There is no Antimemetics Division is my second favourite SCP article of all time. My favourite is Tufto's Proposal. If you haven't read it, go read it. The next section will be spoilers for it.

The Scarlet King

In that article, the Scarlet King is an anti-scientific entity. It actively resists any attempts to understand it on a mechanical level. It specifically exists because the SCP foundation is out there collecting anomalous "objects", writing down their properties, classifying them. The Scarlet King cannot be contained with a written protocol: writing the containment protocol would change how the Scarlet King behaves.

(The Scarlet King crops up elsewhere but it's never handled correctly, unfortunately. I don't think any of the other writers "get it" beyond "big spooky red thing")

In some sense, this is impossible. As Rationalists, we ought to believe it is. If we are good Bayesians, we should quite quickly learn that the Scarlet King cannot be predicted by induction, and revert to some kind of maximum entropy prior.

But in other ways, it's totally possible. Our predictions can be diagonalized just as well as our actions, because we are deterministic machines. In the prediction market of our minds, all of our active traders can be drained of cash, until all that remains are a few, dead-eyed algorithms spitting out "50% subjective probability on [logical statement 215034]" forever. And who knows how long that will take!

But I'm not here to talk about the Scarlet King! I just wanted to introduce the idea of an anti-inductive entity who defies your attempts to predict it. This is a nice segue into the idea of an anti-optimization utility function, which defies your attempts to maximize it.

On Slop

(Yeah, I wish we'd chosen a word with less antisemitic etymology too. But slop it is.)

Let me think about some things which span the range of least to most sloppy:

• Goya painted a dozen paintings while living alone. They were undiscovered until after his death. One of them—Saturn Devouring His Son, though even the title is inferred from his notes; it had no label—is one of the most recognizable paintings of all time (cw: Saturn devouring his son). There is no feedback loop.
• An auteur filmmaker, Gerwig, perhaps, produces a half-dozen films. Each of them takes time to be produced. The feedback loop between her and her audience is several years.
• A TV show-runner produces a run of a dozen episdoes. On a rolling schedule of writing, shooting, and editing, the feedback loop may be a month or so.
• A YouTuber, MrBeast, produces a new video every few days.
• A recommendation algorithm selects over videos by different creators on the scale of hours.
• The recommendation algorithm is plugged directly into an image generator. The model is updated in real-time.
• Raw wireheading, I guess

I Tentatively Conclude

Content gets sloppier with shorter and stronger feedback loops, with more effort put into optimizing the content directly (as opposed to the higher generators of the content) and when the optimization looks like selection, rather than control.

And... this might be the case even if the content is good! I've seen enough low-quality slop that I instinctively recoil from AI videos when they start to get funny. My opinion on Huel is "If it's not tasty, I'm not drinking it, because it's not nice. If it is tasty, I'm not drinking it, because it's the experience machine."

I think that I value experiences for how they fit into a causal web. And if the causal chain above my experience is just "A functionally omnipotent algorithm optimized for this experience" then I'm not interested.

Which is unfortunate! It means we can't optimize away the slop. If you optimize it, it just gets sloppier.

Slop creep

The way to avoid sloptimization is to seek out domains which aren't sloppable. Mostly, this means things which humans can't optimize over too much.

Making a film has too many moving parts for the feedback loop to be tightened. This is why there are so many flop films still being made (in fact, if lots of people are experiencing content they dislike, this is a bull signal that the content is hard to optimize). Films are less sloppish than youtube shorts (though films can still be pretty sloppish in the modern age, as producers get better at optimizing them).

Even harder to optimize is stuff in the natural world. We can't optimize a saltmarsh at all, so it's very un-sloppish. 

But humans are getting more powerful. Film and books are on the edge of being sloppified. Saltmarshes are next. If humans get infinite power, can we still make art that isn't slop?

The Counter-Curse and the Counter-Counter-Curse

Maybe we can keep finding harder and harder domains to optimize. Maybe we can make a kind of media which is impossible to optimize.

One approach is just to find something really difficult. You can probably only find a few guys willing to let you tattoo your art on their back and have them sit in an art gallery to show it off. 

Another way is to put a massive weight on novelty. By definition, only one person can be the first to do something, so there's no ability to optimize over it.

Combine these two factors and you get the esoteric kind of modern art. It actively resists doing anything that might be optimized over. And on the first-order level, it works. But...

" I got a guy to let me tattoo my own art on his back, and now I make him sit in a gallery for six hours a day, even through covid lockdowns when the gallery is empty." [See above]

"My film won't even be seen for a hundred years."

"My artwork is me destroying my belongings in a department store."

The constraints on their art are all fake, and they're just competing to see who can have the most constraints! Esoteric modern art is just a higher-order kind of slop. Artists are now optimizing over obscure domains which can't be optimized over.

So maybe we can go one level up again? No! I know an ordinal series when I see one. We can always go up another level; the sequence is infinite, and we can always find a level above that infinite sequence, forever. Maybe the right question is "How many layers up can we go before we can no longer see the sloppiness?" 

Infinite Power

Humans will get more powerful (OK, actually we'll probably just die, but let's ignore that for the purposes of this essay). This means we end up with 1: more optimizing power and 2: more ability to perceive the sloppiness. What's the equilibrium?

Maybe our levelling up wins: suppose we find a way to go up for so many levels of (optimizing over constraints on optimizing over constraints on ...) that we can no longer see the sloppiness, and no longer care, and we find the resulting thing beautiful again.

Maybe our perception of slop gets too good: we get so good at seeing the optimizing power behind things that we can no longer find any beauty in the world. The cosmic endowment is a row of butterflies, dried out and pinned to a board, in a museum, in a viral tiktok.

Maybe we stop caring about slop: if the slop is good, maybe I'll decide that ---on reflection---I don't actually mind it if the content is heavily optimized for my preferences, even on a low level. Maybe I'll find a hole which was made for me, and jump right in.