Let’s get one thing straight: AI in music is a good thing.
In other words, the bots aren’t coming for Bach. Or, depending on your perspective, they already did. Artificial intelligence made its first inroads into music as a composer when Lejaren Hiller and Leonard Isaacson developed a composition-generating program that produced the ILLIAC Suite for String Quartet. This was in 1957, mind you. Google’s first-ever AI-powered Doodle enabled users to create their own Bach-style compositions. Sorry, Johann Sebastian. You had a good run, but the bots have graduated from customer service, content marketing, HR, and healthcare, and they’ll take the fugues from here.
Or perhaps not. There is no shortage of misconceptions about automation and artificial intelligence in music. People think it’s a relatively new phenomenon. It’s sixty-plus years old. People think it’s a niche area. It’s an integral part of every major production software on the market. People think it’s only suited for electronic or experimental music. This makes sense, but AI-powered composition software can pass the Turing test by mimicking Mozart, not Brian Eno. And the big one: people think it’s going to replace composers and songwriters. Which will, presumably, turn music into a soulless, dead-eyed endeavor.
It’s not. It won’t. None of the above.
This isn’t exactly a new concern. People are understandably precious about music. It is an art form that most everyone has deep emotional ties to in one way or another. This applies to not only specific songs and musicians but to musical formats themselves. Audiophiles are famously ardent defenders of vinyl and detractors of MP3s. Even for non-audiophiles, the loss of live music with COVID-19 felt personal.
At nearly every point in music history, the introduction of a new technology has been met with pearl-clutching over the purity of the artform and the extinction of human feeling. This applied even to the invention of the phonograph and the dawn of recorded music in the late 19th and early 20th centuries. People feared recorded music would destroy regular people’s desire to learn an instrument and rot everyone’s brains. The rest, as they say, is history.
Artificial intelligence and automation have had a rather benevolent integration into the listener-facing side of music thus far. Every time someone sticks their phone in the air to Shazam a track playing at the coffee shop they’re working at, then streams the album later that day, that’s AI in music at work. AI-powered recommendation services are ubiquitous among streaming services and listener-facing platforms. Every time a fan buys a ticket to a show by an artist they first heard on a playlist generated by an algorithm or while clicking down the list of suggested videos on YouTube, that’s AI in music at work. Premium licensing platforms are also starting to implement similar recommendation algorithms designed specifically with content creators in mind. In all of these cases, artists benefit. In none of these cases are they replaced.
People should feel just as comfortable with artificial intelligence and automation on the artist side of music as they do with it on the listener-facing one. If anything, artificial intelligence takes on the role of a fellow traveler for musicians and producers. If automation and AI were going to completely and irreversibly upend how we write, make, and listen to music, it would have happened already.
We’ve had the technology to make drum tracks that can hit every beat down to the millisecond for decades. Even so, weird, manmade odd-time rhythms persist, and good drummers are praised for the personal feel and style they bring to their music. There’s “Money” by Pink Floyd in 7/8 time. “Spoonman” by Soundgarden and “Demons” by The National are both in 7/4. Beyond rhythm, analog synthesizers are more popular than ever, and songwriters the world over remain hunched over their guitars, searching for a style rather than perfection. They type ideas for lyrics in the notes app and hum melodies into voice memos, our modern-day replacement for the tape recorder and further proof that while the medium has changed, the most basic human and creative elements have not.
If some replacement of human songwriters or producers was happening, it would be a major story unto itself. And yet the most sensationalist stories about AI and music are along the lines of Grimes announcing (via Instagram comment) that her next record will be a space opera about a hyper-realistic artificial courtesan implanted in a simulation for the human inventor of AI. But Grimes, despite herself, is still not a cyborg or an AI program. Her record will still be the product of human creativity. So it goes.
Another rather sensational story circulating is the “completion” of Beethoven’s tenth symphony. In September of this year, a group of scientists and musicologists at the creative AI startup Playform announced that, following two-plus years of feeding Beethoven’s oeuvre and evidence of his creative process into a sophisticated AI program, they’d managed a new, completed version of the symphony.
Completing unfinished works by great composers is not a new endeavor. Robert Levin, a Harvard musicologist was asked to collaborate with Playform on the tenth symphony project due to his previous work rounding out a number of unfinished compositions by Bach and Mozart, perhaps most famously the latter’s Requiem in time for the 1991 Mozart Bicentennial.
But the trouble with Beethoven’s tenth symphony is that he never wrote it to begin with. Mozart at least made it mostly through the Introitus and the Kyrie of his Requiem (plus a handful of drafts of other sections) before keeling over. His wife later contracted former students Joseph Eybler and Franz Xaver Süssmayr to finish the job.
Beethoven, on the other hand, made only a few sketches of the first movement of the tenth symphony before illness brought him and it to an abrupt end. In the intervening centuries, musicologists and historians classified it as a “hypothetical work,” and what sketches did exist were assembled properly by Beethoven scholar Barry Cooper in 1988. The fabled tenth symphony is a subject of perpetual speculation and an April Fool’s joke from NPR Classical.
What artificial intelligence has managed here is nothing short of a technological marvel for the sake of art. And that is spectacular! But it’s no completion, and framing it as such is the kind of misleading language that drives concerns over AI’s place in music. We’ll never know what Beethoven’s tenth symphony would have sounded like because he died. And we’ll never be able to check just how close this AI-finished composition is to the would-be real thing.
I’m no detractor here. This is a thrilling development. It’s very much part of the lineage of AI in music launched in part through imitating Bach’s fugues. This AI “completion” is the closest we’ll ever get to the real thing. In fact, I find its existence reassuring. It proves that AI is a boon to music, but it also reaffirms that AI is an aid to us, not some creativity terminator lurking in the shadows. AI hasn’t replaced Beethoven. Instead, it’s helped us know him better.
