Art: A Unifying Model[1]

The Ghost in the Text

My name is Abraham Haskins. I am a cognitive scientist, an engineer, a psychologist, and an AI systems architect. I'm an overeducated millennial with five degrees scattered throughout surprisingly relevant fields for the points I'm going to make, and the most relevant is a PhD.

In this essay I will use plain english to: define a unifying model for art (with accompanying graph), explain the neurology behind how art works, define the mechanism for an incoming AI-driven cognitive catastrophe, fully flesh out a solution, and then close by showing how the application of this model suggests we focus on a specific strategy for helping align AI.

After, I'll abandon the plain english to trace out an equation describing the model and then close with a proposed set of hypotheses and a plea.

Before you read this, let me be clear: I wrote every fucking word of white text on these pages. I know we’re all suspicious of long bits of text these days, and I’m going to spend the next several pages explaining why. But I want to ask for your trust up front: I, a human, wrote this - and I truly want you to read it.

No AI would open like this - it’s both unprofessional and a little embarrassing.

Perfect. Let’s begin.

AI's intrusion into art has quietly and subtly confirmed for us what art truly is: art is compressed intent.

Let’s be precise here. “Art” describes any creation into which a thinking being invested a large concentration of decisions, all in the service of one or more goals [2]. That creation can be a painting, a speech, a book, a bridge, a dance, a makeup application, a headstand or… really anything. Anytime any thinking being makes a high concentration of decisions, and that volume of decisions creates something, that creation is art. [3]

You'll note that definition encapsulates pretty much all of human activity. You are correct. It will not surprise you to hear that under my model anything can be art in the right situation and with the right viewer. We're going to be speaking most about how "artful" something is. It's simple: if more decisions are made in the creation of something, it is more artful (or artsy, if you prefer) [4]. Everyone has their own personal threshold for how densely packed the decisions need to be for the resulting creation to be "art" in their eyes, but we will be instead putting all human activity on a scale of "artfulness." We’re going to make a graph with this, so let’s put that on the X-axis: things on the left are less artful, and things on the right more so.

The Y-Axis defines whether the art is good or bad. Whether art is good or bad depends on whether the artist met the goal implied by the actions taken. [5] This is the pivotal key to appreciation, because you, the viewer, get to decide what their goal was. In doing so, you decide if you agree it was a respectable or worthy goal, and (more importantly) you get to decide if they met it [6].

Art is "good" when you judge the artist's goal as having been met. If you judge the artist's goal to be unworthy of respect but agree they met it, you may not respect the artist, but you will still tend to classify their work as "good art" [7]. A prime example is a skillful speech given by an evil man. Imagine a hypothetical painting that inspired a nation to evil just like Hitler’s speeches did. Would you call that painting “bad”? I feel most would begrudgingly admit that it was technically good in quality - before calling it misguided and evil [8].

Alternatively, failing to achieve a goal worthy of respect results in a "noble failure," like a director biting off more than they can chew. You might agree with some decisions and say parts of it are good, but you wouldn’t call the whole piece "good art."

So when we speak casually about "good" or "bad" art of any kind - what we mean is that the artist accomplished their goals, as we infer them. And that's our Y-axis.

So there, now we have two axes. Let’s draw it out:

The Graph

My (very human) editors suggested I add this section. Previously, I just let that graph stand on its own, but they suggested a bit of exploration here.

Notice how easy it is for you to place pretty much any bit of human activity you can imagine into this graph, and note that it lines up quite well with a second, subjective glance at what is more "artsy" or "artful." For example, you instinctively know that someone who mechanically traces over a stick figure cartoon hasn't made as much “art” as someone who drew a full picture themselves.

Let’s start here. A… or B?

But is this a robust principle - can we subdivide it? What about people who use reference photos? What if you trace over a real photo rather than a cartoon? What if we use a reference photo, but only from memory?

Note that all of these sort of slot easily into the X-axis of the graph. You probably don't have an issue imagining where they could fit unless they’re incredibly close.

We can disagree on the specifics, of course.

Let's do the same thing when moving up and down the Y-axis. Imagine you saw someone go to incredible lengths to flawlessly trace a cartoon using tons of machinery - lasers, and so on. They made the whole contraption. You know exactly what their goal was. They told you what their goal was, in fact. And there's some artistry to be seen in the effort of chasing extreme precision

But in the end - they didn't trace the cartoon properly! It’s super janky looking. Would you call their highly flawed recreation “good art” or “bad art”?

I’d just reflexively call it bad. I think everyone would.

But what if I showed you the exact same tracing - only now from an artist that made stylistic changes on purpose? All of a sudden you look at the imperfections differently [9]. You don't really know the artist's motivation anymore, so you have to guess at why they made those changes, and that guess informs whether you think it's good or bad. Maybe they made a hated politician’s nose larger, and you think that’s a great (or terrible) way to make a point. And that, then, becomes your interpretation of whether the art is good (or bad).

Respecting an artist's competence is not inherently an assessment of beauty; it is the byproduct of reverse-engineering another primate's problem-solving strategy.

Talking about Art

Let’s return to the foundational definition for a moment: art is the product of highly-compressed decision-making [10]. Honestly, the examples are limitless for this concept.

Photography is a great example. I like museums, and I’ve been to one or two that focused on photographers. As part of a tour in the Harry Ransom Center in Austin, Texas, I saw an interesting defense of the medium. As part of a discussion about whether photography “counted” as art – the tour guide dove into explanations of how much care the photographer put into decisions regarding positioning, lighting, film development, and so on [11]. Looking back, I’d swear there was a lot of focus on the importance, the density, and the difficulty of the decisions that were made. Literally, the answer from experts to “prove this thing is art” is reflexively “look at how many decisions went into this thing’s creation” [12].

Counting Decisions

I’ve been describing obvious applications of the definition. Let’s try to break the theory instead. If art is just evidence of decision-making - whether that be a speech or a painting - then let’s address art with very, very few decisions. How can a Zen master drawing their 10 millionth ensō circle be more artful than a child scribbling for days on a single painting?

Pretend this is the 10-millionth ensō circle, please. Doesn’t it somehow make it more interesting?

Compressing human struggle

The answer is the bundling of previous decisions via the process of automaticity. In cognitive science, automaticity is the process of bundling actions into automatic routines, similar to functions in a programming language or muscle memory [13]. It’s the word for what happens when you get in the car at the end of the workday and just appear in front of your house without realizing how you did so.

When a Zen master draws the circle, you’re seeing the weight and the compounded array of decisions that the master has built into their masterpiece. They make decisions about how to control their breath, the angle of the brush, pressure, light, and so on. But they don’t need to think about all of those actions manually. Instead, those decisions were made previously - and effectively the Zen master just needs to activate that muscle memory (“call a function”) in order to reapply all of those decisions once more [14].

These layers add such a compression of decisions to every moment that the child, who has built up no automaticity, cannot match the density of artfulness [15]. So we respect the decision-making, but we include decisions that were made previously.

So the Zen circle does have an answer: baked-in hierarchical compression [16]. Decisions are counted individually, including subordinate and previously addressed solutions. Every decision matters.

Automaticity is the caching of human struggle [17].

But what if you can’t tell that decisions were made at all?

Talking about the Viewer

Let’s talk about the engineer who walks up to an ugly bridge, stares at it for 10 minutes, and declares it to be art.

