Ted Gioia against AI slop music misses the terrible paradox, that AI music has been an utterly human endeavor from start to finish so what, exactly, is even inhuman about it?

Ted Gioia writes as if he’s been a prophet his whole career.  A prophet of what? Art for human’s sake and he at least can formulate zippy and memorable one-liners that distill his key ideas.

 

Now why should I describe him as a prophet? It’s simple, if he writes that culture is now being controlled by fifty people, and if he believes those fifty people are wrong, then this is the kind of thing that prophets in the Hebrew Bible were canonized for writing about. Whether I consult Jacques Ellul on how prophets could work for royal courts or as confrontational outsiders; or Abraham Heschel on how prophets, if we met them, would turn out to be disturbing and unpleasant people who said terrible things about the terrible things people do and imagined alternative ways of being;  Ted Gioia has fashioned himself as a prophet.  This would even apply to the futurist elements of his thought.

 

This colloquial understanding of prophets makes it easy to say that Ted Gioia thinks of himself as a kind of prophet. Self-identified prophets of this kind can reveal themselves by the way they quote themselves from the past. Take this recent case where Gioia talked about how, back in 1988 he anticipated artificial intelligence being used to make art and said that would be a bad thing. 

 

https://www.honest-broker.com/p/my-warning-about-ai-music-from-1988

…

 

This extract from my first book has been widely quoted over the years. And it helped launch a new approach to art—known today as imperfectionist aesthetics.

 

Imagine T.S. Eliot giving nightly poetry readings at which, rather than reciting set pieces, he was expected to create impromptu poems—different ones each night, sometimes recited at a fast clip; imagine giving Hitchcock or Fellini a handheld camera and asking them to film something—anything—at that very moment, without the benefits of script, crew, editing, or scoring; imagine Matisse or Dali giving nightly exhibitions of their skills—exhibitions at which paying audiences would watch them fill up canvas after canvas with paint, often with only two or three minutes devoted to each ‘masterpiece.’

 

These examples strike us as odd, perhaps even ridiculous, yet conditions such as these are precisely those under which the jazz musician operates night after night, years after year.

 

I published this at the end of the 1980s—a long, long time ago. But even back then I mentioned the possibility that machines might some day make music that was technically perfect.

 

The idea didn’t appeal to me back then. And it still doesn’t today. But here’s what I wrote in 1988.

 

Imagine a computer that has been programmed to compose musical works in any style. Even if the computer produced works stylistically and qualitatively indistinguishable from Mozart’s, we would still be unwilling to consider them as comparable to the Austrian composer’s pieces. The two are incommensurable. Mozart’s works are artistic masterpieces, and the computer’s output, however admirable, is something else entirely. The latter’s perfection no more reflects on the composer’s art than the existence of motor boats affects our judgment of how difficult it is to swim across the English Channel.

 

Thus, not only is our interest in the human element in art a justifiable concern, it is in fact a necessary concern….Art, in the words of the great modern aesthetician Benedetto Croce, is “expressive activity,” and lives and dies by the success of that expression.

 

I wouldn’t change a word of that today. In fact, this is more urgent to understand now than it was back then.

…

 

Gioia has said that mistakes are what make art human but this assertion gets to the core of the philosophical problem, you can only ascertain that somebody made a mistake if you can plausibly claim to know the artist had an intention that could not be realized.  Why does it matter that humans can make mistakes (whatever those are) that AI slop can’t?

 

What, I have to guess, Ted Gioia was contending AI slop art could not do is participate in both sides of the hermeneutic circle?

 

What’s that, you ask?  It’s a concept that shows up in philosophy and theology and one of the most legendary proponents of the concept was the liberal Protestant theologian Friederich Schleiermacher. Andrew Bowie has been arguing for decades that the impasses between structuralist and post structuralist literary feuds has largely come about because people forgot Schleiermacher’s ideas about hermeneutics. Schleiermacher is a foundational figure in Romantic philosophy and liberal Protestant Christian thought alike and he also had ideas about music as a form of intersubjective communication, all of which Bowie has discussed over the last thirty years. I have also been reading Andrew Bowie for his work on Theodor Adorno’s philosophy and what, if anything, we can glean from it. In Bowie’s reading of Adorno, the Frankfurt school godfather of critical theory tells us that we must look at the contradictions within a society to get a clearer sense of what philosophical problems are inherent in them.  AI slop art is probably one of the better case studies of such a contradiction, because you might think that plugging all the best and brightest art humans have ever made into an AI prompt ought to work.  Why have the results of “only masterpieces in” led to “soulless garbage” out? That’s the kind of aporia that would warrant investigation.

