Defeating Generative AI: Why I'm Hopeful For The Future of Music
Defeating Generative AI: Why I'm Hopeful For The Future of Music
AI news can give you whiplash. Is the new technology going to end the world, or cure cancer? Will it destroy the livelihoods of artists, writers, and musicians everywhere, or finally provide a reliable way to discover new human music?
The answer is whichever one will make you most likely to click on the article, mostly. But I digress.
What’s clear is that, for better or worse, AI has been used to make music. It seems that it will continue to be used to make more music. That means that we, as music listeners, need to figure out how we’re going to respond.
I’m not going to get into how AI-generated music is made, or the myriad ethical complications this implies, plagiarism and all the rest. AI is complicated, and there are plenty of other newsletters out there to tell you more about it. But I have been thinking about the role AI will have in music, and how it will affect me (and you) as a music listener.
Note 1: for further reading on the impact of AI on artists, check out these links for some interesting takes on the matter (Link 1, Link 2, and Link 3). I won’t get into that topic here, as I’m only focusing on impacts to music fans, but it’s obviously an important issue and I encourage you to read more about it.
Note 2: I’m treating “AI music” as music created totally by AI - written, recorded, produced. There are lots of smaller uses of AI in music that are perfectly innocent, and they are not the subject of today’s discussion.
You might make a case that as long as the music is great, who cares how it was made? If AI makes better music than humans, doesn’t that mean the listener is the one who ultimately wins? Well I have news for you, hypothetical other side of this debate; that argument doesn’t hold water for me. My definition of “better” music means music that I connect with. That means that often the artist’s struggle to express themselves is a key aspect of that connection. The story of the band writing the album, their B-sides and discarded tracks, their fights and their new guitarist’s girlfiend’s influence, all of that is fascinating and affects my appreciation of the music.
In short, I care how the sausage is made.
So for my veggie-forward friends, here’s my definition of sausage for this article:
Sausage: a cohesive product formed out of mysterious inputs that might or might not make me swear off eating meat forever if I discovered the truth.
AI is a fancier way to make the sausage of music. Mystery meat in → convenient, grillable tube out. Get used to that metaphor, because we’re gonna be riding it into the dirt.
Maybe none of us know how sausage is really made. Maybe you care, maybe you don’t. But do you know who really doesn’t care how the sausage gets made? The sausage seller. They just want you to buy as much of it as possible. And you know who normally cares a lot how the sausage gets made? The person eating it.
And I think most people, even non-sausage-eaters, actually do care about how things get made. There are countless documentaries about humans making awesome things - movies, buildings, countries - that are evidence to that fact. How many popular movies have a director’s cut with added explanation about how the scenes were made? There’s a fascination with how things are built, the joy or hard work or sacrifice (or all 3) that went into the final product. For me, it’s fascinating in part because it lets me envision myself creating the thing since I have a glimpse into the process.
This is a key aspect of “the creator economy” and social media. Nobody has to wonder how the sausage is made, because everybody has the means to create anything they want. Photoshop, GarageBand, even TiKTok music remixes. That kind of toolset lets everyone who wants to create something envision themselves as a creator. Their motivation for learning how the sausage is made is so they, too can become sausage makers. So the makers also care a little how the sausage is made, if only in a self-serving kind of way.
But remember what we said before - the only person who really cares how the sausage gets made is the consumer. The seller doesn’t care, and the maker just wants to get the product out the door.
I believe the proliferation of AI music, and other art forms, is not actually driven by the consumer. It’s driven mostly by the sellers - the peddlers of music like TikTok (who “sells” the music in exchange for your attention) or music labels (who just straight up sell it). These entities don’t give a damn how the music gets made; witness the constant rise and fall of trendy songs and pop singers, which are driven by the music peddlers’ endless need for the next big profit grabber. Why would a seller of music care how it gets made? There are only 2 criteria they have to meet - if the method of making music is cheaper, while still satisfying the consumer, they will always choose that option. AI music is undeniably cheaper to make than human-created music. So if all else is equal, including consumer satisfaction, the music sellers will always go for AI music as a solution. Their shareholders demand it.
And what about the music makers? I think they’re likely a mixed bag. Some artists will resent AI because it replaces their own creativity, while others will embrace it because it enhances their own creativity - or grants it where none existed before. Remember the sausage maker’s motivation - to satisfy the sausage sellers. Music sellers are after cheap-to-make music with the only caveat being the consumer must want it. The makers of music, then, are waiting for the sellers to take a stand before acting. But when the music sellers decide what to do, the music makers would have no choice, it seems, but to follow.
