Creating songs with AI is fun, but also unpleasant

SAN FRANCISCO — Fun fact: The closest thing this newspaper has as a theme song is this march by John Philip Sousa, which I'm sure you've heard before. It's definitely a classic, but maybe we can make it even better.

Unfortunately, I'm not a songwriter, so I turned to AI.

This week, Suno, an artificial intelligence startup that lets you create songs just by inserting a bit of initial text, released an iOS version of its app. In doing so, Suno has arguably made it easier than ever for regular people like you and me to make music on the fly.

That probably wasn't good news for the handful of record labels that sued Suno in late June, arguing that the company's tool was only able to generate tracks because it analyzed countless of their copyrighted songs to figure out how to do so. (Suno, for its part, has said its technology is “groundbreaking.”) Still, the app remains live and free to download — at least for now.

And since the app came out a few days ago, what started as a silly experiment to produce catchy, journalistic tunes has become a minor obsession for me. As it turns out, creating entire songs on a whim using artificial intelligence is really fun, but it also started to change my relationship with music in a way I didn't quite like.

Here's what Suno can do and why I felt a little unsettled after living with it.

Getting started with Suno is easy: just create an account, decide if you want to pay extra to create more songs daily, and then start typing 200-character prompts.

Generating these songs can take anywhere from seconds to minutes, depending on whether you've paid for a higher service package. Your requests will always generate two tracks for you to review.

Your taste in music probably differs from mine, but I already knew what my first attempt at writing a new Washington Post theme should sound like. Bright, tinkling guitars were a must, as were meandering, adventurous bass lines and journalistic lyrics.

But when I asked Suno to create just that, what came out was two generic pop-funk tracks that used the words “bright and tinkly” as lyrics rather than instructions.


Stories to keep you up to date

[Listen for yourself: Washington Funk 1, and Washington Funk 2.]

Maybe this genre wasn't right. Next, I gave Suno the following prompt to see if it would copy a particular artist: “Early 2000s Paramore-style pop punk, high energy, female vocals, lyrics about the Washington Post.”

None of the resulting tracks immediately felt like a Paramore pastiche to me, but that might be because Suno completely ignored my desire for female vocals. Still, the songs felt like something I would have heard in high school, and had a surprisingly catchy chorus:

Telling stories we need to know

From the city to the world and back

There is no turning back on its pages”

[Listen for yourself: Postamore 1, and Postamore 2]

I wanted to keep those lyrics (plus a few tweaks) for my final attempt, so I opened Suno's “Custom” mode and pasted them back in for another try. (Interestingly, Suno's website reminds you to only use AI-generated lyrics if you want Suno to build a song around a full set of lyrics; the app doesn't mention this.)

Now for the rest of the instructions. It seemed like the right step to go a little further afield, so I requested that the music style include the following elements: “J-pop, math rock, female vocalist, anime theme, instrumental intro, guitar solo outro.”

And for the first time, I felt like the results of Suno fully embodied my brief—except when both tracks ended abruptly, went silent for a while, and then the faux guitars came back on for one final run.

[Listen for yourself: Washington! Post!! OP1, and Washington! Post!! OP2]

Okay, fine, none of these shows will ever truly replace the “Washington Post March” – but if any of them had a chance, it was “Postamore 2”.

After I finished my AI journalism song tour, I just played around with Suno, writing silly little songs with nonsense lyrics, trying to recreate the style of one-off pieces I loved.

But it wasn't long before I felt like I was using too much of it – and shared the results. My wife was having a rough day, so I sent her a hearty AI song, complete with our silly pet names, to cheer her up. I came up with some really awful rap lyrics and sent a friend four Suno songs built around them, one after the other.

Then it dawned on me – I could easily imagine continuing to write songs and send them to people as carefree as I send emojis.

Music is a force for good, for joy and healing, for activism and reflection. Has this whole sloppy music generation in some way contributed to the devaluation of music in my life?

Max Vehuni, one half of the indie pop duo Slenderbodies, talked me out of it.

“Music is a way for people to express themselves,” he said. “If it's another way for you to communicate with your wife, I think that's really cool.”

Vehuni is clearly not an advocate of AI music doom—he has experimented with Suno and similar services for personal projects and says he sees incredible potential for AI as a tool to improve an artist's writing and production.

He's also quick to admit that this process isn't fundamentally different from what humans do, even though Suno is being sued for allegedly using copyrighted music as training data.

“Artists draw the line and say, 'Well, I'm OK with artists being influenced by me, with people being influenced by me. But when a computer is influenced by me, that's not OK anymore,'” he said. “Is that something you can agree or disagree with? I don't know.”

But that doesn't mean there aren't other things to worry about. The rest of my ongoing unease, for example, stems from the worry that I'm screwing over the artists I love by producing music that sounds a bit like theirs, but isn't.

Fortunately, Vehuni said, Slenderbodies makes most of their money from touring and the band is fortunate to have a fan base that will support them even during the “post-artificial intelligence music apocalypse.”

In other words, choosing to directly support the artists you care about is more important than ever.

Still, he worries about the possibility that record labels could sell their copyrighted song catalogs to AI companies in exchange for access to models that can create synthetic music for which they don't have to pay royalties. Or that streaming services could create and promote their own synthetic artists and reap the revenue. (He's not alone in asking this question.)

It's too early to say how this will all play out. Either way, big tech companies, the music industry, and the rest of us have no choice but to continue to grapple with AI music invading our lives.

“We took it out of the box and I don’t think we’ll ever put it back,” Vehuni said.