Ted
Gioia has stuck by the conviction that the flubs and gaffes in art are how you
know the art is made by a human. I recall reading an author who mentioned that
if you heard Dave Brubeck perform live that some nights he was on fire and
other nights he played badly and that’s precisely how you know he was
improvising. Frank Zappa answered the
question of how and why some nights he sounded bad by saying that that’s how
improvisation is.
Alan
Jacobs has taken up the idea of AI generated art lately and references an
article by Richard Beard, one which I’m going to cite myself:
https://aeon.co/essays/sure-ai-can-do-writing-but-memoir-not-so-much
Richard Beard
23 January 2026
...
Turing’s Imitation Game paper was published 14 years
after the first Writers’ Workshop convened at the University of Iowa, in 1936.
Turing may not have known, with his grounding in maths at King’s College
Cambridge, that elements of machine learning had already evolved across the
Atlantic in the apparently unrelated field of creative writing. Before Iowa,
the Muse; after Iowa, a method for assembling literary content not dissimilar
to the functioning of today’s LLMs.
First, work out what effective writing looks like.
Then, develop a process that walks aspiring writers towards an imitation of the
desired output. The premise extensively tested by Iowa – and every creative
writing MFA since – is that a suite of learnable rules can generate text that,
as a bare minimum, resembles passable literary product. Rare is the promising
screenwriter unfamiliar with Syd Field’s Three-Act Structure or Christopher
Vogler’s Hero’s Journey: cheat codes that promise the optimal sequence for
acts, scenes, drama and dialogue. In the same way that an LLM is designed to
‘think’, these templates are a form of reverse engineering: first study how the
mechanics of Jaws or Witness made those movies sing, then identify transferable
components for reassembly to achieve similar artistic success further down the
line....
To escape the dead man’s handle of cliché, readers
live in hope for organic associations, speculative leaps and surprise
inferences. Whereas, to an AI, which is fed the answer before the question,
‘surprise’ remains an elusive concept. [emphasis added] This
objection to machine thinking was raised as long ago as 1842 by Ada Lovelace
about one of the earliest computers, Charles Babbage’s Analytical Engine (for
historical context, Iowa’s first creative writing get-together, though
informal, took place in 1897). ‘The Analytical Engine has no pretensions
whatever to originate anything,’ Lovelace observed. ‘It can do whatever we
know how to order it to perform.’ Her italics emphasise the contrast with
human thinking where originality, among artists at least, is a cherished value.
Keep
this idea in mind as we go. If an analytical engine can do whatever we know how
to order it to perform what do we order it to perform.
This is
where Alan Jacobs went with this observation:
https://blog.ayjay.org/diy-culture/
February 4, 2026
...
I think Beard could have made it more clear that
what people learn at the Iowa Writer’s Workshop is very, very different from
what they learn in screenwriting workshops, but let that pass for now. He is
right to suggest that our arts education in general had a mechanistic I/O
character. Thus a point I have made on this blog: LLMs can so easily produce
a classic undergraduate thesis essay because the assignment is already so
formulaic that it might as well be written by a machine. [emphasis added]
...
The world seems to be filled with people who have
certain gifts and certain interests but are continually forced to acknowledge
that the institutions that have been created to foster those gifts and serve
those interests have ceased to do so. Sometimes the misbehavior of large
institutions can spark the creation of new units within them, such as the
creation of the many new institutes and schools devoted to classical liberal
themes and questions. But the failure of many universities to steward the inheritance
of the greatest of books is what moved Zena Hitz to start the Catherine
Project. Many people on Substack are trying to renew the lost tradition of the
literary magazine. If the institutions won’t do it for us, we’ll have to learn
how to do it ourselves. And then maybe these amateurish and improvised
endeavors will eventually develop into new institutions.
Do It
Yourself bids at counterculture emerge as people feel that the academic
mainstream has failed to deliver on an often implicit, and at times explicit, promise to train people up in
the liberal arts.
The
long-term viability of such bids at countercultural or new-institution-forming
communities can’t be foreseen. I have seen Mere Orthodoxy go from being
an online cultural magazine to an actual magazine over the course of twenty
years. I have been reading Mere Orthodoxy for two decades and it’s taken
me a while to realize that if I run with the Matthew Lee Anderson admonition
about being called into questions or think about how to ask a good question,
the wall I have run into is realizing that the kinds of questions Anderson
wants to answer as a ethicist have never been the questions that interest me as
a hobbyist composer and writer.
As to
hopes that Substackers will “change things” I can hardly forget Alan Jacobs’
own skeptical bottom line about what Substack has always been.
https://blog.ayjay.org/substack-vs-indie/
November 29, 2024
...
Let’s be clear about what Substack is: it’s a
blogging platform with a paywall — that’s all. And it’s not the first one:
Medium, among others, preceded it. Substack is just the one that happened to
catch on, largely because it offers the best (i.e. least revenue-extractive,
most flexible) deal to writers. But it doesn’t enable any kind of
writing that isn’t already enabled by the World Wide Web.
Whether
or not a bunch of people on Substack are going to catalyze a New Romanticism or
a new renaissance actually doesn’t interest me.
After
twenty years of blogging about Christian issues and arts issues hardly anything
in the world has me more jaded than talk about cultural renewal or fretting
about cultural stagnation.