There is a real risk, in my simplified situation, that the sausage maker and the sausage seller might get together and say “we’re only making cheap, shitty sausage from now on.” Then what’s the sausage eater going to do? They only have one option now. Buy, or don’t buy.
AI could lead to a similar situation, if musicians allow it to. If enough artists embrace AI as a cheap music-making engine, it’s possible that we end up in a world with only AI music to choose from, AI algorithms to force it on us, and the only choice remaining to us would be to listen or not to listen. That’s a scary thought.
But here’s the good news, the reason music is not really like sausage after all - musicians are music consumers too. And remember - the consumers have the power over the sellers. That’s why I think there’s still hope for music.
AI-created music reminds me uncomfortably of manufacturing.
I worked in manufacturing for a decade. I was a manufacturing engineer - literally the person who figured out how the sausage would get made (except the sausage was cars). The world of manufacturing, aside from being the most soul-crushing environment I’ve ever worked in, nonetheless taught me a lot about what car companies (and other kinds of manufacturers) actually care about.
My time in manufacturing was a similar story to my high school friends who worked at Wendy’s, and would never eat a square cheeseburger again after the horrors they saw in those kitchens. I’d never drive a car from my factory because I knew how it was made. It was good enough to keep people alive and get them from A to B, because it legally had to be, but that was pretty much it.
Why the lack of quality, you ask? Why aren’t manufacturers as obsessed with perfect quality as consumers are? What about those shiny, perfect cars we saw on the commercials? Isn’t it in their best interest to make a perfect product?
Unfortunately, it doesn’t work like that. The goal of the manufacturer isn’t to make the best possible product. The manufacturer’s job is to make a product that meets, but doesn’t exceed, the product specifications at the lowest possible cost. In other words, good enough. My company wasn’t interested in going above and beyond in areas that didn’t make them more money, and so they didn’t. The only person who wants the best possible product is the consumer.
And how do sellers set the specifications for good enough? They look for the simplest, cheapest, or easiest to make thing the customer will buy. They also know that less variation will make manufacturing easier, so they offer as little variation as possible. This is because less variation makes automation easier and cheaper, and saves money. Then, all of a sudden the beautiful hand-crafted automobiles of decades past turn into standardized blobs of grey, white, or black painted metal that all look essentially the same. Automation, driven by desire for profit, leads to a standardized end product.
And AI, when applied to music creation, is just a fancy term for automation.
How sad would it be to think of creating music as little more than manufacturing? To think of musicians as just a gang of AI-powered robots putting musical pieces together to create a standardized product that just barely meets the consumer’s specifications? That’s what an AI-led music industry’s worst possible endgame looks like to me. One day the music landscape would become as boring and monotone as the shipping yard at an auto factory.
Music and manufacturing are different on the surface, but some things hold true in both.
First, consumers dictate how the cars are made by demanding hard-to-make features. In cars, these are features like big touchscreens; in music, these are attributes like unique melodies and lyrics. In both cases, the makers are pushed to their limits trying to keep up - and that only helps the customer. (put a pin in that).
Second, the creation of both cars and music requires high-tech tools to keep up with the times, and with high-tech tools comes the idea of automation, driven by the desire to save money and increase profit margins. This is one of the key drivers for artists to use AI in some cases, just like the executives of a car company are incentivized to automate their production lines. They’re seduced in by the myth that automation is always easier and cheaper, and that quality won’t suffer.
Third, the automotive and music businesses are low margin at many levels - which means that those making the bulk of the profits, aka the sellers, are very driven to cut costs wherever possible. This creates even more pressure to use automation. Less people or time needed to make the product = less money paid per unit. Another often overlooked effect of this is the pressure to keep products as low-complexity as possible - to create the minimum viable product that will sell to the target customers. An AI “musician” would have a much easier time making a good enough 3-minute power-pop ballad (and making money from it) than a 10-minute masterpiece that pushes the limits of music, so that’s what AI would do.
And fourth, the automotive industry and the music industry benefit from controlling what the consumer likes. It’s the cheap sausage example from earlier - if they can control the options that are available, the consumer might never know that there was a better option out there. In cars, this is evident in things like self-driving features; this isn’t a feature many people are actually asking for, yet companies have spent billions developing it because it could make them money (that part’s complicated, but it’s true). This is a similar idea to the music industry’s big players forcing boy bands or hand-picked superstars to the front of the mainstream, regardless of whether they are capable of creating music that people want. No music listener asked for boy bands, and no music listener is asking for AI music. But AI could end up being the next iteration of the boy band - a handpicked moneymaker that the industry forces down our collective throats.
But there are some key differences, too, which I believe will be the saving factors for the music industry to avoid devolving into a manufacturing process.
