The songs remain the same

Into the universe of posthistorical music

According to a recent study from Deezer, a music streaming company, 97% of people polled “couldn’t tell the difference between fully AI-generated music and human-made music in a blind test with two AI songs and one real song.” If everyone guessed randomly — if it were literally impossible to tell the difference — wouldn’t the figure still be more like 67%? So what is going on here? Do people hear a difference but then invert its significance in their mind? Or do they just reasonably default to labeling anything that sounds like music as “real music” regardless of whatever machines were used to assist in the process, especially under contrived survey conditions when the music they are being exposed to has no other context.

What does “fully AI-generated” even mean here? It’s not like software just spit out these songs without some human developing the systems, training them on human-produced content, and prompting them for song-like outputs according some specific parameters. Human intention put those sounds into those study participants’ ears. And how were the “human” songs chosen? I’m guessing these were not Jandek and Shaggs records. If it were all a bunch of contemporary smooth jazz, would you be more or less surprised by these findings? What if it were EDM? What about if it were a collection of wan, acoustic covers of pop hits from the 1970s, or some other genre populated by interchangeable anonymous performers?

Does it mean anything to distinguish between “AI-generated” and “human-made” when music in general is so often governed by predictable formulas and a limited number of notes, when what makes music “real” to us is not necessarily how it was made and how many synthesizers and computers or whatever were involved, but the conditions under which we consume it? Much of the music I hear in hotel lobbies seems “fake” to me, whereas anything a friend plays for me I will take as “real.” And I can’t tell “real” instruments from electronic ones, live performances from overdubbed ones, loops from human-executed repetitions, and so on.

The study seems premised on a latent idea that consumers will be harmed somehow by “artificial” music, as if this were akin to being sold adulterated baby formula or something. But the line between machine-generated and human-made music is not especially clear or stable, and it’s not especially relevant in light of the larger question of how music is circulated. I think the various forms of automation involved with that have more of an effect on how we appreciate music than the degree of automation involved in how it’s made. Or to put that differently, maybe the line between music production and music circulation is also being increasingly blurred, and the anxiety over AI-generated music could be better understood in those terms: that music can be made solely to be distributed to us, and to render music less social, less of something that people would bother to share or bond over.

I personally don’t listen to nearly as much music as I used to (I can’t read anymore while music is playing, I get too distracted), but I still spend a disproportionate amount of time worrying about my relationship to it. I haven’t fully adapted to the problem presented by having a nearly limitless amount to choose from — to the problem of seeing that as a problem. Like the proverbial Depression-era grandparents who can’t help but hoard food later, I grew up with a sense of cultural deprivation that makes it hard for me (like many people my age, I assume) to navigate an apparent era of abundance. Media technology and media companies have changed faster than my means of formulating aesthetic interests and investments could adjust.

This is hardly a new condition, and I’ve lamented about it in probably dozens of posts before, but now the problem manifests for me more specifically as a crisis of novelty, and of being unable to separate value from it. Knowing that I will never run out of new music to listen to and that no scarcity will ever force me to listen to anything more than once, increasingly I don’t. I treat unfamiliar albums as essentially no different from mystery novels, as if the only satisfaction they should afford is allowing me to move on to the next one. I’m reading the books only to find out how some particular and arbitrary crime was solved, and I’m listening to the albums only to hear how a set of generic conventions, familiar reference points, and influences were worked out in this particular instance. It’s become a kind of procedural listening.

Consequently, my relationship to music has become more ephemeral and perhaps even more like data processing than when I was fixated on assembling MP3 libraries and perfecting the metadata. Now I find about some records or musicians or genres I’ve never heard, cue up a series of albums, listen to them once, and then mentally check them off as completed. More often than not, I am working my way through a “never played” playlist; it’s as if my favorite kind of music has become “new to me.”

