.
That is not to say that AIs won’t be able to convincingly “sing” songs from scratch at some point: with enough data and stronger algorithms, there is no reason why they couldn’t. And that leads us to the core element of this whole discussion: is it that bad? Does this mean that human musicians will become a dying breed, as Brian May seems to believe?
The short answer is — no. The longer — very unlikely. And why is that? For a couple of reasons:
Fear factor
Any major innovation, in music and elsewhere, is usually met with the same kind of skepticism at first. Such was the case when DJ’ing and sampling became a thing: people started claiming that computers would take over, forgetting that you need actual samples to work from — and that DJ’s are also very human artists.
Such was also the case when autotune got big: what Cher started as a one-off became an industry standard, which T-Pain caught enough slack for amplifying. Until everybody realized it was merely another tool for musicians to keep experimenting, creating new sonic experiences — and correcting the very human mistakes artists still make. Adele doesn’t use it, but many do.
These are two glaring examples of similar patterns, where a new capability suddenly threw everybody in turmoil and where the reaction was overwhelmingly negative before the industry as a whole started to adopt it. The least we can say is that DJ’s didn’t kill musicians. Neither did autotune: if anything they made musicians out of amateurs…
Napster 2.0
There is another great parallel to be drawn between the digital revolution and AI today. In the late 1990’s and early 2000’s, as Napster and friends began taking over the Internet and kids — including yours truly — realized they could access the entirety of recorded music in compressed form through the network, labels did what they know best: nothing, except try and fight an irreversible trend. It took many years and the rise of third parties — iPods and iTunes, Spotify… — to create new business models for the music industry that were based on the dual revolution of the MP3 and broadband.
Today, labels still exist but definitely do not hold the position they once had, while streamers have become the new core industry player. And musicians have both more opportunities to distribute their creations than ever before — and more difficulties living off of those. Because labels and publishers don’t have the controlling hand anymore, negotiations with streamers are difficult, which could have been avoided had these legacy actors taken stock of the change and actually embraced it from the beginning: a distribution system that cost next to nothing compared to the old system, which involved producing physical discs and setting up complex logistical processes should have caught their eye… The long term — likely — scenario is that musicians have a better chance of improving their profit sharing by working outside of the historical label framework, which is slowly happening.
AI is going through similar steps today: labels and publishers are actively fighting this new technology, yet again in the fear that it will replace musicians. And yet again ignoring the obvious revolution in all of this, from a strict business standpoint: music owners can now create new music at virtually no cost. Leveraging an existing AI model to create a track out of already recorded songs and/or compositions requires a little bit of bandwidth at best. Gone are recording studio times, mixing and mastering costs and the rest of the traditional music production process. Put simply, AI is a potential cash cow for the music industry that is probably greater than what any innovation brought to the field in the past 50 years. Yet, industry players keep fighting it…
Whenever a true revolution comes, most people tend to fear the unknown and focus solely on the negative changes they can observe: automatic cash registers at the grocery store mean that cashiers are out of a job; OCR receipt analysis means that common accountants are now redundant. That is somewhat true: you only need a few humans to check that the process runs smoothly – that customers use the registers properly and that OCR didn’t make mistakes. But, and more importantly, the jobs that are directly threatened are also arguably the least interesting: accountants mass analyzing receipts will likely confirm. And the jobs that are left are far more interesting: accountants today are increasingly advisors helping clients navigate business — and legal — decisions. Something that cannot be so easily automated. Most importantly, innovations of that magnitude actually create more opportunities for more people: developing new software (which requires developers), creating new companies (which means business), all the way to creating entirely new fields (prompt engineering and design with AI). In other words, an overwhelmingly positive-sum game.
Music today is going through that myopic clash with AI, just like accounting is — or rather was. Those who understand the opportunity the technology represents stand to make billions, all the while creating millions of jobs — and content that will entertain the world. Others will protest until they fade away in the rear-view mirror…