Tom's Ugly Bridge [sic]

You see, art is indeed in the eye of the beholder. And the eye of the beholder is defined explicitly by the error bars on the two axes of our original graph. First, you may disagree with an expert about how many decisions went into something. I might look at a grey government slate bridge and say that obviously they used boilerplate nonsense: this is about the most uninspired bridge I’ve ever seen. But an engineer who specializes in city bridges might walk up and look carefully at the concrete mixture selected not just for this region, but literally for this street. They might look at the way various columns are made slightly thicker or lighter, deviating from the standard design to achieve goals I wouldn’t even consider [18] [19]. Notice that as I say that, you can feel your appreciation for this non-existent bridge grow. All I did was describe the decisions.

Experts see more decisions in artifacts of their specialty. They understand the goals of the creator faster. They may even disagree with you on how much something is art at all. However, that doesn’t make it good or bad art, just whether it is art.

This explains the scene in the movie where one person might laugh and cheer at an explosion, but the nearby bomb expert stares at it with a tear in their eye and declares it to be art. The bomb expert sees all the decisions that went into the explosion and can appreciate them.

I think everyone does this within their own expertise, you included. My fun, personal example is that I hate it when I hear people trash IKEA manuals, because I specialize in interfaces. Yeah, my specific specialty is interfaces involving AI, but I still can see the artistry in the way IKEA tries to communicate so cross-culturally and effectively. I see little details that could only have been made in service of a particular clever goal, and it makes me smile. I can’t fully appreciate all the decisions they made in creating it, but I do appreciate at least a bit of it.

Ikea manual designers, I see you. Splendid work.

Reverse-engineering the creator

So, here I’m going to define the word appreciation. To appreciate something is to identify the actions taken to create an artifact, then to use those actions to infer the goals of the creator, and to subsequently connect with and learn from the creator if those goals are deemed worthy of respect [20] [21]. I expect you'll appreciate that definition more if you ever glance at it after the neuroscience section, but let's make it much simpler for now: Imagining all the decisions and work put into something is how you appreciate it. You can appreciate a good story by admiring its theming and seeing how the author guided you to feelings of surprise or sadness. You can appreciate the work that your spouse does for you by noticing the care put into your favorite so-and-so. You can clearly appreciate a beautiful painting. It is the act of looking at an object, using the part of your brain that models other people to understand how they created it, and inferring their intention [22] [23]. That is the act of appreciation.

It is the way you are able to empathize with a director, an actor, and a costume designer involved in the making of a movie scene while standing hundreds of miles away from every one of them. You connect with them because you understand the decisions they made, and you can infer what their goals were through those decisions.

Art is learning. In a way, it’s forced learning. By hijacking the automatic process of learning through intent, it forces you to simulate the creator’s mind and to change your mind in such a way that you could make something similar [24].

Because of that, art is also a virus [25].

One day we may learn who this famous plague doctor is

Talking about Aesthetics

Let’s talk a little bit about aesthetics now. I’m going to define this, but first, an interesting perspective that might help make sense of it.

What if there is no free will?

Seriously, most philosophical arguments and cognitive models are equally valid in a world with and without free will. Annoyingly, so is this one.

When you think of things from this perspective, you often come to weird conclusions where individual people act like information distributors in a semi-collaborative node network or something [26]. You act in accordance with your neural programming, pretty much a robot. In this case, we could see a person come by a piece of art, stare at it, and just for reflexive reasons, try to learn how such a thing was created. Then, with that new capability, they can (when useful) create a similar tool, perhaps leaving it aside for someone else to see and copy. Art is passing from brain to brain, slowly evolving in concept. We’re into memetic theory at this point [27].

Strategically including memes as an artifact of your model to justify sprinkling them throughout your essay.

But wait. If there are no decisions, how can we call decision-making “art” anymore? We can’t. And you’re right. When you look at it like this, the art starts to look viral.

So now, as this sort of unfeeling node in this network of intelligences, you have gained the ability in one sense or another to create this thing that you have seen. You’re better at making this thing that you have observed than you were before you observed it. And due to just the natural way you prefer to interact with your environment, you are now more likely to create something like this in the future that can then be copied, and so on and so on.

Effectively, this is an informational or memetic virus. It is a meme that infects your mind and forces you to learn how to make copies of it in the hopes that you might, in the future, propagate it further. Obviously "hope" is the wrong word here, but this idea of memetic theory is not new. Of all people, it was Richard Dawkins in the 70s that pioneered it [28].

Oh, and just to complete the thought with something cool: if art is a memetic virus, artists are the people who specialize in creating the most viral memes. They are teachers, soothsayers, nature's marketers, maybe even plague doctors. Pick your favorite.

The Autonomic Honeypot

Alright, so now let’s define aesthetics. Aesthetics is the honeypot. It’s the trap.

Attention-grabbing, no? Why did you look at that picture?

What possible value could an AI-generated picture of honey give this paragraph?

It is the word for how much an object forces you to stare at it. It is a clever manipulation of your mental shortcuts, heuristics, background processes, and overall subconscious – things like the tendency for movement in your periphery to catch your attention [29]. It is something attention-grabbing (but not necessarily attention-keeping). It is an invitation to force you to appreciate something, to learn from it, and to absorb the knowledge that went into its creation so that you can be a little bit better at creating something similar, thus passing it on.

Now let’s combine these concepts to imagine how it feels to appreciate nature. There are two ways one might “appreciate” nature, and let’s talk about them both. First, you can admire the sun. However, unless you think there was some decision-making that went into the creation of the sun, you can't really "appreciate" it like you do a gift. To appreciate something, you have to acknowledge the decisions that went into the creation of something. If you’re appreciating the aesthetics, you’re appreciating the effort that went into the honeypot's creation [30]. Religious people can appreciate the sun by thinking of the care their god put into its creation, but the nonreligious folks can only admire its brightness or feel gratified by its warmth. Appreciation requires the acknowledgment of decisions, which usually requires a creator.

Appreciating Nature in Two Ways

But not always.

This feels like definitional quibbling, but I wanted to highlight this because of the other way one can appreciate nature: a sort of engagement with natural law, formed in part by the decisions of all its creators. This kind of appreciation is actually very analogous to interacting with other art. It’s simply that you’re attributing all of the decisions made by every one of the creatures to a conglomerate called “nature.” “Nature” isn’t some kind of blind watchmaker—it’s the interaction between natural forces and the shared power of all of earth’s animals' neurons in a network [31]. So a nonreligious natural appreciation might be a blind appreciation of the sunset, without a full acknowledgment of its creator - and a part of your brain seeing the ancient beaver's intentionality in the mountain's position.

It’s an awareness of all of our brains - every human’s, kitten’s, and worm’s - creating. And your brain trying to understand it and learn from it [32].

Extra Credit: Art Mysteries Answered

▶Postmodern Art

Let’s go all the way to Marcel Duchamp’s "Fountain." This was a urinal that he simply bought and then declared to be art [9]. He didn’t make it. He didn’t do much of anything other than pick it out and then sign it with a fake name.

And part of me hates it, but I now believe it was art.

Art in its natural habitat

This is an example of someone pointing out, through acknowledging what art truly is, that the decision-making is itself the key. He demonstrated that the essence of art is the communication of decision. Duchamp made a highly unusual, noteworthy, hard-to-understand, but compelling decision. And then he declared that single decision to be art. In fact, when asked whether he had created the urinal himself, his explicit defense was that whether he made it with his own hands had no importance. He chose it. It was the selection, he said, the choosing, that mattered. The decision [33].