None of this prevents me from indulging in dreams of a deeper kind of relationship to music while having these disposable experiences. As a record plays I’ll begin to situate it within a narrative history of the progress of my personal taste. I might be telling myself how great it is and how fortuitous that I now know about it, and that it will certainly become a “staple in my listening diet” or whatever cliché I’m using with myself to make this fantasy seem plausible. I’ll often remember a line that has stuck with me from a Spin magazine article I read when I was in middle school, where the writer joked about gluing a copy of the Soft Boys’ Underwater Moonlight to their turntable because they couldn’t foresee needing to hear anything else for a while. Isn’t that the goal, to love a record that much?

What I think I want in my encounters with art is something that will interrupt the consumption feeds that technology has integrated me with — that rhythm of having a very minor interest aroused and disappointed and rekindled again for a cycle that can persist for as long as I surrender to it. Yet instead, I find myself adapting nearly all my listening habits to that cycle, as if all one could ever want is to say “Next.” As much as I might rationalize this mode of listening as a kind of “omnivorous curiosity,” that omnivorousness seems more like indiscriminateness, and the curiosity like distractability.

So what’s to stop me from taking this approach to listening to what might seem like its logical conclusion and start getting into generative music? I complained in a previous post about generative fiction, in the form of Nanni Balestrini’s 1966 experimental novel Tristano. It seemed totally pointless to me to read a book no one else would ever read, and I couldn’t find any compensation for that pointlessness in the idea that other readers of their unique copies of Tristano and I were playing the same game, participating in the same process. And I felt exhausted by the idea that I was compelled to be the “real author” of the text, having to marshal my own creativity to invest its mishmash of signifiers with significance.

Would it be different with music, given that listening to music doesn’t necessarily entail looking for meanings or indulging any desire to make interpretations? Why not embrace not only the random bleeps and bloops of procedural electronic music but also software-synthesized generic pastiches like Breaking Rust, whose AI-generated music recently hit Billboard’s Country charts? Better still, why not have a set of algorithms produce music ad hoc based on my browsing habits, so my listening can become even more purely feed-like? Every song would be guaranteed to be completely new — shouldn’t this make me feel like I am on the frontiers of the listenable, discovering the heretofore unheard? What could be more gratifying to one’s curiosity than that? And yet, at the same time, I can be fully assured that I am hearing what I am supposed to hear, the music that will magically conform to my tastes while conforming me to social norms or at least to how “the system” understands me. And no idiosyncrasies of any human musicians can interrupt or muddle my direct consumption of the ideology encoded in a particular kind of song, the quality that resonates with who I think I am when I hear it.

In a recent London Review of Books essay about Brian Eno, Ian Penman highlights Eno’s self-professed “obsession” with generative music, which, as Penman points out, is “an odd thing to be evangelical about, somehow both cutting-edge and old hat.” This also describes not just the idea of generative music, which has been around since at least the early 20th century, but the music itself, which is necessarily “new” on one level while being necessarily formulaic on another. Eno, Penman writes, is “so stoked by the idea of having ‘no one definitive version,’ he even asks whether our current habit of listening to favorite pieces of music over and over again will one day seem ludicrous.” My knee-jerk reaction on reading this was to think Eno was the one being ludicrous. But then I began to think about my own problem, how rarely I listen to anything “over and over again,” and I wondered whether I was on Eno’s ludicrous side despite not being all that stoked about it at all.

It would be no more than a fatuous, provocative pose to advocate for listening to infinite versions of the same song rather than trying to develop a canon of music worthy of repeated listening. But viewed from the sort of perspective that media theorist Vilém Flusser sketches out in Into the Universe of Technical Images (1985), all of what seems like the history of music can also be construed as basically all the same song anyway. Once we remove the blinders that are imposed on us by outdated, text-based modes of cognition, we see that everything in the world is just a “heap of particles” that will eventually be configured in every possible permutation. All music is, in that sense, generative music, a “program” that gives arbitrary elements some more or less familiar form, much as photography in his view is basically a systematic concatenation of pixels (and not a documentation of some given reality).