Yet
discussing the crises that large language models and artificial intelligence
generated arts present to us really does interest me. I blogged in 2025 with some thoughts about
why the priests of Art Religion regard AI music as a crisis, you cannot offer a
virtual ram or ox or lamb or goat on the altar.
I ran with a figural reading of the Levitical sacrificial rites and
thought about how real sacrificial animals have bowels that get emptied and
part of the sacrificial process is taking that excrement outside the camp.
Artificial intelligence generated art has no byproduct.
For
that post head to the following link.
https://wenatcheethehatchet.blogspot.com/2025/07/why-priests-of-art-religion-cant-accept.html
But I
also had to point out that for people who used the AI band The Velvet Sundown
as an opportunity to take down how bad and formulaic pop music is (and by
implication always was), the reality is that AI was used to generate classical
music back in 2010. AI generated
classical music long before it generated pop songs.
https://wenatcheethehatchet.blogspot.com/2025/07/in-wake-of-velvet-sundown-lets-recall.html
Beard’s
article zeroed in, ever so briefly, on the nature of what large language models
have been programed to do, by being given the canonized correct answers for
what counts as at-least-good-art before the model has ever been prompted a
question of “can you create X in the style of Y with Z influences?” Alan Jacobs
caught this part and pointed out that the Iowa Writers Workshop honed
generations of formulaic “answers” from which a large language model can
draw.
The
crises of what AI generates when prompted is that it feeds our monolithic
monomythic tropes back to us, minus all the digestive work and associated crap
that reveal that a flesh and blood person or group of persons worked on X, Y or
Z.
AI
music presents a crisis of two centuries of Romantic and post-Romantic music
pedagogy in Western culture. I find it
telling that some of the most inventive theoretical debates and explorations of
the last forty years I’ve had the pleasure to come across have not been from
music scholars looking at the 19th century (we don’t need yet more
books about Brahms or Beethoven) but from scholars looking at the 18th
century. As Richard Taruskin put it, the gap between what Haydn actually did in
and with his music and what pedagogues said he did in his music has been one of
the great gaps in the history of Western music education.
The
crisis of AI-generated music could be telling us that the academics who set
down the top-down schematic formulas of the last fifty to even two-hundred
years have come up with answers that a large language model can spit out on
command. No amount of effort to “interrogate”
or “problematize” these formulaic “answers” upon which large language models
draw is going to be the same as coming up with something that surprises us.
Contrary
to the reception history of Romantic music, sonata forms became a lot more
rather than a lot less formulaic. The celebrated masters of the 19th
century, let’s say someone like Chopin, could write a Piano
Sonata No. 2 in B flat minor between 1937 and 1839--wherein the first
theme of the sonata exposition is never brought back in the recapitulation.
That’s considered daring and innovative when Chopin does it. Would anyone say it was daring to omit Theme
1 from a sonata recapitulation when Franz Simon Molitor did that in his Op. 7 Guitar
Sonata in A minor that was likely composed around 1806?.
I have
discussed Molitor’s
Op. 7 sonata in the past but anyone reading this can safely assume the
answer to my earlier question is “No.”
Even if Molitor’s guitar sonata predated Chopin’s piano sonata by three
decades just choosing to not bring back the first theme in a sonata exposition
isn’t particularly daring in itself. Never bringing back Theme 1 in a
recapitulation was relatively common, common enough that Hepokoski and Darcy
have “Type 2” to describe all variations of this formal schema. If a canonized
composer does something that isn’t “textbook” it’s held as proof of “breaking
the rules” and if an obscure also-ran military officer who hosted musical
parties and sometimes wrote music for the guitar did the same thing three
decades earlier it’s largely consigned to Biedermeier era oblivion.
I’m not
saying you should go out and be Simon Molitor fan. I’m writing hundreds of pages about Wenzel
Thomas Matiegka’s guitar sonatas and not Simon Molitor’s because I think
Matiegka’s sonatas are way more fun to listen to. Matiegka
also wrote a sonata form where Theme 1 never comes back and you get Themes 2
and 3 instead.
By now how
this relates to Chopin can be spelled out. Before it became “textbook” to
insist that Theme 1 must always come back in a sonata form
recapitulation, there was no “rule” out there to be broken. Molitor and Matiegka couldn’t be credited as “breaking
the rules” of sonata form if music theorists and pedagogues had not firmly
established that that was one of “the rules”. You can’t break a rule that hasn’t
been made standard -issue, whereas by Chopin’s era that’s how the decision to
not bring Theme 1 back in a recapitulation in a sonata form was received. Ergo
Molitor and Matiegka were has-beens by the time of Chopin and their decision to
not follow the playback that hadn’t been standardized meant nothing.
Contemporary
musicologists critical of the reception history of the long 19th
century on musicology and music history have very real reasons to wonder why we
should be beholden to the academic cliches and tropes of that era.
All of this connects to the use of AI in a very simple way, if it turns out fifty to two-hundred years of scholarly consensus was used to create a reception history of canonized works that constituted an incomplete or wrong “input” then we can’t be too surprised if AI-generated music predicated on those kinds of inputs creates a troublesome or troubling output. The fault is not necessarily with the large-language model, as Alan Jacobs has proposed, as that generations of scholars have been pushing generations of students towards the kind of identikit school work hand-in projects that large language models can now very easily replicate.
No comments:
Post a Comment