I am a fan of Kirk Hamilton's Strong Songs podcast in which he analyzes popular songs, showing you how they work musically. He does not hold back on explanations of rhythms, melodies, arrangements, and sound engineering. It’s at the right level of expertise for me, which means it’s a good deal over my head. But he’s such a great teacher that he opens my eyes even when I’m not 100% sure of what I’m seeing.
Evidence of his excellence: He got me to listen to a song by Björk all the way through.
But.. . .
Fine. I'm allowed to disagree with someone who knows far more about music than I do, and who can play instruments from just about every section of an orchestra. Plus he sings and whistles beautifully. So, I already can't stand the guy. Except he's funny, self-aware, insightful, respectful of others, self-deprecating, and a great teacher.
Anyway, here’s why I’m writing this: He says at the beginning of his discussion of the GoT theme that when he first heard it, he didn't much like it. But upon repeated hearings, as he recognized its complexity, he came to love it.
Thanks to the podcast, I now admire that theme song’s construction as well — yet I still don’t like it. I find it repetitive and embarrassingly genre-based, with its big warlock drums pounding a deadening rhythm. Yes, I hear more in the song now, but I'd still fast forward over it in the unlikely event that I ever re-watch GoT.
But not @KirkHamilton. He didn’t like the GoT theme when he first heard it, but after analyzing it, he now likes listening to it. So here’s my question: Why? You could prove to me that "Tea for Two" is a work of musical genius and I’d still pay to never hear it again.
The answer is pretty obvious: It's analyzing the structure and elements of the song. Consider the four chords it relies on and what’s unusual about them. Notice how the theme goes from minor to major. Do you hear that the strings repeat the melody at double speed as the cello plays it more slowly? Thanks to the podcast, I can now hear that the song is good in ways I’d never have known.
But it still annoys me.
Let me put this differently: In an earlier podcast, Kirk does a great analysis of the song "Satisfied" from the musical Hamilton. He shows how characters' themes are altered and interwoven in ways that reflect their complicated personal relationships. He shows us details of its construction that deepened my appreciation of both the musical and Lin-Manuel Miranda's genius. But if the song grates on your nerves, I doubt he's going to make you like listening to it. Kirk’s analysis only made the song more beautiful to me because I already found it beautiful. His analysis of a song I don’t find beautiful does not make it beautiful to me.
What, then, is that beauty that analysis can deepen but not instill?
For at least 2,500 years, philosophers from Aristotle to Kant and beyond have offered explanations of beauty. I’m not going to pretend to do so myself. Instead, I'd suggest that machine learning’s way of “thinking” about things points in the direction of an answer, albeit not a very satisfying one.
If you asked me to characterize the music I find beautiful, I could tell you some of the big patterns I've come to recognize. In classical music I like to be able to hear each part distinctly, and thus prefer chamber music to orchestral. But I also love some orchestral music. I love the Beatles' “Strawberry Fields” because it is so unpredictable, and "There Are Places I Remember" for its simplicity. During COVID I've been bowled over by a lot of music from northern Africa and the Middle East because of their rhythms and instrumental precision, but I generally don’t like heavy metal music that exhibits the same characteristics.
That sort of answer is unsatisfying for at least four reasons:
- First, it only talks about what I happen to find beautiful, not what beauty is beyond my personal taste.
- Second, I don’t know why I find music with those traits to be beautiful.
- Third, not all music with those traits is to my liking.
- Fourth, I can’t find patterns to my patterns. They seem to be a quite miscellaneous set of characteristics, as if they’d been scattered over me like glitter.
That’s where we may have something to learn from machine learning. I don't doubt that there are meta-patterns of my patterns, and I expect that machine learning will eventually get good at finding music that matches those patterns. I say “eventually” because right now my streaming service insists that I am huge fan of Arcade Fire. But at some point I’m confident it will learn the odd set of factors in the right proportions in the music that pleases me.
The recipe is unlikely to be simple. Rather, it may be that a heavy bass line sounds good to me but only if the song is light on mid-range instruments, heavy on the snares with noticeable gaps in the drumming, and with no tinkly bells. (Of course, cowbells make everything better.)
I suspect the patterns are far more complex than that, though. That’s why I can’t always predict what song or even what type of music I’m going to like. In fact, the patterns may be so complex that the AI can spot them but not explain them. They may depend upon too many factors in intricate interdependent weightings to be intelligible in part or whole by my little human brain, or even Kirk Hamilton's big musical brain.
And what's true of what I find beautiful in music may well turn out to be true of what I find beautiful in people and life more broadly: not a repeating set of characteristics or patterns, but delicate, infinitely complex networks of interplaying signs and clues that, for reasons we will never fully understand, burst our hearts and give meaning to our days.