Remember the hierarchical nature of decisions here. If you think about the concept of art, think about how it fits into 1917, think about the meaning of toilets in that time... layer all of these decisions on top of one another, alongside a layer of understanding of the art world and its culture in a way that I, as a non-artist, cannot easily have. All of these decisions compressed into one moment were decided by enough people to be art that you kind of have to look at it and question [34].

And that question is, specifically, "was there intention to this that I don't understand? Was there a goal to this that I can't perceive?"

Humility can be a surprisingly beneficial strategy, neurologically.

That's pretty much always the core puzzle we try to solve at the center of all art and assessment thereof. By definition, if you're doing it then you're appreciating art. And unfortunately Duchamp's Fountain successfully demands that you ask.

Effectively, what Duchamp did with "Fountain" was to boil away all the trappings of art and lay bare its most skeletal form [35]. I suspect that, alongside mockery, that was exactly Duchamp's goal. He was quite successful at accomplishing both his goals, and did so with such a high ratio of effectiveness to action that it frankly beggars belief. Ergo, it is technically at the very least "good art" because it accomplished its goals. It wasn't necessarily artful, though.

▶Cy Twombly's Scribbles

Honestly, so many of the puzzles of art become immediately obvious underneath this framework. We’ve always been curious about why art is in the eye of the beholder, or why someone might say a phrase like, "I don’t get it, it’s just so bad." Those phrases fit perfectly into this model, answering the paradox of why Cy Twombly’s scribbles are seen so differently by society than your toddler’s scribbles: the automaticity layering and the gap between them. We know that a child has no built-in automaticity, and so we don’t seek it [36]. There’s no point admiring a child’s art because there’s no density of decisions to be discerned. They’re still building up their hierarchy, and we know this.

Can you tell which is worth 2.4 million?

This framework helps us immediately identify "correct" art by outright counting how many decisions were made in the process of its creation. For example: modern art is sometimes accused of being a scam. We have a solution to that. If you have The Black Square by Kazimir Malevich, an artist that clearly put in a lot of thought to painting this piece of canvas pure black, alongside a simple recreation of that by someone just painting black on a canvas - you can actually mathematically distinguish why one has high artfulness and one does not [37]. There’s a lot of intent and many nested decisions that were built into one, and the compressed intent is what we find interesting, compelling, and something that demands examination so that we might learn and grow.

I saw this one once in a museum.

No joke here, the audio commentary really helped me properly appreciate it.

My tattoo artist tells me I'm dead wrong, though, and that this kind of thing is just a quick scam. And that's fine - we're always free to disagree on how many decisions went into the creation of something.

Simplicity without a dense underlying decision tree is just empty data; simplicity born from extreme compression is a masterpiece [38].

▶Subconscious Intent

The "Death of the Author" is answered as well. The author’s intention clearly does matter, but we evaluate the intention of all parts of the artist - including their subconscious, the intent of which they may not have understood at all [39]. So, the artist’s intention absolutely matters, but the stated intention is only part of the picture.

Any time you ask whether a functional thing could be art, like a chair, it’s goal layering that answers the question. You, as a chair designer, clearly tried to make this chair serve some kind of function [40]. And I, as a maker of chairs, can appreciate the artistry - in a Ron Swanson kind of sense - of this specific chair.

Can Nick Offerman even make chairs?

The question of whether the chair is art has to do with whether the creator of the chair just wanted to meet simple base requirements by making minor modifications to an existing design, or whether they had complex goals of their own they were optimizing towards by layering interlocking decisions. As always the path to full understanding is to understand intentionality. And when you go into high art, the answer could be something like "it’s a commentary on other groups of art or on artistic cultural artifacts." It can get rather meta as you get into layering decisions upon decisions, as it should.

We can easily answer questions about randomness-derived art such as Jackson Pollock or glitch music. Let’s take a moment to talk about Jackson Pollock and the amount of choice that went into the creation of those paintings. The setup was not easy, neither were the choices likely simple. Jackson Pollock, who used a mix of understanding the cultural landscape combined with a set of skills arranging paints and ropes to create desired effects - that array of decisions did result in art. You can try to extract why each decision was made from the paint splatterings. It is the difficulty of this extraction that is inherent to the art of Jackson Pollock, because it’s often difficult to differentiate between the randomness of the painting's process and the subconscious of its artist [41]. In doing so, it plays with the concept of what art is on a fundamental level, but in a way that is quite obvious to us using this model.

Pollock was (in a way) an early AI artist.

This may be the first will-smith-spaghetti attempt in history.

We can also casually categorize things like shock art because as not artful. Toilets aside, they can often be highly aesthetic pieces - honeypots intended to grab your attention with garish, loud, or excessive displays. But they can’t really hold your attention easily because part of the magic of holding attention involves presenting this kind of puzzle to the viewer: "What series of actions could have resulted in this and what were the goals of its creator?" These puzzles are compelling, and shock art removes that follow-up. It is literally just raw aesthetic attentional capture without the follow-up in observable decision-making required to maintain that attention [42].

▶What experts like

Why do experts like such weird stuff?

Let’s talk now about the other axis of our graph, good art and bad art, and how this ties into the concept of respect. Effectively, respect for the artist refers to whether you agree with the goal of the artist. Full stop. You won't respect a failing director's competence, but you may still align with their goals. Respect for competence is a different component, and refers explicitly to the y-axis of the original graph. Recall: if I respect your competence in achieving your goal (through agreeing that you did indeed accomplish it), then that was "good art" to me.

You aren't allowed to write about art without mentioning this movie.

If I respect the artist - say I hypothetically really admired Tommy Wiseau and I had a lot of respect for him, but think that the movie The Room was a disaster - in that case, I might treat it as a "noble failure." Frankly, as someone who knows nothing about directing, I actually do have that respect for Mr. Wiseau because I’d imagine it was a lot of work, especially for an amateur. And when I try to imagine how much effort went into this - by which I mean all the decisions - then that means I actually have very little difficulty appreciating Wiseau. I would imagine that it is even more compelling for directors to see such a thing and to understand more of the decisions and why they were made.

It is the language of expertise that unlocks the ability to observe the hierarchies of decisions in others. Expertise just means having made lots of similar decisions yourself – building out your own hierarchy of action through automaticity [43]. I believe, famously, some of Kurt Cobain’s favorite music was unknown "outsider art." You can find a video of Teller’s (of Pen & Teller) favorite magic trick – and it is confusingly bland and short, requiring a lot of explanation from him to understand why he’s so thrilled. These are creations from truly inexpert creators who nonetheless made interesting decisions that only a true expert - the master becoming the student - could appreciate. As they observe the decision-making of the students, they can judge each decision and choose to incorporate that into their own larger hierarchical structure of nested decisions inside of automaticity.

Kurt Cobain, Teller, and other masters of their craft often have... unusual favorites.

I would also argue that we can, if we want, also respect authority. Which is to say that other people tell me that what an authority says, or the actions they describe, align with my own base goals. You have to believe in the authority-giver in the first place, which can then endorse this figure which can endorse the next in a kind of chain [44]. I myself prefer to only respect competence and kindness. But I would be willing to bet that has to do with perhaps my own autistic sensibilities. Still, we all must do this to at least an extent – you cannot easily judge the skill of those whose nested automaticity functions contain pieces you aren’t familiar with. Judging an expert as a novice will leave you frequently unable to even notice the decisions the expert made, let alone intuit the goal of those actions.

