How I Got an AI Theme Song for My Substack

Or why AI music in 2023 isn’t about innovation or creativity—it’s just a crude cost-cutting tool

I recently got a phone call from a power broker in NY arts and culture—a guy that can make or break careers like a bartender makes and breaks ice.

He had just one question for me: What’s really going on with AI music?

---

The Honest Broker is a reader-supported guide to music, books, media & culture. Both free and paid subscriptions are available. If you want to support my work, the best way is by taking out a paid subscription.

---

He was responding to my recent claim that a respected musician had made a record with a singer, but my behind-the-scenes sleuthing had discovered that this singer didn’t actually exist—it was an AI program.

Yes, that’s a true story. But I didn’t reveal the name of the artist.

My caller wanted to know why this was happening. And, even more important: How is this going to play out in the future? How will AI change music, and how fast will it happen?

“Radio Man” from Popular Science magazine (1939)

---

Here’s how I responded:

“Don’t believe anything you read in the media about AI music. They will tell you how innovative and exciting this is. Those folks are always looking for hip new trends, and they will try to treat AI music like this.

“But it just ain’t true.

“The real leverage point for AI is cost savings. That’s because the music itself isn’t very good. Sure, there’s a certain novelty factor here—but that will wear off very soon. The real hook is that AI works for cheap, it’s almost like slave labor in the band.”

AI doesn’t ask to share publishing rights or composer credits. You don’t even need to buy it a drink or take it out for lunch. Even more important, AI doesn’t even know what the words union scale mean. So it’s never seen charts like this from the NY musicians’ union.

That’s a little odd, because AI is supposed to know everything. But the robots are apparently warned to stay away from the Local 802 AFM website.

“That’s the real story,” I told my caller. “That’s why AI is going to take off and have an impact on the music ecosystem. It’s a cheap date, my friend.”

(Of course, that’s just phase one—I’ll discuss phase two below.)

---

AFM ad campaign from 1930s

---

Let’s consider the case of soundtrack music—especially for student films, training documentaries, corporate videos, indie movies, YouTube videos, and other projects on a tight budget.

Filmmakers want original music in their movies, even if it’s just a student film—it’s a mark of pride. But composers charge anywhere from $50 to $1,000 or more per minute of finished film music. You soon run up a huge tab even with the most frugal use of their services.

And I’m not even referring to superstar movie composers—the John Williams and Hans Zimmer elites who can get $2 million for their music.

Sure, filmmakers could go to a stock music company and buy something off the shelf.  Check out the options here—where you can buy prepackaged albums of soundtrack music for $99.95. 

But even if stock music sounds good, serious filmmakers are embarrassed to use off-the-shelf tracks for their movies.

That might change in the future. I wouldn’t be surprised if the massive re-use of the same backing songs on TikTok will make everybody more receptive to stock music. Youngsters are not only getting accustomed to hearing identical tracks on different videos, but find the repetition comforting. So don’t be surprised if that trend spreads elsewhere.

But right now, most filmmakers want original music. And if you don’t have a lot of money to spend, you must hire the robot.

And the robot is a long, long way from putting John Williams out of business. Williams composes robot music a lot better than the robot composes human music.

My article on “20 Startups That Want to Control the Future of Music” is behind a paywall—but here are two tech companies I included on the list.

SOUNDFUL

Soundful recently raised $3.8 million—and this was a rare situation when a music tech startup got money from the entertainment industry, with Disney and Universal Music Group taking a stake in the San Diego company. It’s worth asking why this startup should interest major labels, and I note with uneasiness that Soundful focuses on music composition which doesn’t require human composers. Tracks are created by AI, and the technology “enables anyone to create incredible music quickly.” Here’s an example:

SOUNDRAW

Soundraw allows video creators to buy AI-composed music easily and cheaply. Here’s the pitch on the website: “Select the type of music you want—genre, instruments, mood, length, etc.—and let our AI generate beautiful songs for you.” The company is located in Tokyo and has raised at least 65 million yen. Soundraw is at the forefront of a new movement which I describe as “music without those irritating musicians.”

---

I decided to take Soundful for a test run. Maybe I need a theme song for The Honest Broker. So I set up an account.

I signed up as an “Influencer” and “Artist” and “Social Media Creator.”  I didn’t really like those titles, but “Writer” wasn’t an option—and the others (DJ, Rapper, etc.) didn’t fit at all. 

