Can AI make meaningful covers?

You might be tempted to think this is a book post, but recently I’ve gone down several musical rabbit holes. This is one such. Today we’re going to talk about the phenomenon of musical covers.

I was in my yard the other day when my neighbor ran over with his phone out playing a Justin Bieber country cover and wearing indignance. It was an AI cover that to me just sounded like the original. I could hear the country but there was no depth, nuance, or difference to it that was notable. My neighbor talked about how concerned he was that the human musicians had to compete with AI. I felt a curious hollowness. Because if that is the new standard of a good cover of a song — something almost unnoticeably different from the original — then I think we’ve all lost a key joy of listening to music, right up there with obscure lyrical rabbit holes and intertextuality.

What makes a meaningful musical cover?

The key element of a meaningful cover of a song is reinterpretation or reimagination. My favorite covers don’t just take on the trapping of a new musical style. They speak to the meaning of the song from the perspective of the performer. Take, for example, a famous one: Cake’s cover of Gloria Gaynor’s “I Will Survive.”

This cover does a lot of really interesting things. First of all, some of the urgent defiance is gone – Cake’s interpretation is almost relaxed, containing a certain amount of ennui. As opposed to Gaynor’s bright and urgent arrangement, Cake’s voice seems resigned. This song, for me, is characterized by grim resolution tempered with self-deprecation. In contrast, the original is a proclamation, a celebration of resilience. There’s more to be said, but suffice to say these songs feel very different in tone. And no wonder: the singers are bringing very different cultural contexts into their interpretation of the sentiment, one a black woman in the 1970s and one a white man in 1996. The first time I listened to the Cake interpretation, I didn’t even know for sure what I was listening to.

In my mind, this is what the best covers do. They take something familiar, and cast a new light on it, one whose angles and shades change the perception of the shape beneath. The words are, usually, the same. Occasionally there are some editorial additions, but not generally anything that is by volume significant.¹

Can AI do this? Not so far. These choices require a level of intentionality and volition, an attention to detail that even if prompted well in an LLM would likely fall short of the vision. But my neighbor does have a point. If platforms like Spotify and Google are flooded with hundreds of AI covers, will the standout examples from smaller bands ever be discovered, much less rise to the top? How will you find music that helps you think in new ways about your life, about the memories and person you were when you heard the original, about the sheer sentiment of human experience?

I don’t know. But we’re all going to have to figure it out.

1. A notable example of this is Sylvan Esso’s cover of Gillian Welch’s “Everything is Free,” where the singer substitutes “If there’s something that you want to hear/fucking sing it yourself” for the original “If there’s something that you want to hear/you can sing it yourself”. The single word introduces a level of confrontation that is not overtly present in the original song, despite the cover otherwise being mostly true to the original in terms of the arrangement. Ironically, this change is more in keeping with some of the sentiment Welch has expressed in interviews regarding what inspired this song in particular. It’s a song that’s become even more relevant in our current situation. ↩︎