Expertise is simply the cognitive resolution required to see the load-bearing scaffolding beneath the paint [45].

▶The Flowing Iceberg

But there’s something else “artists” can do, too, that goes one step further. You see, modern psychology tells us that we do not always understand our own motivations.

Your subconscious - the parts of yourself that make decisions and can influence your behavior but that you do not have direct access to - is part of this mix as well [46][47]. Because sometimes when you finish making something - perhaps you’re an engineer making a bridge - you sit back and admire the bridge afterwards, appreciating your own art for a moment, if only to try to understand why you made certain decisions. Because in doing so, you learn more about yourself and your preferences and why you might have come to certain conclusions [48]. And indeed, I would argue that classical artists do have a certain mental trick they play on themselves where they mentally relax their intention and then allow this part of themselves, this subconscious, to make more decisions as they themselves make fewer. Cognitive scientists call it flow state [49]. Maybe there’s a bunch of initial conscious planning, and then you do something like prompt an internal LLM perhaps, and then just start to curate what it generates in a way [50]. I do suspect that we will find some processes like fast-thinking speech generation end up working in a very analogous way to AI structures such as LLMs.

People love pictures of the iceberg-subconscious.

Because it's... bigger? I think that's genuinely it.

It’s arrogant even to claim to “fully appreciate” art because that implies you understand why every single decision was made in the creation of that piece. But even the creator doesn't know why they made every decision. You can only really understand the decisions that you climb above, that you yourself can perceive the goal of, and understand all the constituent automaticity functions of, and finally judge how effectively the steps taken led towards that goal or not.

One can imagine the case where a sculptor claims to create a perfect likeness of a king, and in so doing creates a beautiful piece of art that people appreciate, but you couldn’t say you fully appreciated it because maybe the artist subtly disliked the king, and that caused them to create a slightly more sour visage using one subtle method or another.

Christopher Anderson shows why you don't piss off artists.

They can do this shit on accident, even.

And so even if the artist tells you the explicit goal and how they accomplished it, you can still try to appreciate the art even further by disagreeing with the artist and implying that they do not know why they took the actions that they took [51]. This is probably the main method through which the master learns from the student.

Talking about Neuroscience

Let’s start with the learning aspect, and why it’s clearly central to your functionality. A little modern cognitive science for you: stimulus-response is a bit out of date these days. It’s better to imagine yourself as a sort of prediction machine. First, you learn to predict your environment. Then, you use your predictions to error-correct your way into a future that looks the way you want it to look [52]. If you want to Wikipedia-dive, the terms you’re looking for are the Free Energy Principle - or when AI agents use the same mechanism, Active Inference modeling [53].

Effectively, this perspective states that you constantly have two goals: to become more certain about your environment, and to use that certainty to guide your environment into whatever you want it to be. Learn things you don’t already know, then use them mercilessly to maximize your goals (such as they are). And while we are constantly doing both, we’re only going to be engaging here with the learning aspect.

The Thousand Brains of the Galactic Senate

Let's tie this all together really quickly with a metaphor to explain the Thousand Brains theory of consciousness in simple terms (while baking in a few other models for your benefit) [54]. Imagine the neurons in your brain as something like a much, much larger and more diverse version of the galactic senate from Star Wars. Each little hovering repulsorpod with an alien in it is a neuron. Also there are a lot more of these senators-on-repulsorpods in your brain - trillions.

Jar-jar is, in this example, a small part of a single neuron. I love robust metaphors.

Some neuron-senators are at the bottom, and can physically see the "ground" truth: raw sensory data. Then all of them yell and argue about what they think they see. Above them is another layer that looks down and can't see the ground truth - but can hear the arguments [55]. There's a fog. At some point, someone in that second layer who can hear all of this will yell "we're touching a curved, smooth object! A lot of you are saying that!" and everyone below who isn't yelling that shuts up. And now the second layer starts arguing until someone in the layer above them hears the noise (and maybe the people in that layer can hear a little of the argument on floor one) and yells "we're holding a cup!" This continues up the floors of the galactic senate until you get to the top floor, where - the supreme chancellor is missing. All we have are 150,000 or so top-level senators voting on everything. Maybe in this case they're voting on "is this cup of coffee mixing well with the soy sauce I poured in?" or something.

Now, the higher levels of senate aliens care a lot about when the lower levels are wrong. Note that the senators doing higher-order reasoning aren't generally using raw sensory data. They're using the perspective discussed below to inform their reasoning (this is how you don't actually "see" reality, but rather your own predictions of it). They're keeping track of which senators below are often right or wrong, and updating their own trust and voting ledgers as they do so. Each senator has a ledger: it helps them keep track of how to vote given what's below.

I want everyone to note how cleanly groups of people seem to act like neurons at times.

I feel like there is a general field of study here about... intelligence... and it's interesting.

Two things I want to get out of this metaphor. First, when a lot of senators are yelling at the same time, it's costly. You only have 20 watts to run your brain with, and you like it when you can have senators positioned above that can yell "quiet" often because they correctly figure out what the deal is early [56]. You learn how fire works, and you don't need to spend time re-understanding smoke when you have a senator that knows how to identify it quickly. Even better if the senators above see that he's right frequently, because the second thing I want to introduce is how surprise fits in here.

Once the senator above yells "quiet" to all the incorrect shouters below and declares they've figured out what's going on, everyone down below who wasn't correct has to not only update their voting ledger so they don't mess up again in the future quite so badly - but they also have to tell all the neurons below them to update their weights too given this new information. This combined work is costly, so much that you can actually feel it. It feels like being surprised. The Free Energy from the free energy principle that we try to minimize as the learning half of the active inference model is simply the effort that all of these senators have to spend updating their voting ledgers. The more wrong they were, the more they have to change, and we try to minimize that overall effort [57].

Now that we have this model of galactic neuron-senators (my own metaphor for the thousand brains theory of consciousness), let's attach it to what we've been talking about.

Mirroring Intent

Mirror neurons have long been associated with the concept of empathy (affective empathy, specifically) [58]. Fun note: mirror neurons are a little out of vogue right now, in part because we mimic things more comprehensively than their function would imply. Mirror neurons are lower in the galactic senate, effectively acting as our eyes into the emotional world. We use them for what we call affective empathy, sure, but if anything their limitations show that we clearly do more than just that. Enter embodied simulation. Embodied simulation is a more active process, using cognitive empathy instead of affective (lower-level and emotion-driven) empathy. Take a look at the following photo.

Can you feel it?

Even without a specific reason for your mirror neurons to activate, I bet you can feel it: you know how you would feel holding that ball, how it would feel to throw it, and what your muscles would do to accomplish that exact goal. It’s not a muscle flex, so much as a reflexive sort of awareness. You aren’t empathizing with anything: there’s nothing here to empathize with. Your lower level neuron-senators are quietly refusing to mirror anything, but the higher-level senators can still use their previously filled voting ledgers to figure out the details of how this could be executed and yell upwards anyway. It happens almost without you noticing: it’s a reflexive engagement with the world. It is embodied simulation, driven by the Theory of Mind network in your frontal cortex [59].

Let’s talk about that Theory of Mind network because it is vital. Specifically, I'm talking about the neuron-senators in the middle of this particular chain: the ones that read from your lower-level emotion-aware mirror neurons.