Then I had to pick my genres, but the choices here were limited too. I wanted to be jazzy and bluesy and totally rock out. Or, as a backup plan, I’d side with Beethoven and Mozart.

But none of those options were allowed. Instead, I picked these three.

Frankly, I was surprised to learn that “SocialMedia” is a genre. But, hey, why not?

I then had to pick a template. I chose one called “Road Trip”—although I’m not quite sure what template means in this case.

There were many other templates, everything from “Real Estate” to “Study.” I wasn’t especially happy with any of the options, and I tried out a few of them. “Road Trip” worked best on my AI test drive.

Then I picked a tempo of 125 bpm in the key of F# major.

That’s all it took—and, voila, I now had a possible theme song for The Honest Broker. You can listen to it, and decide for yourself.

This isn’t quite at the level of Lennon and McCartney—or even the Lennon Sisters and Jesse McCartney. But it’s at least as good as your typical on-hold music at the customer service center. If I had a podcast I could use it without shame, and if I ever switch careers and become a meteorologist, this would work nicely for a rainy day weather forecast.

On Soundful, users can create songs for free, and have some limited use rights. But to own the copyright and have unlimited use for monetization you need an account ($89 per year) and a purchase fee that starts at $50.

You can’t argue with the price. Especially if John Williams is out of your budget.

But this kind of AI composition isn’t very sexy, and doesn’t make for a good media story. The people talking about AI music are serving up grand futuristic visions, and asking the bots to finish Schubert’s unfinished symphony. But the reality on the ground is several notches lower.

Of course, that will change in phase two—which is at least 5-10 years away, by my reckoning.

---

That takes place when AI actually composes hit songs.

At that juncture, a robotized AI sound will begin to enter the mainstream. Instead of cost cutting or deskilling, the goal will be building a large mainstream audience. Similar mainstreaming has happened before, even in our own lifetimes. So we know what it looks like.

It happened with drum machines. It happened with Auto-Tune. And it will happen with AI.

For example, at some point in the late 20th century most music fans decided that they preferred the sound of a drum machine to a real drummer. I’m not in that demographic, but I can’t deny that software now sets the beat for most commercial music. The same thing happened more recently with Auto-Tuned vocals—many listeners, especially young ones, think they sound better than a real human voice.

That’s how these shifts happen. New tech enters the marketplace as a cheap alternative, and gradually becomes the preferred alternative—because the ‘ears’ of the audience have changed.

I once did a gig with a famous trumpeter who admitted in an offhand moment: “Even I’m getting used to the digital imitations of horn sounds. Frankly, I’m surprised when I hear an actual trumpet or trombone on some of these tracks. I’m wondering why they didn’t just use the software.”

That is the inevitable destiny of AI music. Its first successes will be driven by cost-savings, but after audiences grow familiar with it, many people will prefer the new inhuman sounds—especially young listeners. Human intermediaries will merely flip the switches, but the robots will handle the actual music-making.

But not for everybody.

Real musicians and composers won’t disappear. They will still possess an emotional authority that the machines will never quite dislodge.

Here’s a short passage from something I contributed to a new book of philosophical essays. I wrote this more than a year ago, and will probably share the whole thing here soon, but this extract is especially timely:

Duke Ellington once said that the blues was the “music of romantic failure”—and just consider how many hit songs are about that particular way of falling short of expectations.

But can a machine ever even begin to understand such matters? Will AI ever have a broken heart? Will AI ever grieve the death of a loved one? Will AI ever know about the music that helps Alzheimer’s patients or Parkinson’s sufferers or war veterans with post-traumatic stress disorder? Will an AI ever sing a lullaby to its child? Will an AI ever need to pick a song for the first dance at its wedding reception?

And consider even more trivial uses of song, far beyond the experience of software, no matter how smartly designed. Will an AI ever need a shanty to help it hoist the sails on a ship? Will an AI ever embarrass itself at karaoke? Will an AI ever sing in the shower, or along with the radio during a daily commute?

These are limitations that the robot can never overcome. The human element in music will always be beyond its scope—and that’s a large part of what songs are all about.

It’s just like the Tin Man from Oz. What’s missing is the heart.

But that doesn’t mean that AI won’t put a lot of human musicians out of work. And the more those flesh-and-blood performers simplify their songs, the more likely they are to lose their jobs to the robot. The less they play from the heart, and the more they rely on formulas and stylized poses, the easier they will be to replace.

There’s a lesson there, for those human musicians savvy enough to learn it.