I've been having a hard time sleeping, and this entry is an artifact of that. As you may be aware, I love the Arts—especially music and creative writing. I also love technology—and I think the intersection between these is fascinating. But what treaties and truces do AI and the Arts hold? That's what this entry is about (sort of).

The eloquence in the room

Sleeplessness leads to ideas: some good, some poor, some half-thought and obscured in a mist of brain fog—but all thoughts are valid.

I love thoughts and I love thinking, and sometimes jumping off certain trains of thought is mighty challenging—it's a non-stop track.

Right now it's 2:45am and I'm pretty sure all of my family/friends are asleep and would not welcome a phone call at this time, so I decided to, once again, have a discussion with AI. Tonight's topic: creative writing and AI—and very specifically is there anything that AI can do that humans cannot.

The question

This is my initial prompt:

So I woke up at 2:15am and sleep is impossible sometimes. I was dreaming and had an interesting sort of epiphany that I thought I'd tease out. Less of an epiphany up front give feedback, more... I'm wanting you to function as a concept exploration partner for me. You're also here to sanity check me, as always, in case I'm being dumb or am otherwise overexcited and just cooked too hard in the middle of the night.
As you may know from prior conversation (or just this one who tf knows), I'm interested in creative writing and storytelling and artsy projects that fuse technology and art in a way that's meaningful and interesting. You might also be aware in your knowledge source that there is a MASSIVE anti-AI movement going on that basically is like, ALL AI ART IS BAD.
Frankly, I think to throw it all into the bin is a categorical error. All art is the expression of what could be other than what is. Different tools allow different things. If we use new tools for old projects that make it so much easier, especially during early adoption, usage of those new tools is deeply frowned upon. However, if the tools are used to make things that would otherwise be impossible, I think a case can be made for it.
So with that said: for what is effectively a static project, where once it's done it's done, and no back-and-forth LLM calls during the final piece... what sorts of creative writing projects can be accomplished via AI where they would be simply otherwise impossible without the usage of AI?
The argument is this: AI does not have domain exclusivity over text production; humans have done that for millennia. So creating text by itself (i.e., writing poetry/prose) is not something AI can do humans cannot. What CAN AI do, in terms of storytelling and creative writing, etc., that humans CANNOT do without the usage of AI?
While I can't think of anything that is impossible, I can think of some types of stories that would be incredibly challenging to put together without AI as an assistant. But I'm interested in your perspective first.

Wax philosophical with me and, if you can think of any that satisfy the criteria of "is text and static" and "AI can create this but humans cannot", share with me what you think those could be—or at least the kinds of questions to ask that could lead us to illumination about what those things would be.

Because AI tends to be wordier than I am (which is hard for even me to believe), let me summarize the core concepts and quote the important parts.

Initial response (with bonus commentary!)

This message early on in its response showed clear "understanding" from the Mansplaining-As-A-Service (MAAS):

If the standard is "a human literally cannot produce the final static text without AI," then the room gets almost empty.

The whole thing about it being static is very, very important. If the final product is chatting with an LLM and producing output an LLM can produce, that's not human out of principal, and if the written work is interacting with LLMs on a live basis, then yeah obviously humans can't do that.

However, if the final work is finished and unchanging, the MAAS concedes there's not really anything AIs can produce that humans cannot as far as creative writing goes.

With that in mind, the MAAS reframes the question in a way that I think is kind of fair.

The question cannot be about artifact possibility alone, otherwise the answer collapses into a boring technicality: "if the output-text exists, then a human could type it." We need to look at it as one of these instead:
Could a human have conceived of it?
Could a human have produced it within a sane lifetime?
Could a human have produced it while satisfying the same constraints?
Could a human have produced it by the same process? I.e., what kinds of static text works become artistically available when, and only when, MAAS is part of the authorship process in a constitutive way rather than just a convenience?

Okay, so these are not direct quotes, but you get the point (actually, I'm rephrasing a lot of what it said to push my point—because this whole thought process is best captured through dialogue in the middle of the night for me).

Anyway, the MAAS suggests:

The bottom-line:
There are probably very few, if any, static text works that are metaphysically impossible without AI.
There are many static text works that are effectively inaccessible without AI because AI enables forms of scale, compression, constraint-satisfaction, or corpus-transformation that humans cannot realistically do.
The strongest defense of AI art is not “AI can make text humans can’t,” but “AI can participate in forms of authorship humans alone do not have access to.”

I think the last one is the biggest sum up. Let's tease out what this entails.

AI authorship's signatures: questions

Where does the irreducible artistic value live?

We have a few contenders. I'm going to go through them real quick. At this point, a lot of my takes are my personal preferences, which you may or may not agree with. Either is fine, but both can be a springboard for your own perspective if you haven't cultivated a clear picture yet.

The final wording

The end result of the text is critical to the artistic value of creative writing. At the end of the day, the contents of a book are more important than the Foreword and online review and summary. The painting is more important than the plaque, etc. I think, at a point, there is too much reframing that has to take place to make the writing itself not the centerpiece, at which point designating where the artistry is becomes more of a carnival trick than anything honest.

The procedure

This can be a part of the art, but is secondary to the final wording. Listening to a virtuosic guitar piece written for one string and one string only—the music itself is the most important, but learning about constraints, procedures used to generate the music, etc., can enhance or cheapen how we feel about the piece.