First, to go back to our car manufacturing example; people who make cars aren’t car people. They’re money people, mostly working for wages or to maximize sales, not because they just love cars with their hearts and souls. The opposite is true of music - it’s art, and is meaningful for the artist, which is a large part of the driver of its value. Of course there’s often a goal to make money, but the greatest are primarily after making art. It’s that deep love that has led to the great music we’re blessed with today, not advances in technology.
Second, the measure of success in music is always changing, while in manufacturing the finish line is set. I’ve automated a lot of things, and if there’s one thing that automatic processes hate, it’s unpredictability of the outcome. Making the same car on repeat is simple, but making ten different models on the same line is exponentially more complicated. That’s why they’d prefer to just make one car, over and over again to infinity. One measure of success. But music isn’t like that - everybody has a different measure of good music. AI, of course, is built to handle fast-changing and complex requirements, but it still requires a definition of success to finish its job. Is success, in AI music creation, defined as a song that sounds like a human made it? Is success making “good” music? Or making money? Or is success simply giving the consumer the bare minimum they’re willing to buy? That answer will change depending on many factors - the consumer, the sales model of the music industry, etc. - but the fact that the target can and will change means that AI, no matter how adaptable it is, will still rely on humans to tell it what direction to run. Humans ultimately decide what good music is.
Third, the music consumer isn’t like the automotive consumer. Buying a car might feel like a hard decision, but the reality is that most cars these days are good enough, and there’s very little differentiation between them, relatively speaking. Whether you chalk that up to industry power or to government-mandated standardization, it’s the truth. That means consumers often settle for good enough, whether that’s driven by price or the fact that every car satisfies their basic needs and there’s nothing that compelling to make them splurge for the deluxe model. That’s not true in music. People don’t want good enough music - the journey of a music fan is one of always seeking the best music possible. Think about the FOMO you’d feel if you found an amazing album from decades ago that you never knew about - that FOMO is driven by the fact that you’re seeking the best possible music, not music that’s just good enough.
Finally, and perhaps most importantly, there is JOY in making music for yourself. I can tell you for a fact that there’s very little joy in manufacturing cars by hand, so the automation encounters little resistance on that front. Automation is used for the boring, dangerous, or unpleasant tasks. But using AI to automate making music, something humans genuinely enjoy, doesn’t seem destined for success. Nobody is eager to weld a car together once robots have taken on that burden, but people will always want to make music, even if AI takes over the industry, even if AI music is somehow better than human-made. We’ll take that job back because we love it, profitability be damned.
So what does this all mean for us as listeners?
Music isn’t a commoditized business - it’s built on differentiation, meaning better music will always trump cheaper music in the eyes of the consumer. And that means that no matter how many corporate boardrooms decide to push AI music into the market, no matter how many people are tempted to use AI to fake their way into being an artist, no matter how high the marketing teams predict the profits could grow if they could just get people to buy AI-created music: the consumer will get what they want, eventually.
So all we have to do as consumers, if we don’t want AI music, is not want AI music.
Tech companies are notorious for pushing “cool” (or “fun to make”) technology that doesn’t solve an existing problem. Remember Google Glass? It was just a cool project that nobody wanted. AI, when applied to music, stinks of that exact problem to me. It’s not solving a problem - music is already high quality, it’s fun to make, and nobody needs a tool to take away the joy of making music except the talentless. So if history tells us anything, it’s that the product of AI applied to music making will not succeed, except maybe in smaller use cases where a real problem/need exists, like mixing or sound cleanup, for example. But the song creation process as a whole - the act of steering the sounds into a cohesive unit - is not where AI will take root. So even if an AI makes the bass sound a little clearer on a track, the song will still be human-created, and I think we’ll be OK with that.
To me it would be a lot scarier if consumers were the ones demanding AI-created music. But I don’t believe that’s the case. Consumers want the best music, not more of what’s just OK, and part of the way they identify the best music is by connecting with how the music was created. Nobody would watch a documentary on the making of an all-AI track.
In the end, AI music is like the mystery sausage at my high school’s cafeteria. Nobody really knows what went into it. And nobody actually wants to eat mystery sausage, so why would anyone keep making it?
Music will be fine. All we have to do as listeners is continue to seek out the music that speaks to our soul. AI won’t be making that leap anytime soon, because our souls want music created by humans like us.
And if AI music does speak to your soul? Well, I have news for you…
Image credit: Possessed Photography on Unsplash
So what do you think? Will AI-generated music actually take over the industry? Or will the music fans and creators band together and stop it? Or is the advent of AI music actually a development to be embraced?