This network is what raises us "above the animals," so to speak. It is the robust structure that is one of the hallmarks of the neo-mammalian brain, something nearly uniquely human given how specialized we are in it. A lot of animals have mirror neurons and limbic systems, and some even have some capacity for cognitive empathy (great apes, dolphins, whales, elephants, crows, and ravens have more than normal).

I wonder if there's a moral culpability that comes with having this brain structure? I can forgive a spider - but the dolphins know what they did.

But no one went quite as hard into specialization as we did, and the robust structures in our brain that hyper-specialize in this sort of higher-order empathy are quite uniquely human. Effectively, we have a beautiful superpower: we can model other brains with incredible accuracy. We can use cognitive and affective empathy, using each to error-correct for the other [60]. We can use Theory of Mind to try to understand what other people are thinking and how their perspective works.

The neuron-senators on the ground floor are the mirror neurons, the source of affective empathy. The ones above are your theory of mind network. Just like your senses help you error-correct your simulation of the world, your mirror neurons and affective empathy help you error-correct your automatic simulation of other people's physical intent. [61]

A quick aside: I'm glossing some of science here. For example, your theory of mind network and the mirror neurons in your limbic system are part of distinct and separate networks, but often work together for certain tasks. So it's more like the neuron-senators from those floors are often jointly members of special committees on human behavior.

There’s one very specific behavior I want to point out. Obviously, art is very tightly coupled with the Theory of Mind network. When people view an image and are told that image is “art,” those regions of their brain light up. The Theory of Mind networks activate.

The Theory of Mind Network is the seat of Cognitive Empathy, and error-corrects using Mirror Neurons

Something interesting happens if you tell people that the image was computer-generated or is random: almost immediately, those regions go fully dark. I think you can actually subjectively feel this; we’re all familiar with the sensation by now. When you’re viewing a picture online and halfway through realize it was generated by AI, part of your attention slams off as one of the larger parts of your brain… just stops caring [62]. What we are subjectively feeling is this Theory of Mind network turning off [63][64].

The exact mechanism for this works through the Default Mode Network, another brain network that has something it cares about. It is the network that decides who and what currently has control over your mental processes. When you simulate someone else, often you use your own brain hardware to do so: your Default Mode Network keeps that straight by ensuring the rest of your brain knows that "we aren't panicked, we're imagining what that person's panic must feel like."

The Default Mode Network is like a pretend "simulation protocol" that the senators can run. They disconnect from everyone else, and just... daydream, or pretend, or simulate. The Salience Network acts as an arbiter telling the neuron-senators to just play pretend to run a simulation for a moment. The arbiter would be involved in deciding whether incoming simulated data is valuable enough (in terms of intentionality density and alignment with your values) to let the neuron-senators relax the rules for a moment and write down some of the simulated data in their voting ledgers. This is how learning occurs.

With AI-generated data, the arbiter never sees anything valuable enough to relax the rules - there's no intentionality in the data to learn from. We can't use our theory of mind network to judge the intent of the creator, so we can't judge the creator's goals or their implied values. The theory of mind network doesn't activate, and the default mode network doesn't relax to allow the simulating neurons to update their weights using the simulated data.

There is a part of your brain continuously seeking evidence of actions, assessing those actions as effective or ineffective, and trying to learn the effective ones by intensely empathizing with the creators [65]. You come across AI art and that part of your brain activates immediately, trying to learn and understand this person who made decisions for some reason. Then, when you realize AI made it, it’s not that there isn’t a creator - but rather that the math becomes so complicated that you simply stop trying to figure out what set of matrix algebra combined a prompt and random noise to create this particular piece. It is not understandable to you at all. The black box is so unbreakable you may as well not even try.

Left: AI Art. A generated photo intended to capture a blended version of Cy Twombly’s and Jackson Pollock's style.

Right: A famous painting from a master at the top of their craft. Experts, in particular, seem to admire it.

Please take a moment to look at them both.

Are your eyes drifting right? Are they almost… sliding off… the picture on the left? Now you know why.

I would liken it to the feeling of being in a magician’s audience. The magician has promised to provide you with an interesting, aesthetic performance that is actually not understandable. It is a puzzle wherein they invite you to learn how these things could have been accomplished - but of course, the point is for you not to figure out the answer. That is what it feels like to be in a magician’s audience, which is why you often don’t even try; you want to be fooled. Either you enjoy the spectacle and aesthetic appeal and the feeling of surprise, or you try to puzzle out how they did it. Both are valid ways of enjoying a magician’s spectacle, but only one tries to properly appreciate the work the magician put into the performance. Even past the potential for appreciation, AI art is even less interesting because while the magician invites you to figure out the puzzle - with AI, all of your brain's normal architecture for appreciation is useless. Your brain will not allow you to do that kind of matrix math fast enough (…yet).

This is also a process for distant learning. It is one of the main processes by which we engage with society, I would argue - this kind of distant, empathetic learning. The current dominant model of learning explains sitting in a classroom as follows: you hear a teacher give a speech, and you rearrange the relationships between the neurons in your brain (you adjust the weights!) such that you could produce the same speech [66]. Those of you familiar with how LLMs can map their internal weights to clone each other’s outputs as part of a distillation attack will find this a very familiar-looking process. And it is. And with the power of your Theory of Mind Network, you don't need to even watch the creator in person. As you’re looking at a sculpture, if you can figure out how it was made, you can now make one yourself [67]. It is a method for survival, learning, and connection over a distance. It is the way we are constantly refining how we interact with the world as thinking beings. We seek evidence of intentionality so that we can learn from it.

We learn by reverse-engineering the decisions that shaped our world [68].

Talking about Communication

I also want to take a brief moment to talk about mimicry. I used to tell a joke when I was teaching classes on this. I would say that most languages - almost every language, I believe - have some pun, joke, or word that connects primates with the idea of mimicry. In English, it is "to ape" something, as in copying it. But interestingly, when we study apes in the lab, we find that they are pale shadows compared to the masters - the true masters of copying any and everything they see: humans [69].

Laughing Together in One Way

If you walk into the middle of a room with a box and an ape, do a "box dance," and then grab a banana out of the box, the ape will simply walk forward and grab the second banana from the box. But if you do that same thing for a human child: the child will walk up, and their Theory of Mind network will activate. They look at your decisions, and then try to figure out why you did the box dance, fail, and because of that failure: they just have to do it too. Because if that child respects you by believing that your motivations are in line with their own, and they also respect your competence by believing that your methodology is likely superior to their own or otherwise aligned with their own interests… well, then they'd better do the box dance too [70]. So they reflexively watch you and learn how to do that exact box dance so that they can do it themselves.

The Firewall of Disgust

By now you see that to learn you first simulate someone else, then relax the "rules" of simulating to allow the senator-neurons to write down a few of the simulated voting rules. But what if you’re simulating someone awful? You do have protection - though it’s not perfect.

We aren't disgusted by art we deem unacceptable merely because the motivations it reveals are ugly. We're disgusted to ensure our brain rejects the neural weights (the numbers in the neuron-senators' voting ledgers) because the target's simulated data was corrupt.

I would argue that at least a little of the disgust people feel at AI art comes from this effect. I have no idea how to prove that one, though.

I just want to highlight how much we are, in a very real way, the copy ninjas of the animal kingdom through the magic of our Theory of Mind network. We don’t even have to look at you to copy you. We can just see the results of what you’ve done and then do it ourselves. I can see a video of a knife fly through the air and, with enough appreciation over time, learn to do the same thing myself just through careful knife-study.