Flusser argues that “technical images” — i.e. media — have displaced our relation to a shared reality; everyone is instead awash in generated representations that have no fixed connection to a base truth, if there were one. Flusser, who works in a similar vein as Jean Baudrillard, is prone to dropping statements like this:

The world has become meaningless, and consciousness will find nothing there but so many disconnected elements. We are, absurdly, in an absurd world. Bending toward the world is therefore an unsuitable stance and must be abandoned.

Media replaces the meaningless and random expression of matter in the world with something that is potentially meaningful, because it can be construed as having been shaped with intention and directed toward some aim and some audience. The content of media, Flusser claims, is not what’s in the images so much as how they are targeted, or “programmed.” The function of images is to program people, to condition how they see certain specific things but also to condition their entire experience of reality to resemble the nature of images, which don’t emerge from some necessarily linear development of causes and effects but can be generated to depict whatever. They don’t speak to “history,” as texts in Flusser’s view do. “What we call ‘history’ is the way in which conditions can be recognized through linear texts” he argues. “Texts produce history by projecting their own linear structure onto the particular situation.” Images instead articulate the world as cyclical, progressing stochastically through all the possible combinations of the stuff that happens to exist. “Current events no longer roll toward some sort of future but toward technical images,” Flusser writes. “Images are not windows; they are history’s obstructions.” (This is similar to Baudrillard’s theories about the “precession of simulacra.”)

With that understanding generalized to the entire population — the view that events happen only in order to make media, and not that media exists to record a history that unfolds independently of it — society becomes unsustainable and superfluous, replaced with atomized individuals programmed with their own bespoke sets of images, existing in a world that requires no outside social confirmation to cohere from moment to moment.

This, of course, is one way to understand what “social media” really is, the replacement of the social with media. Flusser’s work can be mined for prescient descriptions like this:

The penetrating force of technical images drives their receiver into a corner, puts him under pressure, and this pressure leads him to press keys to make images appear in the corner. It is therefore optimistic nonsense to claim to be free not to switch the television on, not to order any newspapers, and not to photograph. The energy required to withstand the penetrating force of technical images would project such a person out of the social context ... However, the reception of technical images does not end the communication process. Receivers are not sponges that simply absorb. On the contrary, they must react. On the outside, they must act in accordance with the technical images they have received: buy soap, go on holiday, vote for a political party. However, for the interaction between image and person under discussion here, it is crucial that receivers also react to the received image on the inside. They must feed it … This feedback enables the images to change, to become better and better, and more like the receivers want them to be; that is, the images become more and more like the receivers want them to be so that the receivers can become more and more like the images want them to be. That is the interaction between image and person, in brief.

And since social media, in this view, isn’t about communicating with other people, social feeds can readily be populated with AI-generated material that more efficiently establishes this isolating (and nourishing) feedback loop.

AI-generated music could work the same way, opening another intensifying feedback loop between the isolated self and the “system” or “ideology” or “power” or have ever one wants to characterize what Flusser describes as “the envisioners.” The generated content absorbs the receiver’s actions to produce more content, distracting them from the fact they can no longer act on a shared world itself, as figured by an idea of history. “Human freedom” goes from an ability to intervene in history to an ability to get more compelling images sent at you. Get stoked.