The concept of the hermeneutic circle can be described as a dynamic in which you come to some act of communication (let’s say a text) and you distinguish between the formal, technical elements of the text and the range of interpretive possibilities within the text.  You could think of it as “what the text says” and “what the text means”. These two elements are always intertwined in human interpretive and communicative acts.  Schleiermacher proposed that language as language can be construed as objective in the sense that it precedes our existence and we have to use it. It can be subjective in the ways we use it and changes in use can lead to changes in meaning.  Moreover, there must always be an interpenetrating synergy of technical analysis and interpretive engagement.  I’m coming across this a lot in my reading, it’s even showing up as an early concept in Justin Michael Reed’s The Injustice of Noah’s Curse.

For something to even be a mistake we have to have an idea of an intent that was not realized in the language that was used. If there is no capacity for error then all we have is the technical deployment of recombinant language that isn’t necessarily communicating something. It might be, but you have to guess.

 

Well, when Ted Gioia claims the mistakes in jazz are what make the art human he leans on the surmise that intentions can be known and the failure to attain to an ideal is what makes the art “human”. To frame this in light of Schleiermacher’s hermeneutic circle, Ted Gioia has proposed that AI slop has only the technical pre-given aspect of a “language” and there is no interpretive “there”.  The AI slop is the perfect realization of a technical possibility for recombination in a pre-given language but this doesn’t mean that new sentence means anything, or it has the uncanny valley effect or insert-your-preferred-flaw-of-choice-here.

Now it’s one thing to say the second half of the hermeneutic circle needs to be accounted for, to say that we have to have a guess what person A meant in saying X so that person B can understand X to be X. Contrary to “the death of the author” doctrine we have a historical and ethical responsibility to try to understand what person A meant in saying X. We may be free to accept or reject X once we have understood X to be what A meant, but we have to do the work.  A death of the author gambit could say that because the author used words Y that X can’t be what the author meant.  This would be, so to speak, rejecting that person A said X or meant X because words Y got used.  In the second part of the hermeneutic circle, it is possible to infer that person A used words Y to actually convey X based on details such as usage and context.  Philosophers actually talk about the possibility of shouting “Water!” in a theater when you intended to say “Fire!” and people could actually understand that you were trying to warn them about a fire. 

Why would such a silly example matter? It’s the basis upon which we construe intentions vs usage in language.  It’s how we can think about someone saying one thing but meaning another can be applied in a situation. Gioia’s attempted argument hinges on this interpretive capacity among humans.

 

In other words, what Ted Gioia seems to have meant to say is that the problem of AI-produced art is that it only features the technical or “objective” elements of language and skips the interpretation of intentions.  Complete mastery of merely the technical/”objective” side of the hermeneutic circle is complete mastery of nothing. His appeal to the “mistake” is predicated on a concept he has not successfully articulated.  If the gap between expression and intention is what makes for a mistake and that is what makes art actually art then AI slop is more human than human, it has no imperfections and perfectly executes what humans ask of it by drawing upon the best of the best.  So how could AI slop be the satanic counterfeit to real art as extolled by someone like Matthew Arnold?

 

This is where Gioia bungles hard:

 

…

Art is a human expression. That’s why we don’t consider a sunset or rainbow as works of art—no matter how beautiful they are. The essence isn’t the beauty, it’s that persnickety human ingredient.

 

Without it, you don’t have art. That’s true no matter what an AI song sounds like. Or how much money it makes.

 

You don’t need to be a critic or philosopher to figure this out. People call this stuff slop because they know instinctively how different it is from human creative work. They feel repulsion when confronted by it.

 

It’s like visiting a wax museum. The attempt by the inhuman to imitate the human is always creepy.

…

AI is human expression therefore AI is art.  Whatever is actually wrong with AI slop it cannot be that it isn’t human expression.

 

Gioia insists that sunsets and rainbows aren’t art because the lack the persnickety human ingredient. What is that ingredient? A sunset or a rainbow are not works of art because humans didn’t make them. Can’t a human decide to perceive a sunset or a rainbow as a work of art?  The stunt inherent in John Cage’s 4’33” is that whatever you hear during the performance is the music.  That’s not art?  Roger Scruton would say “No”, and he did, but what’s Ted Gioia’s response?