Art allowing revelation

This may take a couple attempts to get the point across, but I think if I compare creative writing and AI as the tool to other media we can get to the core idea well enough.

If paint is a stand-in for AI, can we create art with paint that tells us about paint better than we could without paint? Can AI reflected through art be part of the art? And, because AI is an extension of humanity, can we tell ourselves about different aspects of culture, experience, and so on through art that uses AI in a ways that we wouldn't be able to without AI?

This last one I want to tease out a bit.

Here's an AIdea

One of the approaches that a person trying to make novel creative writing via AI could take is making the AI the centerpiece. What this could look like is creating clashing scenarios that you simply could not do effectively without AI.

17th century French literature takes on the WWE

So hear me out. What if we took an LLM that was minimally trained to such an extent that it knew no historical or factual information about the modern era or the past or could hypothesize about the future—basically just knew English (or I guess maybe French)—and we trained it exclusively and fanatically on 14th century French literature—what would the consequences of this be? How would it affect the output of said model?

Sure we could have it write more French literature that's very 17th century era, but what if we used it to instead create a retelling of the early days of WWE (an American professional wrestling group with storylines that's basically a soap opera with more machismo than sense and more drama than a holiday family dinner where you and your partner are coming out to your parents and immediate relatives as adopted (I don't know how that works, but that's not the point; the point comes after these closing parentheses)). I, for one, would think it would fascinating to hear a 17th century reimagining of any modern media, like The Office.

While, yes, it is most definitely possible to do this without AI (if you haven't, check out Two Gentlemen of Lebowski for an idea of what I'm talking about), the purity of the idea simply cannot be approached without AI.

However, the trouble is that, at a point, the actual art is no longer the words produced; it is the AI itself. The carefully crafted something was not the work—because AI that is based on words is notoriously good at entertaining word output—it's the thing that creates the written work.

It's like if someone made a painting with Mars dust intermixed with Venus dirt and called it "Two Dirty Lovers" and just splashed it on a canvas and added some glitter and macaroni for good measure—the point is not at all the end product; the point is, "Wow, you actually got stuff directly from Mars and Venus and made a thing with it. That's wild."

We can go further, though; you could train an AI on all of Shakespeare's plays and correspondences, on Einstein's, and HP Lovecraft's, on Hypatia's, on Egyptian funerary rites, etc. You could then could craft a story where the AI assists you in writing perfectly imitative correspondence across time and space between unlikely recipients.

I think there's potential for a really beautiful story there.

But also, the AI being involved feels like cheating and less integral to the art itself—especially when AI is responsible for only some of the literature produced rather than all of it.

Spontaneous literature

This is what it's used for most nowadays: non-static literature, like fanfiction. I know easily a half-dozen people personally who have used, on numerous occasions, AI as a means of teasing out fanfiction ideas because they just want more of the story they love. Or they want to feel different angles and nuances of it.

It's conceivable to make a procedural novel generator that is semi-AI powered, and to write 10,000 custom novels for 10,000 people, and each novel can be part of an interconnected web across all of these novels.

But this defies artistic tradition: the idea is to make something amazing, something that is broadly appealing, and to share it with many so we're enriched collectively.

My takeaways and takeoffs

To me, these are my personal evaluations of some key points:

The final text has to carry the weight. If the text cannot survive without proper staging and framing lore, it's too weak to survive in the wild.
Most AI writing is artistically empty because it substitutes labor, not form. AI cosplaying as an artist is insulting and that's all people really know how to do right now. Until we get past this stage in some way or another, the anti-AI art movements will continue.
The best defense of AI writing is not speed or volume, but medium-specific... weirdness. To put it another way, churning out 10,000,000 lines of poetry in record time enhances no one, even if some of the poetry is legitimately profoundly written. We don't care because it's failing per (2). Thus, AI will only ever be seen as a viable medium for creative writing and other art if we can expose the weird and uncanny qualities of the new medium.
Hot take: the strongest AI-native literary works will likely be nonlinear, combinatorial, or architecturally inhuman rather than merely stylistically impressive. I'd expand on this more, but it spells out too much what I'm considering working on as far as a new project goes—because sometimes 2am ideas are too much to pass up I guess.

To synthesize these, we end up with something like: AI use is artistically justified if and only if the finished text is strong on its own, and the knowledge of AI involvement clarifies some formal property of the work rather than merely adding trivia.

This is where taste and aesthetics comes in—which are presently incredibly ill-defined.

To take an earlier example: let's say I train an LLM on 8,000 train schedules and lyrics from John Dowland super depressing Renaissance music. I use that LLM and generate a story that's relatively aesthetically pleasing about the coming and passing of bumper cars and how the sole contact we feel in our hearts is momentary collision, at which we laugh, hurt, and move on.

Cool story—but like, how does the 8,000 train schedules and lyrics part come in? Why do we care? Why does it matter?

Don't know. We need to train our taste I guess.

Final remarks

Me being the person I am and with the technical abilities I do, I want to see if I can take a stab and creating an AI literary work that entirely challenges the notion of what AI literary works can be. As per the earlier remark, I want the work to stand on its own, but also want it to feel even more incredible knowing that AI created it and how it is even more unlikely to have ended out the way it is because of the limitations of AI (there are plenty) and the sheer hassle that one has to go through to use such a tool.

At any rate, that's all from me for right meow, ilu goodbyeee~.