Arguing against Eno’s “self-interested” promotion of generated music, Penman suggests that “some music is eternally unfinished: you can play it for decades and it can still make you dizzy. (It just happened to me with the Rolling Stones’ ‘Street Fighting Man’.)” In other words, lots of musical works have depth, they perpetually provoke new responses in us, they change with the context in which we hear them, we continually hear different things in them. They are infinitely interpretable because their human makers are truly infinitely complex, as are the other people in the audience who also invest meanings into these works. Generated music may come from a black box with trillions of parameters, but it is not infinite (not even in the “bad” sense) or in any way free. “The paradox of generative music,” Penman writes, “is that you would have to play it over and over again in order to notice any of the gazillion tiny differences. An infinite wibble. The eternal return of the vaguely familiar.” There is a temptation to see “infinity” in pure permutations rather than in the unfathomable depths of the other, the ultimate unpredictability of their intentions, at least in theory and often enough, it seems, in practice. But the “definite version” can be understood as ultimately far more mysterious and compelling and unfathomable than something ad hoc and random, procedurally singular but with no one invested in its being any way in particular.

Though I’m often flummoxed by his jargon and his gestures toward cybernetics, I get the sense that this is what Flusser is getting at as well. The media-centric world pushes all of us “into our corners,” as he says, and encourages us to engrossed by a kind of raw novelty, the “eternal return of the vaguely familiar” that Penman describes. This accounts for most of my music listening, where I listen to something for the first and only time and then break it down into its elements and draw up apropos comparisons to other things I’ve heard. I assimilate to it purely private history, one that is basically ad hoc and driven by random connections and exposures. For example: I recently heard a song from Hank Garland’s Jazz Winds From a New Direction one afternoon and wondered who the vibe player on it was; that set me on a streak of listening to Gary Burton records, up to the ones he made in the 1980s, and then I read that percussionist Jack DeJohnette passed away, and there was some overlap with Burton there, and that sent me in another direction I followed for a while. There was no overriding purpose to any of this, just momentum.

From a Flusserite perspective, I am a “functionary” of the “apparatus” drifting entropically toward “a closed feedback loop,” as “a cloak of endless, eternal boredom” comes to “spread itself over society.” Like many AI critics today and critics of postmodernism before, he is worried here about the “universe of technical images” absorbing everything history had to offer and finally exhausting itself as it works simultaneously to eliminate history. “History is about to dry up, and this exactly because images are feeding on it, because they sit on historical threads like parasites, recoding them into circles.” My listening momentum carries me down this same spiral on a personal level, as I absorb more and more only in order to process it at the level of its surface, its arrangement of elements, denying it could have any depth that would reveal itself in other ways.

The antidote to this would not be to stumble upon some especially deep and great music that compels me to listen to nothing else — there is no miracles awaiting in the hope for the perfect artwork that would unfold infinity before us. It depends instead on changing my relation to what Flusser calls “telematics,” his term for media networks. When he glimpses the future of telematics in 1985 — at an expo where “one could see people synthesizing images on computers, storing them in memory, and transmitting them to others in dialogue” — he comes to the conclusion that once generalized, this “empty chatter” will amount to a “form of distraction at the intellectual, political, and aesthetic level of the nursery.” It will make everyone into narcissistic babies, if not fascists. Flusser claims that

Media form bundles that radiate from the centers, the senders. Bundles in Latin is fasces. The structure of a society governed by technical images is therefore fascist, not for any ideological reason but for technical reasons. As technical images presently function, they lead on their own to a fascistic society.

But because he imagines a sort of top-down model, where the control and programming comes from the “center,” he hopes that a “dialogic” kind of communication that bypasses the center could suspend the “empty chatter” — the feedback loop of consuming content selected or generated for you alone — and have real “conversation” and “creative play” in a “society of players” that was evolving toward becoming a collective mind.

At this point, Flusser sails into visions of cybernetic utopias. For him, there is no way back to the idea of great works and great makers. His hope is in some future participation in a giant decentralized network, which I find hard to give much credence to in the wake of social media, crypto, and AI hype, which all employed similar tropes. Baudrillard’s “fatal strategies” seem more plausible to me, even though they to seem constitutively impossible to implement. I’d rather it be as simple as Penman seems to suggest, that I could just will myself into listening to “Street Fighting Man” again and be dizzied by it, whirled out of my private feedback loop into some larger, untraceable orbit. Maybe if I got it on vinyl.