A Squad of Copy-Monkeys in their Natural Habitat

We’re constantly seeking evidence of intentionality, evidence of activity from other thinking beings in our world [73].

And it is through this constant state of seeking to learn - and learning through looking at any individual object - that we are reforming ourself into the kind of person that could make that object.

We spent so long trying to figure out what our ultimate weapon was: endurance sprinting, throwing things, pattern-matching… when it very well may have been the semi-telepathic nature of empathy, empowered by viral meme-learning all along [74].

We are Earth’s apex predator, wielding our truly mightiest weapons of all [75]:

Memetic Predator has a nice ring to it, doesn't it?

(with the power of empathy!)

Talking about AI

By now you should have your own ideas on AI, and you may even be ahead of me on this one. Basically, depending on whether you can intuitively understand how matrix algebra results in a picture, AI art is either anti-art or hyper-art. I’ll explain.

Let’s acknowledge that it is technically possible to create such an array of AI art that the sum total of all of the decisions would approach the decision density of more traditional artists. You and I both know that no matter how much you hate AI art, if you met someone who said they had spent four years crafting a single piece of AI art - a picture generated entirely by code, but with the work of four years of continuous effort going into it - you would be a little bit interested and compelled to see what such a thing looks like, AI art or not.

But effectively, when you show me standard AI art, I process the number of decisions that went into its creation, and I acknowledge it to be minimally artful by definition because of the low number of decisions made. You’re left with the feeling that you could do that very easily. It is not impressive art at all. It may be "good" (achieving the creator’s goal) or "bad," but few prompts would be considered artful by most standards - and other than model selection, how many other decisions are being made?

It is actually worse than that. Because at this point you should see that art is a kind of distant communication - a way to empathize with people who aren’t present.

Artists are masters of the telephone game.

AI doesn't know how to play yet.

Part of that communication relies on the fact that everyone has their own idiosyncratic mechanism for disrupting data, for changing the environment. You, as a particular artist, may hold the brush with your left hand upside down, and that gives you a particular style. That decision-making tree is your particular set; it is well-explored and becomes your artistic style. This kind of idiosyncratic order is a level of personalization that comes in part from you and in part from your subconscious.

AI art is a direct removal of this.

If AI art is a kind of average of everyone’s perceptions and perspectives - which it is - then your own idiosyncratic order and the decisions that you specifically make are removed [76]. They’re simply no longer present. And what’s more, we aren’t just filling in the gaps with what makes sense. We’re filling in the gaps with the average of humanity. We are regressing towards the mean, away from the decisions you are making. You said you want to make a fish, and you have an idea of what kind of fish you want to make. But the AI generator will move you away from your own mental idea and towards what everyone else thinks a fish looks like. It won’t even give you a random fish (cough assuming a temperature of 0 cough). It will give you the average fish. A common, uninteresting fish.

One fish, two fish, your fish, slop fish.

The latent space is a graveyard of idiosyncrasies, ground down into a frictionless paste of human expectation [77].

By definition, this removes the connection. It makes it harder to identify who you are. It makes it harder to empathize, not just because you made very few decisions, but because you used a tool that fills in the gaps not with your idiosyncrasies, but with the decisions of everyone else. It makes the least interesting thing in the space.

It is, in a very real way, anti-art. If art is your decisions, then AI is everyone else’s decisions in conflict with your own. It’s worse than Jackson Pollock. Jackson Pollock used noise to fill in the bricks of his decisions with the mortar of randomness. But if we do this with AI art, we don’t just get random noise; we get boring noise. We get the kind of noise that everyone makes. It is utterly uninteresting by definition [78].

It is a "sanding down" of the most interesting parts of any creation—the parts that allow you to see a little bit of the artist, the eccentricities and idiosyncrasies, and to see the originality. AI art sands all of that away and returns us to a world of the world’s average [79].

Since we are learning, seeking beings actively engaging with our environment through the sense of intentionality, trying to learn all we can about the world around us, it is novel or creative chains of actions that we find most compelling. Therefore, AI art - the opposite of something that is novel - is the opposite of compelling. It is a regression towards the mean. It is anti-art.

Unless… you’re some kind of future-version of a human with a chip in your head and the aforementioned ability to intuitively understand how pictures can result from a prompt after some matrix algebra [80].

You'd best start believin' in cyberpunk dystopias, you're in one.

Then, you might admire this as some kind of proto-version of whatever kind of art you might be familiar with. You may see it like some kind of an ant farm – primitive, but fascinating nonetheless. Perhaps, in tracing how a particular piece of AI art is made, you might find whispers of intentionality or decision-making, and that would also be art. Some kind of… early… hyper-art? I really hope the future is as awesome as all our new words suggest it may be.

Alignment

I personally believe that the alignment problem is the most important issue facing humanity at this time, and has been for my entire life. Successfully aligning an emerging superintelligence’s will with our own is the difference between a Star Trek heaven and a sudden and unexplained biological strike resulting in humanity’s immediate extinction [81].

There are numerous papers on this idea, and I won’t add to them here except to say that if one of the core powers of humanity is the extraction of intent from the observation of behavior – it provides a clear path forward to solving the alignment problem by creating frameworks that build off our empathic brain structures to intuit intention. One of the core troubles of the alignment problem has always been that we don’t understand what we want [82]. This sort of… juiced-up inverse reinforcement learning… would absolutely allow us to make something better than we are at cracking the very problem we can’t solve ourselves.

Currently, we use reinforcement learning from human feedback: we have machines train on human ratings of generated text [83]. But these ratings are inherently going to be variable because they will be either built off errors from cognitively exhausted raters seeking nonexistent ghosts in text - or on shallow aesthetics. There's nothing else present in the rating data to capture!

All else being equal, this means the expected output is shallow, error-prone shock art. Slop [84].

Coming soon, from a facebook near you.

But we don't value text with surface-level attention-grabbing openers like "The Solution? Synergy." We value text with depth: with multiple layers of hierarchical decisions within.

If we want to move forward with aligning our AI, we can't use this noisy surface sampling. We need to use cooperative inverse reinforcement learning [85].

Arguably, one could say that one of the greatest powers of the human mind is its ability to align itself to the unseen goals of others. Let's steal that architecture, enhance it as far as it will go, and use it to solve the alignment problem for us.

Alignment won’t be solved by coding better guardrails; it will be solved by building a machine capable of reverse-engineering the goals of a collective soul we ourselves can’t yet understand.

And wasn’t that the way it was always going to happen? We always needed to figure out how to build a framework that creates its own goals in alignment with our presumed perspective, but derives them intelligently from us somehow.

A clear description of the problem

I don’t know if you’ve noticed, but the world has become a LOT more confusing lately. We are all having trouble separating fact from fiction, understanding our rapidly-developing world, and being understood by those inside and outside of our communities.

The signal-to-noise ratio of human connection is collapsing [86].

This is a cultural, ideological, and biological problem.

Memetic Stagnation

The cultural argument is easy. I think it’s fair to say that there is more to culture than art, but I won’t bother proving that art is critical to culture. Do you like good movies? Words worth reading? Those directors and authors needed clear evidence of intentionality in front of them to study, for years, in order to become what they are. Without the ability to find and grow from seeing intentionality, they can’t impress you with what they learned.

And without an audience that can see the work they’ve done, the motivation to make things diminishes. There is an innate want to share the things we make - that’s part of the way the memes spread so virally [87]. I wrote this paper to be read. If I knew no one would ever read it… the motivation to write it wouldn't really be there. There will be less art of all kinds in the world, and less quality movies, books, and works.

