OK, Computer

Listening is thinking, and thinking is human

An Epistemology of Listening

Let me tell you the story of how I found this marvelous—the blurb says “A signature collection of exquisite works for onde Martenot and piano by Olivier Messiaen, N’Guyen Thien Dao, Jacques Charpentier and Tristan Murail, exploring the many voices of this extraordinary instrument … The compositions here exploit every unearthly aspect of the instrument and are beautifully performed: a remarkable and exquisite collection,” and that is 100% accurate and honest—album. I was digging through the virtual bins at Bandcamp, looking for examples of a specific kind of instrumental music to put into a guide I was writing for them, when I somehow ended up on this album. The record doesn’t have anything to do with the topic and won’t be in the guide, and is sort of two clicks away from the central topic, but it looked interesting as hell to me and Bandcamp lets us preview things, so I hit play, and took it all gloriously in.

That’s “music discovery.” I put that in quotes because it’s just another of the business bullshit¹ terms that have infested our language through the tech-mouthpiece-to-end-user funnel since the 1990s. Even twenty years ago, no one who went to a record store to browse through the bins said or thought they were going to perform the activity known as “music discovery.” They were going to look for records, for music. This is not a listener’s nor a music lover’s term, it is a software engineer’s, and it exists because people use streaming services and, like any technology, it has often become a mental and aesthetic crutch. Instead of standing in temporarily for a bum leg, it takes over what had been a useful portion of the brain, where curiosity and adventure met. They use the service to “discover” music for them, and when I see social media accounts talking about the own listening experiences using those words, I know those people are lost to the machine.

Streaming services can play music new to your ears, but only up to a point. They rely on a chimerical quantification of the elements of music that not only fail to capture incredibly important elements of timbre and mood, but can’t work at all with classical music because not one streaming service (with the qualified exception of Apple Classical) has even bothered to build correct data tables for composer/artist and piece/movement details. To see how this works, and why it is so limited as to be nearly worthless,² it’s useful to look back at 2008, when Spotify launched, from 2022 on, the Large Language Model/ChatGPT/etc era.

Garbage In, Garbage Out

Music discovery uses the same method as the so-called “Artificial Intelligence” LLMs, which is to take a bit of information, whether a musical artist/album/song or a command typed into a text box, and use brute force computer processing to run a statistical analysis and return an answer that is the closest to what is most frequently found in its database. Fancy autocomplete is a clear and honest way to put it, the medium just changes from text strings to another track on a playlist. The contents are beholden to the most common pieces of information, so what the algorithm discovers for you is what everyone else is listening to. Its answer will always end up being: conform.

When’s the last time a track picked by a streaming service surprised you? Not pleased you, not made sense, but surprised you, opened up something new? Just as LLMs cannot produce original writing because they don’t think, they can’t recommend anything because they have no taste. And they have no taste because they can’t experience things. It baffles me how little this is pointed out, and after reading this fine editorial in N+1, I’ve come to see it as a problem for how professional context narrows a viewpoint so that one cannot see essential things. For software engineers, that means never using a Turing Test to ask a machine, “what did you do last night?” Because if a machine is going to fool a person into thinking they’re human, they better come up with an answer.

From the outside, what N+1 calls “reading- and writing-heavy fields— chiefly media and academia,” are understandably focussed on text, both the quality of information and of language. They worry about writers being replaced, and certainly the bland and mediocre language of ad copy and LinkedIn posts can be ably replicated by LLMs, because that’s much of their source material. But there are types of writing LLMs will never be able to replicate, like criticism. ChatGPT cannot “think” critically about an album or a gallery show because it can’t put on a record or go out to an event—and even if it could, it can’t react or bring any context. And criticism is reacting in context, not just the accumulation of events but the ability to feel and respond to information coming from the outside.

The takeaway is this: just as using LLMs will make you dumber, so will using streamers for “discovery” will dull your taste and make you a less interesting, more conformist person. Trust the critics to, if not show you things that will please you, at least show you things you wouldn’t ordinarily run into. Because we’re irreplaceable.

Album of the Week

Here’s an example of how experience is irreplaceable. Sometime probably in early fall of 1993, I saw the Borodin Quartet play Shostakovich’s last three quartets. That has since been embedded in my mind and emotional memory, careful and commanding performances that used precise articulation and dynamics so that Shostakovich’s despairing and death-haunted music could come across without driving everyone in the audience mad, like in John Carpenter’s “Cigarette Burns.” They played the last quartet with no lights except for candles burning at the sides of the music stands.

So every single recording I listen to of the Shosty Quartets has this extraordinary context, and so the ones that impress me, that speak to me, that I will listen to again, need to survive contact with that experience. The ones that have are the Borodin’s own on Melodiya, the Emerson’s, the Pacifica Quartet, and this final installment (and complete set) of the Casals. The playing is beautiful, technically and emotionally, there’s great expressive power, understanding, and that comes first, last, and always.

1. A cursory search for that term turns up a series of articles that are trying to train musicians to use streaming services in hopes they will be “discovered.” It’s a fucking exploitation racket.

2. Why You Like It: The Science & Culture of Musical Taste, by Nolan Gasser, the “chief architect” of Pandora’s old Music Genome Project, is perversely fascinating in this regard. In short, he identified types of listeners (in his mind) and typical examples of genres—i.e. already far too complex to quantify—and ends up with things like a “Rock Genotype” without Led Zeppelin and Rolling Stones but with Lynyrd Skynryd and the Arctic Monkeys.