So I’m having lunch with two of my adult children, and we're talking about the impact of AI on the job market – in creative domains in particular, given the advent of generative apps like ChatGPT and musicologist Rick Beato’s warning that deepfake pop-rock vocals are about to become a thing.
Both Josie and Trevor are studying and training for creative careers, so this topic is of particular interest. Their older brother Steve, in whose restaurant we are dining, comes by to check on us. He, of course, has an opinion: there are some things AI will never be able to do, deepfakes or no, he insists, so he’s not worried.
On the other hand, I point out, most art and music out there in the world is used for marketing, and AI’s creative output will fool most of the people most of the time.
As I sat there, I flashed back to a lecture I saw AI prophet Douglas Hofstadter give in 1996 at Indiana University. He recalled an experiment performed by music professor Steve Larson, who competed with a computer program to compose a piece that could pass for authentic Bach. His composition and the computer’s creation were played for an audience, along with a real Bach piece, and the audience voted.
Larson was hurt when the audience took his work to be computerized, but was somewhat consoled when the crowd decided that the computer’s composition was authentic Bach.
“Isn’t it interesting,” Hofstadter mused in his lecture, “how stodgy, mechanical Bach can be so effectively usurped?”
As I sat there with my kids, a Queen song came on the restaurant sound system. It was, serendipitously, “Radio Ga Ga”.
“What about that?” I asked the three of them. “Could an AI create a song like that?”
All three – Josie in particular, as Queen is her favorite classic rock band – said no.
“Are you sure?” I pressed. “Listen to it: it’s incredibly simplistic. The rhythm is completely mechanical. The structure is bare-bones. It already sounds like a computer produced it!”
They didn’t take the bait.
“No computer could ever deepfake Freddie Mercury,” Josie insisted.
Now we’re getting somewhere!
Recently, a deepfake vocal of Paul McCartney was dropped into his song “Scared” from his 2013 album New. McCartney was 71 when he sang the vocal in the original recording, but the deepfake version de-aged his voice – making him sound like he was still in his 20s. McCartney’s voice is every bit as distinctive as Mercury’s; doesn’t this prove Josie wrong?
It doesn’t. What sets vocalists like Mercury apart is their ability to weave half a dozen emotional states, rife with nuance and subtlety, into a single three-minute song. In “Radio Ga Ga”, he projects wistfulness, longing, apprehension, disdain and endearment in a seamless progression. Sure, an AI could mimic the sound of Mercury’s voice; but could it mimic the emotions, with all that subtlety and nuance? And even if it could, would it have the slightest idea how to transition between them?
Then there’s the lyric itself. The song both celebrates and defends the AM radio format that birthed the era of rock music, which in turn birthed Queen.
I'd sit alone and watch your light My only friend through teenage nights And everything I had to know I heard it on my radio
You gave them all those old time stars Through wars of worlds invaded by Mars You made 'em laugh, you made 'em cry You made us feel like we could fly
That warm, nostalgic sentiment then gives way to a cloud of cynicism, as Mercury considers the modern musical landscape:
So don't become some background noise A backdrop for the girls and boys Who just don't know, or just don't care And just complain when you're not there
Mercury anthropomorphizes his radio, speaking to it as he would an old friend, and goes so far as to express encouragement:
You had your time, you had the power You've yet to have your finest hour Radio
Finally, he professes his undying devotion:
Radio, what’s new?
Radio, someone still loves you...
That’s quite an emotional journey. Can AI match it?
I gave ChatGPT the following prompt: “Write a pop song lyric, in the style of the band Queen, about nostalgia for AM radio and the decline of modern pop music.”
ChatGPT served this up:
Remember back in the day
We'd turn on the AM and play
All the hits that we loved to hear
It was the sound that brought us near
But now the airwaves have changed
The music's all prearranged
No more soul or rock and roll
It's all just beats that sound so cold
We used to sing along
With every word, every song
The DJs were our friends
And the music never ends
But now the airwaves have changed
The music's all prearranged
No more soul or rock and roll
It's all just beats that sound so cold
We miss the sound of the past
When the music was built to last
But now it's just a shallow game
And we're left with a feeling of shame
But now the airwaves have changed
The music's all prearranged
No more soul or rock and roll
It's all just beats that sound so cold
Bring back the AM radio sound
Let the music once again astound
We'll sing along with every line
And the memories will once again shine.
Well, that’s certainly interesting. But it’s nowhere near the style of Queen; it presents the requested emotional notes with the grace of a dump truck, is littered with desperation rhymes, and reads like the work of a 15-year-old.
I think Freddie’s safe for now.
And I think my kids are, too, at least for the time being. Will AI end up winning Grammies in the long run? Almost certainly; but by that time, my kids will have raised their own kids on Queen...
Comments