The Rabbit Hole of AI & Music

Above Sam Altman’s desk there's a piece of paper that reads “No one knows what happens next”. And if he doesn't know, then certainly, neither do I. We find ourselves at a pivotal moment in history as the world starts to shift in ways we couldn't imagine. “AI is going to kill the music industry.” We hear people sounding the alarm. “AI music has no soul. It will never do what humans do,” others counter. Both of the above claims, while seemingly contradictory, may hold elements of truth depending on one’s perspective. This ambiguity underscores the urgent need for clearer definitions and deeper inquiry. Which part of the music industry? Every part? How do you define “soul"? How are we discerning between human and machine creativity? Who owns the music once it's made? In exploring these questions I aim to uncover nuances that bring us closer to understanding AI’s true impact on the music industry.

We already run into trouble when we ask whether or not AI can produce music indistinguishable from that created by humans. I believe it already can in some instances, but how do we definitively decide this? Is it when you can't tell who made it? When a song is able to elicit an emotional response in a human? Who is the listener? Who gets to say? What kind of music? Music is already so subjective that one song might move one to tears, while another finds it trite. An AI-generated song could sound wholly unoriginal to me, lyrically and harmonically, but someone else could love it for the very same reasons. For the time being, some genres, styles, and eras might be harder for AI to pull off than others, and especially inventive, original lyrics, but who says original lyrics are required for an emotional connection? And if we say that AI is merely imitating, isn’t it drawing from the data it's trained on, much like an artist might draw from their influences?

Another component that requires a better definition is what is considered AI-generated music and human-made music. At the extremes it may be easier to define this, but things quickly gets muddier as you look toward the middle. I expect sample providers like Splice will soon launch a feature with the ability to generate loops and one-shots from text prompts. A personalized audio file, just for you.

I've already explored this idea using Meta’s MusicGen. I generated several 30-second audio files which I then layered in my DAW and tweaked. I conceptualized the idea, wrote the prompts, assembled the elements, and declared it complete, using only AI-generated audio. So who made the track? What's the difference from using an ethereal pad loop from Splice?

I imagine people's stance on AI-assisted composing or AI-augmented music will mirror the debate of whether to use loops and samples or not. Many creators and listeners shun any use of these premade elements, considering the resulting works lacking in originality and authenticity, and maintaining that the integrity of the art has been compromised. There are also those that use loops and sample packs freely, as building blocks to shape new compositions. Creators in this group typically encounter no philosophical quandary.

This debate will likely remain unresolved. AI in music production will be viewed as an artist's tool kit, allowing people to express themselves musically with minimum barriers to entry. It will also be viewed as undermining traditional musical craftsmanship. Inauthentic, and for many, akin to cheating.

Another tricky point: Once AI-generated music becomes ubiquitous, what will be its perceived value over time? Will it make human-composed music even more valuable due to its rarity? Last week, a colleague told me that if he heard a whole album start to finish, and loved it, only to find out later it was made entirely by AI, he would feel duped. Deceived. Would you?

Perhaps the music itself is good and is able to elicit feelings, but aren’t we still missing the essential connection to whoever wrote it? When listening to a song, we often seek solidarity with its creator, and wonder what they might have gone through to bring it into existence. There is no backstory, no blood and tears, no greater dimension embedded in the music a robot ‘writes’, which may be essential for many listeners. A single piece of music from Taylor Swift is part of a greater story. There is a brand, authenticity, fan engagement, point of view and advocacy that all play into the value and connection to her music.

This leads us to a related question: Are some kinds of music more “at risk” than others? Infringement cases like deepfakes are currently rampant in commercial music, but is that sector of the industry actually the most insulated once we get a handle on those? What about other sectors like production music? Some editors may not care so much about where the music came from, how it was made, and by whom. Their task at hand is to find a foreboding 3-min track for their edit that’s due in 2 hours. Others might care more. Say, for example, it doesn't seem on-brand to source an AI piece of music for a B Corp ad campaign that champions local communities. Perhaps royalty-free libraries are the most vulnerable, as they primarily serve the lower budget B2C sector. If cost, ease, and volume is crucial for these customers, AI-made music might be the perfect solution. I imagine there will probably be a split in library positioning soon, dividing those who take a stand for human-made music from those who secretly embrace the prolificacy of AI, and each will likely find a corner of the market.

Another gray area -- likely the most discussed, and for good reason -- is the topic of copyright ownership. When music is created by AI, determining the rightful owner becomes complex. Is it the company that developed and trained the AI model? Is it the person who spent several hours iterating prompts until they got the desired output? And what happens if the model was trained on publicly available but unlicensed data? What’s the legal standing of its output? And shouldn't the rights holders of the data an AI was trained on be fairly compensated? It’s encouraging to see proactive responses to issues like these from Rights and AI, a platform that advocates for the importance of protecting intellectual property from unlicensed exploitation in the context of AI training, and serves as a place to preserve those rights.

As we contemplate the future of AI in music, it's clear that resolving these issues requires both introspection and collective effort. We’re in the wild west, and it will take human perspectives to decide how this unfolds. We’re only beginning to scratch the surface of imagining what a world with AI will look like, and there seems to be an endless supply of philosophical rabbit holes.

At the very minimum, it requires all of us to re-examine our understanding of creativity, authorship, and what it means to be human. Perhaps no one knows what will happen next, but at the same time, perhaps any outcome is possible, depending on whether or not we let it.

-Ryan Claus