 

As to that persnickety claim about sunsets and rainbows, Gioia would have us forget that Psalm 8 and Psalm 19 were written and canonized into the Hebrew Bible.  That the Psalmist described the heavens as the art and handiwork of God reminds us that poets of the past regarded the cosmos as art, as beautiful, and this idea that the beauty of the universe was a cogent argument for the existence of a creator of the universe is the founding concept of natural theology. Viewing the heavens as part of the art of the cosmos is an observation written down by humans. Ted Gioia’s hoist on his own petard here.

 

As for repulsion, The Velvet Sundown happened, right?  Breaking Rust’s “Walk My Walk” happened, too.  Somewhere, somebody is going to be exposed to AI-generated art and like it for whatever reason.  There’s no shortage of AI slop art out there and people wouldn’t be demanding that there be some way to identify that AI slop is AI slop if everyone could recognize it. My own take on the problem of AI slop is what I called the fact that you can’t offer a virtual ram on the altar of Art Religion. Real sacrificial animals had crap that had to be taken outside the camp and burned. What Gioia called mistakes can be translated into another idiom, crap. AI generated art can be thought of as lab-cloned meat rather than an animal that was bred and raised the old-fashioned way.

 

Speaking of animals, Gioia’s whole notion that art is human expression could be easily dismissed as a form of speciesism.  How do you know that beavers haven’t created their own version of Stonehenge?  Of course, the image of a bunch of beavers assembling Stonehenge is exactly the kind of thing that AI-generated art can provide us now. That Gioia insists upon the persnickety human ingredient keeps begging the question of what that ingredient is and it can’t be “mistakes” without some clarification. The clarification, on paper, is that something is attempted and not pulled off. Not all mistakes are signs of artistry; they can be signs of incompetence. We players of stringed instruments know that you can’t ask for triple stops to be played on a two-stringed instrument because that third string isn’t there. Gioia’s humanistic essentialism starts falling apart the second you actually start thinking about it.

 

Ted Gioia has liked to talk about music as magic. Through the power of music, he claimed, we can all be wizards.  What kind of wizard, though? I think that is what gets closer to Ted Gioia’s complaint about AI slop music. In The Last Unicorn we see a dualism in the story, where there are people who are adept at a magic of seeming and others who have mastered a magic of doing. The first kind of magic fools a lot of people and really is a kind of magic but it’s not as powerful. The second kind of magic is very powerful but is much harder to control and is mastered by fewer people. 

 

Ted Gioia wants to defend the arts as this second kind of magic and I respect that, as far as it goes, but his mistake is in treating the arts as “only” that second kind of magic.  The reality has been that, as Ethan Hein has put it, AI slop predates AI.

 

It seems, so far, that AI slop repackages the already mastered and the already canonized.  Humans can perceive that a gap exists between ideal and execution because interpretation of X by persons Y and Z is the other half of the hermeneutic circle—the circle of interpretation doesn’t just happen with one reader of one text but many readers of many texts.  If one half of the circle entails technical study of what the text appears to say the other half of the circle is taken up with interpretations of intent.  When Pee Wee tumbled off his bike and then told a bunch of kids “I meant to do that” in Pee Wee’s Big Adventure we laugh because we all know he could not have possibly meant to do that.AI generated art can’t have a gap between intention and execution and, as we’ve seen, a gap between intent and execution is an engine of comedy.

 

Part of the hermeneutic circle as a process of interpretation means new people coming to new texts with new interpretations.  I’m reading Justin Michael Reed’s work on Noah’s curse as an unjust act because he has decided to read with the text of Genesis over against various racist readings of the text that have been given over centuries.  Instead of surmising that Noah’s curse was justly said and that Canaan was somehow cursed with cause despite having done nothing, Reed proposes that Noah got drunk and in his inebriation uttered an unjust curse. The new Adam of Noah turns out to be not just as bad as Adam but curses one of his own descendants without good cause.

In theology and philosophy these kinds of new readings or counter-readings to established reception histories are part of scholarship.  AI, so far as I can tell, has none of that. AI could spit out a synopsis of Stuart Hall’s idea of resistance readings but it’s probably not going to do any resistance readings of its own any more than it’s going to embrace a gay camp reading of Adam West Batman of its own initiative.

