Day 3
Wednesday, 2:18 p.m.
For the A.I. portion of this experiment, I gather this time not in a state-of-the-art recording studio but in Conference Room 04-E5-249 at the New York Times headquarters in Manhattan with two bandmates. One is Juliana Castro Varón, our senior design editor for A.I. initiatives. She makes A.I. projects. As for musical ability, she professes to have “zero.”
“I cannot play an instrument,” she says.
Not to worry. Our other bandmate is Juli’s laptop computer, which is like having every instrument and singing voice in the world. On the internet, Juliana pulls up Gemini and Suno. She asks Suno to write a song that has the same themes as “American Muse.” “Make me a song about the thrill of being creative and the thrill sometimes disappears and you don’t know where it goes,” she types. She adds caveats, like “don’t be cheesy.”
Wednesday, 2:19 p.m.
The song is finished!
It’s got acoustic guitar and a woman’s haunting voice that sings: “Make me a song about the thrill of being creative …” and that’s when we realize that Juliana accidentally wrote her song instructions into the program where it asked for lyrics. It’s made a song of her instructions. We’ve wasted nearly a minute!
Back to work.
Wednesday, 2:58 p.m.
By now, we’ve gathered ourselves and clarified our commands. Juli pastes the lyrics I wrote for “American Muse” and asks for “folk, blues.”
15 seconds later
The computer plays a melodic track sung by a slick country-western singer with a voice Jake could envy, and some harmonies that would require a group of backup singers. The song sounds decidedly country, which was not exactly what Juli had asked for. But if it played on the jukebox as the bar was closing up, someone might ask to hear it again. Might. Depending on number of beers consumed.
We modify our prompts. It’s much easier than band members telling a lyricist, “I love it, but don’t quite get it.” This process is clinical and spontaneous. Eventually, Juli had to tell A.I. to stop making country songs, which seemed to be its go-to.
Over the next 40 minutes, we make at least 30 songs. Suno takes mere seconds to spit out songs in genres as basic as pop-rock and as hyper-specific as experimental classical meets Japanese indie pop; Latin trap; and Italian disco. Most are not inspiring, but a few surprise. The computer has endless range. Country, rock and an uncanny ability to sound like Olivia Rodrigo. Regardless of genre, the songs embody a particular equation: Made in an instant, four minutes in length, but only 10 seconds to listen to and dismiss as “just fine.”
But.
Are they just fine, though, or something more or less? I put it to you, as this is, after all, a group project. Now that you know the context, please again lend your voice to our creativity cage match. Does knowing the process change what you hear — which is A.I. and which is mere human — and which do you now prefer?
Readers, please let us know.