AI’s hypothetical entry into the Beethoven legend and catalog in no way diminishes the public’s fascination and endearing affection for the man and the music he completed himself. His fifth and ninth symphonies will remain indispensable to the global musical lexicon to such an extent that people who don’t know anything about classical music still recognize the melodies. Für Elise will still be dreadfully overplayed at piano recitals to the chagrin of parents everywhere.
And – for the naysayers – while the tenth symphony project is proof of AI’s songwriting capabilities, so to speak, it’s not the death knell of human creativity. The program only worked because it was fed Beethoven’s works and was taught the ins and outs of his creative process. Quite frankly, the AI program here only worked in the first place because a human music genius did first. From there, it took an entire team of human musicologists to go through and edit the results, evaluating and re-evaluating what sounded right, and what didn’t. Plus, AI or no AI, we were never going to get any more human creativity out of Beethoven anyway. A rather drastic bout of post-hepatitic cirrhosis in 1827 made sure of that.
The tenth symphony project is so exciting! So exciting, in fact, that it makes the lion’s share of how AI functions within music seem a bit anticlimactic. And maybe that’s the moral of the story here. What AI does in music is practical, useful, helpful, and all rather non-threatening, and nowhere near as dramatic as the fretting over AI-powered composition software cannibalizing songwriting might make it seem.
There’s also a key distinction to be made between AI’s involvement with Beethoven’s symphonies versus its involvement with more atmospheric genres intended as background music. AI may very well take over some of the songwriting duties for playlists filled with hours of lo-fi chillhop beats. That’s not a problem, and we’ll be none the wiser for it.
So however understandable concerns over replacement may be, they’re somewhat misplaced. On the whole, we find automation and AI more on the post-songwriting processes anyway: production, mixing, and mastering. Automation is already standard practice across digital audio workstations (DAWs) like ProTools, Logic, and Ableton. That’s a good thing – it makes producers’ jobs easier, faster, and more productive. It effectively empowers them to do their work better.
Plug-ins that automate various processes fall under the umbrella of VSTs. (VSTs is short for virtual studio technology.) A VST is a virtual instrument or piece of equipment that exists within a virtual studio. Think of it as a virtual fuzz pedal that’s strapped to the software rather than the pedalboard, accessible by a few clicks rather than a foot tap. Tap a fuzz pedal with your foot in the middle of a riff, and it will scuzz it up. Apply it over a recording of a guitar in Ableton, and it scuzzes it up the same way.
The production space is one area that automation has transformed for musicians and producers. Consider the humble but powerful noise gate. Noise gates are devices and VSTs that cut off any signal in a recording that recedes below a certain volume. This controls echo and unwanted noise. This isn’t new technology, but it is technology that, until semi-recently, required a manual process.
But beyond the confines of the studio, automation and AI has transformed music. Sampling used to require the extremely precise handling of multiple vinyl records on a turntable. Automation has turned sampling into a fully digital process. AI has made it easier and more precise than ever too. This process is ongoing, as machine learning technology becomes increasingly adept at distinguishing and dividing instruments within a song a producer wants to sample.
Artificial intelligence has also helped musicians and producers across genres make new sounds. Anyone familiar with the basics of synthesizers knows computation in music is nothing new. Similarly, AI is the latest iteration of music’s longtime electronic and innovative impulse. (Synthesizers, for the record, were another one of those major music technological developments that failed to live up to its antagonistic, soul-sucking, creativity-killing potential. Guitars managed just fine in the synthesizer world, after all. And good luck arguing Kraftwerk at their best aren’t brilliant, particularly when impersonating androids.)
As with synths before it, AI makes it easier than ever to make new sounds at a higher quality. This, in turn, makes music itself more interesting. If that’s not a direct counterargument to the hand wringing over AI in music, then what is?
AI and algorithms have made strides in other parts of the music industry too. The old school A&R man went extinct when the industry took a nose dive thanks to file sharing in the early 2000s. Now fans are the new A&R, lifting unknown independent artists to fame after finding their music deep on recommendation playlists curated by AI. For the handful of A&R folks whose jobs have survived, there’s Asaii, a music analytics platform that helps labels track trends and hear more rising artists. All of these are positives.
AI is everywhere within the music industry, production process, and listener-facing platforms, which makes it hard to predict what’s next. The future of AI in music is not a single development. Rather it’s countless improvements on ongoing developments.
Why not go to the extreme here? Will an AI program ever contribute enough to a song to earn a co-writer credit? And in that case, who gets the royalties? But I digress.
I believe one thing is likely to be certain moving forward. I believe automation and AI will retain their ‘fellow traveler’ status. These technological developments will probably not usher in the next dominant medium of music distribution, whatever it may be. Your average music fan in the 1990s thought CDs represented the end of the line in terms of listening tech. We all know how that turned out. AI will probably not change the precarity of aging master recordings, as demonstrated by the catastrophic 2008 Universal fire that engulfed masters by Buddy Holly, John Coltrane, Joni Mitchell, Nirvana, Snoop Dogg, and Tupac, to name just a few.
AI is not an all-powerful force within music. It is a force for increased productivity, discovery, enjoyment, and creative advancement. In other words, it’s a force for good. Despite being dead for the last 250 years, Bach will survive AI. Engineers will continue using his music to make AI better. Not only that, AI has brought us as close to Beethoven’s tenth symphony as we’re going to get. Not bad for two guys who never saw the dawn of recorded music.