And it’s not just traditional art. Remember… a bridge is art, to an engineer.

I care quite deeply about the continual successful transmission of bridge-building skill.

I feel this is something we can all agree on.

We aren’t just training authors and directors: intentionality trains doctors, scientists, and engineers, too. Their “art” will degrade as well. Bridge-building engineers learn in the same way as other artists, and strive to be appreciated for the quality of their decisions too. With fewer people looking for those decisions at all because they “assume AI did most of the work,” we can expect the motivation or even capacity for learning to start to decrease as well.

We are also headed towards ideological catastrophe. Without evidence of intentionality, we will not be able to develop our cultural memes properly. My favorite Martin Luther King Jr. quote has always been “the arc of history is long, but it trends towards justice [88].” We are, all of us, very slowly making a better world. We do this with speeches, with ideas, but also with diagrams, plans, and tools.

Remember when I called us memetic hunters? That was true – but we’re also clearly sort of memetic farmers as well. We keep making our environment better, building good ideas on top of other good ideas [89][90]. Political memes have moved from tribes to slaves to kings to democracy. The best war memes of our forefathers were pointy, but ours have lasers.

We're the only species with a tech tree. This is why.

We cultivate better and better memes over time, and they are our real power.

We really, really shouldn’t ever stop doing that.

20 watts of caring

The cognitive catastrophe is the most concerning to me, though. Uncertainty - lots of it - is something that is incredibly uncomfortable for our brains. Recall surprise is precisely the thing you are optimizing to reduce. Remember the Free Energy Principle? I’ll bring it back in the appendix. It is a unified theory under which your brain acts to minimize uncertainty as its primary function for organization and learning. Uncertainty is expensive not just because you’re more likely to be wrong, but also because your brain spends a lot more energy rewiring [91]. Being systemically more uncertain about everything means you are, by default, more cognitively taxed at all times than you should be.

You are stressed. And you’re stressed from overthinking, specifically [92].

So you’re stressed from the constant surprise and thinking required by the world, and that means you’re paying a metabolic tax [93]. Your brain only has about 20 watts of power. All that thinking has a real cost. So your brain de-prioritizes the most complex actions available, and you spend less time and effort using its theory of mind network.

In the front of your brain is the most potent weapon this planet has ever seen, and you’ll have a harder and harder time using it.

Apathy is the final evolutionary defense against a high-noise, low-signal environment [94].

Abiding.

Your brain wasted so much energy in the past seeking the ghost of an author in the world around it. Nowadays, it keeps being disappointed after its search for meaning. That disappointment becomes a habit of early disconnection, all driven by cognitive fatigue [95].

You will literally become too tired to care.

As a result, you lose empathy, compassion, and learning. You even lose your memes.

You lose access to everything that truly makes you human.

…And a Solution to it all

I saw a problem, and all I had was a hammer - I used Human Factors Psychology to solve this. I saw a solution, and built out the framework myself.

It's surprisingly simple: we need to fix the interface. Right now, our brain’s main weapon (our theory of mind network) fires constantly. We seek - and find - intention in almost everything, whether it’s there or not.

We need to bake in another layer of information when we communicate that solves this problem. And we need to make it easy to understand, reflexively discerned, robust against future technological advancement, simple to apply, and something that unfamiliar people can pick up through affordances. We also want to somehow address bad actors structurally, if possible.

Conveniently, this is literally my job (…what I wouldn’t give for such a clear problem description).

Alright, time for the big reveal!

Meta is the new black.

I’ve actually been doing it throughout this whole paper!

We need to actually sign our work with intentionality.

We have to stop treating authorship as separate metadata and start treating it as a primary visual affordance [96].

We need to illustrate, throughout the document, which parts of the text were made by the user and which parts were the result of interaction with the AI. That way, we don’t devalue all text and lose the ability to use it for connection, understanding, and learning.

Alright, so how do we do this? Honestly, it’s very simple.

The Ghost Scale

First, let’s establish that the background color of any document, image, or creation is the “AI’s color.”

The text? That’s my color.

So any text at 100% opacity has never touched an AI –human edits are fine. Since this’ll be pretty rare (we all love AI-assisted spellcheck), it allows us to play a steganographic trick. We’ve now made 95% opacity text the most common type ("minor edits or translation made by AI"), and then we allow even authors who prefer to stay at 100% to indicate opt-in through making a single sentence 95% opacity. There’s even some artfulness to how one might do that through picking a particular sentence.

Effectively, this means that this model is backwards compatible with all previous human text. Shakespeare didn’t use AI, and so his text is 100% black when on a white background. Or, if the background is some other color, his text is fully opaque – which is how we’ve always done it.

Conveniently, semi-transparent text is something we absolutely avoid in UX design, for obvious reasons. Text exists to be read, so we dial up the readability to 100. In doing so, we also dial up how attention-grabbing it is to the max. Amusingly, everyone does this constantly. The only reason not to is when making art (when you have some other motivation). So there’s a huge hole in the middle of everyone’s design space. Let’s use it.

Let’s dial the opacity of artificial text down just a hair, and let your brain reflexively de-value the text. Let it relax until it’s needed.

Semi-transparent text needs to fit web standards, which means we’re going with WCAG constraints. They use contrast ratios as their baseline, but it works out to an equivalent of around 54% opacity across all colors [97]. Let’s make that an even 60%, so we’re a little bit more readable than the “legal limit for readability.” Now we have our “curator/synthesizer grey” (we’ll sometimes call it “grey text” conversationally rather than the more technically precise “60% opacity text”). Let’s define it: “a human and an AI were both directly involved in the writing of this specific text. The human either wrote alongside the ai, curated this output, or generated the core ideas for this segment of the document using the aid of AI.”

Everything outside of direct synthesis or ideas that you did not explicitly create is grey. I honestly think there's just no way to subdivide it any further and remain honest.

I'll demonstrate with another meta-example. Remember when I brough up the Free Energy Principle earlier? One of my editors (a friend) called me on it, and said that because they personally knew for sure that I had learned that from AI: it needed to be grey text. But I had learned it about a year ago, totally unrelated, as part of a general tendency to say things like "look into my history and tell me things about my field that I should know but potentially do not, then explain them simply and cite them." Then after, a few months later, I used that concept at work. I spent time in rooms talking about it with other engineers. I read a book on the topic.

So does the text need to be grey?

I told you it gets murky in practice, so the definition needs to stick to a hard line: the answer is no. My text can remain 95% black. The Principle originates from a human, no issues there. The combination of these models into a global model centered upon an explantion for art, along with how the Free Energy Principle fits into that model, is entirely mine. I certainly spent time having many models try to break my ideas, and sometimes changed the phrasing to pre-emptively address arguments I hadn't thought of. But the core idea I'm presenting is not something I curated. Do note I'm somewhat near the line. If I had said: "find other models for my unifying theory and I'll choose which to add," then I am now curating external ideas and the text needs to turn grey. Bulk grey text is fine, and will be addressed just before the appendix.

Finally, you don’t have to be so totally suspicious of all text. If you use a mod that tells you when text is “claiming” to be human with 95% opacity, your brain can relax just a hair on that text. Sure, maybe they’re lying, but your brain always has to keep an eye out for that anyway. At least now your brain won’t have to engage with every single paragraph and picture like it’s some kind of puzzle-challenge, working out whether there is intentionality or whether it’s just “AI slop again” [98].