 

What I think Ted Gioia should have tried to go for is that within the arts we revisit and rethink and that this is what AI slop does not do. Back in the mid-20^th century microtonalist composers questioned the dominance of twelve-tone and integral serial music by just pointing out that tuning systems don’t have to be standardized. Ben Johnston pointed out that the so-called crisis of tonality emerged within a few years of the trans-Atlantic standardization of twelve-tone equal temperament. Ben Johnston developed a set of different tuning systems and then composed variations on “Amazing Grace” where every variation uses a different tuning system we understand the intention to create a musical work on a well-known tune that subjects it to searching variations. We can marvel that he pulled off the feat because there are so many ways to imagine failing to make a set of interesting variations on “Amazing Grace”, first of all, and second of all, because he made interesting variations on what is all too often a hackneyed tune and did so using tuning systems that are not normal.

 

Now AI slop music probably can replicate something like that with a prompt telling it to do something like “Write a string quartet on Row Your Boat and have it be a set of variations where every variation uses a different tuning system and a different mode”.  Now why anyone would want to hear a variation on Row Your Boat in the octatonic scale in extended just intonation I can’t actually answer.  I’m just making this up as an example of the kind of thing that AI generated music obviously can do but it’s something only a human, as best we can tell, would actually want to do.

 

Now I have described Ted Gioia as fashioning himself as a kind of prophet and his neo-Orphic sympathies are something he’s trumpeted over the last ten years.  In his telling of music history everything great and amazing came from magic and sorcery. In Gioia’s telling the evil technocratic algorithm sucks all the life out of everything human. In his kind of folk religion mantic states and trance are the way to connect to reality.  The possibility that technocratic divination was preferred to ecstatic prophecy in the ancient near east because it was considered more reliable is something for biblical scholars and historians of Mesopotamia. Similarly, proposals that prophetic religion collapsed in the wake of the conquest of Samaria and Judea in the cultural memory of Israel led to a shift from prophetic insight to sapiential and scribal traditions comes off like the kind of thing Gioia doesn’t consider.

 

When you approach the history of divination and prophecy in terms of getting that people believed that, if you properly understood and observed the universe, it could be a conduit for wisdom then Gioia’s persnickety human ingredient, itself never defined, crashes up against his fondness for magic and sorcery. As Annette Yoshiko Reed has put it in her work on the evolution of Aramaic scribal literature about demons in ancient Judaism, demonological tracts included astronomical and astrological information; the stars in the heavens were sentient and played a role in charting times and seasons and knowing when things happened helped you navigate life. The gap Gioia presumes to have existed between “magic” and “algorithm” fizzles out if you bother to tackle primary and secondary literature on the history of technical divination. Magic was the science of its day, a point that Ted Gioia can’t really account for.  Magic wasn’t against the algorithm it was the algorithm that was supposed to make human life more predictable and safer.

 

The unsolved tension that runs through the heart of Ted Gioia’s whole project is whether music “is” the magic itself or whether it is a catalyst. Now if Gioia had been saying that the problem with AI generated art is that it takes Schleiermacher’s hermeneutic circle; cuts out the half that involves interpreting human intentions; and then leaves us with just increasingly arcane mastery of technical assemblages of languages I wouldn’t have suggested, just now, that this is what I think Gioia has been trying to get at but can’t.  I could agree with that idea if that’s what he has meant but this pushes things back to a deeper and scarier question, if this slop is the best AI can do who taught it to make music that way?

 

I think there may be a simpler and more troubling explanation for why AI-generated art comes off so badly.  We humans invented AI and fed it the prefab answers and when AI generated art recombines things and spits it back out to us. We collectively fed AI wrong answers which it has then fed back to us.

 

AI is just feeding us the slop we fed into it, even if that slop has paradoxically been the best of the best in every canon of the arts we have accumulated as a species across our entire known history. As a Hebrew prophet once warned, the heart is deceitful above all else and who can understand it?

 

I come at this as a hobbyist composer and a classical guitarist. I have been fascinated by how profoundly theorists have been reconsidering the “answers” we were given since 1820 about how you can even define what a sonata form is. Yoel Greenberg wrote a fascinating book a few years ago called How Sonata Forms: A Bottom-Up Approach to Musical Form. He proposed that the field of formal analysis has been hamstrung by commitment to top-down classification.  A sonata form must have all of ingredients X, Y and Z or there’s no sonata form. Yet for Hepokoski and Darcy to propose five flexible scripts in Elements of Sonata Theory showed that a couple of their scripts lacked things presumed to be necessary components of a proper sonata. Greenberg’s idea was to take a bottom-up approach, and to entertain the idea that a sonata could be defined as having a majority of a flexible array of traits, bottom up and polythetic classification rather than top-down monothetic classification.  If you have two thirds of the traits of  a sonata then the missing third element doesn’t matter. Greenberg proposed that in musical form we can see a kind of convergent evolution, similar to how bats and bees and birds all have the ability to fly but they “arrived” at the capacity for flight in different ways.

 

For me the implications and applications of such a proposal couldn’t be more obvious.  You can write a monothematic sonata form disguised as a set of continuous variations on a 12-bar blues theme. It will have all the behavioral traits of a sonata form and yet, on the page, look like a set of variations on a theme. Another thing you could do is play with the idea of double variation (which shows up in Haydn) as a way to create a sonata form using strains from a ragtime by Joseph F Lamb.  Instead of slavishly sticking to Joplin’s AABBACCDD pattern you could go AABBABB and have the strains take on the tonal trajectory of an 18^th century sonata form. “Peacherine Rag” already has a sonata exposition in it, all you’d have to do to turn it into a Type 1 sonata of the Hepokoski and Darcy kind is to just bring back the B strain in the tonic key.  It’s as simple as that, a sonata without a development from the hands of Scott Joplin.  There’s nothing that requires sonata forms in slower tempos have to have modulating transitions.

 

In other words, I’m saying that if we follow Yoel Greenberg’s lead and approach sonata forms from a bottom-up approach we’ll discover that sonata forms have been hidden, latent within Scott Joplin’s music for more than a century since the King of Ragtime died. But you’d never know that if you took your cues about musical form and musical history from generations of jazz bros, rockists and classical music purists. I suspect that the problem with AI-slop music is it has been governed by a top-down conception of how to make this form or that style.

 

It’s nonsense to claim that the real problem with AI slop music is that it lacks the human element. I can believe that it lacks “mistakes” but mistakes aren’t what make us human (I’ve seen cats miscalculate how far they needed to jump to reach their landing point).  I can believe that AI cuts the hermeneutic circle of Schleiermacher in half so that all we’re stuck with is a feedback loop of technical refinement about the mechanical and pre-given elements of language lead to variations of sentences bereft of communicative intent.

 

To read Ted Gioia is to get a sense that AI reveals that Matthew Arnold’s vision of the liberal arts as a summation of the best that humanity has thought and felt has come back, via AI slop, as some kind of taunting satanic counterfeit. Ted thinks the missing element is “mistakes”. I think Alan Jacobs just scratched the surface of an idea when he pointed out that AI couldn’t be so good at cranking out slop essays if the Iowa Writers Workshop hadn’t honed the formulaic essay to perfection. Adorno was pre-emptively complaining about the long-term impact of American music appreciation programs teaching Americans the worst possible lessons about actual music when he was working on what was only posthumously published as Current of Music. If you tell people Haydn standardized the symphony and the nature of sonata form, he groused, you were shoving two very big lies down the throats of an uneducated American public. The possibilities for deception in reducing Beethoven symphonies to a highlights reel and a glib explanation of “form” was taking shape in American music appreciation programming on TV and radio half a century ago.

 

Adorno’s unheeded warning back then was that if you feed garbage in about how to understand and appreciate music then the garbage out will be a musically uneducated public.  Adorno was an elitist snob of a very high order but if he were alive today to hear about and read about AI slop he would have said “I told you so.” If we want to try to understand how it’s possible for AI to produce slop music despite having been fed the best of the best in every style, Adorno’s suggestion would be that we look at the lies we’ve perpetrated and perpetuated about music in how we talk about it, even down to the nature of what we say it even is.

 

Generations of conventional wisdom from music partisans (in every style about every form) fed the prefab canonized answers to what counts as great music into AI. Then, when AI spat out the results people want to say AI music is slop and that it lacks the human element.

 

But wait. What human element has been lacking? AI has been a human endeavor from start to finish. Humans conceived of AI, designed AI, built AI, fed generations of canonized music into AI and then when AI creates slop music the problem is AI lacks the human element?  That is a contradiction of the sort Adorno would be writing about if he were alive today. He would tell us the existence of AI slop hasn’t just come about because the culture industry is the culture industry. We find ourselves troubled by the inhuman aspect of AI slop music because it so frictionlessly replicates the lies we fed it about what good music is in the slop it produces.