And this leaves a perfect bit of design space for AI text, too. It’s easy to imagine a world full of bots on bots on bots leaving messages on webpages using tiny white-on-white text. This allows for that thematically, if you like - ghost text. But it also means we have another way of using white text. Here, let’s formalize this one too:

I like the idea that text from AI should be inherently attention-dismissing, and so to justify its inclusion it has to be precisely curated for incredibly dense, high-value statements that I ideally couldn’t have produced on my own.

If I want you to read text that is only half mine, it had better be twice as useful and half as long.

As much as possible, I’ve tried to demonstrate a variety of ways throughout this paper that this model may be used. The first ghost text block used it for meta-commentary to introduce the concept, the second showed a block of highly relevant, high-density information that I thought was a particularly well-phrased way of communicating a concept. The third is a summary of a long bit of workshopping.

You can also find bits of grey text scattered throughout. I used the grey text in a few ways. Sometimes I inserted a bit as information-dense punchy statements used to control paragraph flow. One time I used it to highlight when I used information that I want to be a bit distant from because I want the ability to walk it back if I’m wrong about something a bit too far outside of my field (see the sentence about contrast ratios on the previous page). Grey text is clearly most useful, however, for technical writing: writing in which understanding intent (while always important) is not the primary goal of reading. And so the appendix demonstrates the third obvious use of grey text: a very carefully synthesized information-dense follow-up. I am openly owning that I made it in part using AI, which means it needs to be even more interesting to capture your attention.

Creators need a way of using AI “correctly,” so that the true value of their creations won’t be lost [99].

Particularly clever authors may even like the meta-layer of attentional control this gives them over their audience, and selectively use AI to help control user attention across the less interesting parts of the text. Even if an author wanted to stylistically use 100% black text to indicate that the text is fully unedited, that author could quite easily put the first sentence at 95% to indicate that they are using this structure. There’s room for nuance and exploration within this framework, as well.

Picturing Ghosts

We can apply this to visual art, as well. A border, shadow, or some other kind of coloration around the edges of the picture can be used to indicate its source. This part makes me the most uncomfortable, because I’m asking the most of artists here - but the average artist hates AI enough to want to do this anyway. That border can look like anything. It can be a single pixel line, a shadow that technically has some presence around all four edges, or even a stylized dragon border.

Let’s be a little bit more formal with this one, and allow artists and curators to both opt-in. Artists, if you want to declare the AI usage amount for your art, then you can either make a full border using a black (human) to white (AI) scaling, as with text (we have to standardize around a hypothetical “white” background color in this case) - or you can just color all four corner pixels for a machine-readable version. And curators - people like me who are just pulling stuff from google? We can put a second border if we choose, one that lines up with our best, honest assessment of this image’s origin. Citations can optionally be included in place of a prompt for black borders, and having no border is assumed to be black.

And, to future-proof this concept, let’s allow for authors to write their own citation snippets within the border. I want to allow space for future “hyper-art”. If two artists spend 100 hours each on their pictures, but one uses AI – that doesn’t necessarily mean that the AI-user made fewer decisions. It is unfair to then suggest that the AI artist stick an attention-dismissing border around their art. The solution is to allow for anyone to write in “citation information.” I would encourage future AI artists to write down precisely how long it took them to create the picture, which is an easily-measurable approximation of decision density. The grey border is then a tacit acknowledgment of the pull of AI towards the mean, and perhaps respect from the viewer for the work it took to break free of that.

This does limit artists slightly, as it constrains them near the edges of pictures they want to use this system with. Furthermore, they can choose not to engage with this at all, and allow future curators to judge the source of the art and then apply their own borders (with respect to the backgrounds of their own creations, like this essay). I want to leave art as free as possible while still bringing artists under the umbrella of this standard. Just like authors, they want you to appreciate the work - not just look at it.

So how about an example, eh? Hell, let’s try to make a meme about it. “How deeply should people drill into your text?”:

View Simple UX Framework (Figma)

In my professional opinion, using this framework consistently would also have two additional benefits that reinforce its use: an automatic attentional wiring and an increased ability to discern transparency over time. It wouldn’t take but a year for you to start to seek intention and learn from darker text and art with darker borders (sorry again for that limitation, artists). In addition, you would almost certainly get better at identifying shades of transparency. Most children would gain the ability to visually spot the difference between 100 and 95% opacity. It would slightly reshape our brains, a sort of grammatical application of Sapir-Whorf [100]. It would slot right into culture rather neatly.

The Best Security: Game Theory

And as for bad actors? That problem actually solves itself, believe it or not. One of my undergrads was in economics, and I focused as much as possible on game theory (mostly for fun, at the time). The term for what we're going to take advantage of is Zahavian signaling. Creators take on a particular cost by honestly marking some of their output as AI-assisted, but they also gain a more nuanced relationship with their audience along with a sort of permission to use AI when appropriate [101]. Authors can always opt out of this deal by returning to flat 100% text and accepting the baseline level of suspicion on all of the words they write.

Bad actors, by definition, want to appear human. Currently, the ones that are trying to build a reputation are doing so in part by taking advantage of a squishy rule in terms of AI usage: how much is too much? As a bad actor I could always claim with full black text that I used AI for some light editing and didn't really notice that it had rewritten the whole sentence. But I pay a much higher reputational cost if I do this with 100% black text in a document while explicitly using the ghost scale. A larger portion of my audience will correctly have their theory of mind network converge on the idea that there's just no other explanation for my words other than that I have explicitly chosen to lie. In a way, the use of this is self-policing.

So that's bad actors taken care of.

Ghost Alignment

It’s good for authors, it’s good for the brains of readers - it’s even good for AI labs. And speaking directly to the AI labs - please note that if this practice spreads, it would be one of the more effective measures we could implement for preventing model collapse over the long-term [102]. I suspect our current tricks won’t work forever, but maybe I’m wrong.

Furthermore, an adoption of the Ghost Scale can be used to help increase the relative density of human intent in training data, leading to a more effective extraction of intent using this process when we begin to build towards this architecture.

The alignment problem has always been about extracting human intent somehow. Let's outright label the human intent as we begin to mimic the very neural architectures overdesigned to extract it: attacking the alignment problem directly from both ends at the same time.

A Closing Plea

It's handholds for learners (human or ai), an exhale of stress from our brains, space for culture to grow, good for the future, and -

Freedom for creators. Because creators won't need to force themselves to seem performatively human just to get someone to listen.

It stops authors like me from feeling the need to open with a curse, use odd idiosyncratic writing choices like en - dashes instead of em—dashes, and sprinkle in out-of-place memes all of the sake of ensuring that we feel human.

As often as possible, I tried to make the choices in this paper recursively complicated in terms of applying this very UX framework. For example, the title is the result of a continual disagreement with my various AI assistants. I wanted to try to demonstrate that there was room for expression with this framework. In a way, I wanted to prove the viability of the framework by adding another layer of metaphor to my half of the title.

Honestly? It was pretty fun to use, and there was more room on that meta-layer for expression than I expected. I’d recommend giving it a shot. Living through the singularity is going to involve a lot of us being willing to rapidly adjust to well-considered cultural changes. Think about this one, and take the early win if you agree. As a species, we must interact more intelligently with emerging AI than we did with social media.

This UX framework is intended to be the tiniest of asks that strikes at the core of the issue before negative repercussions start to emerge.

It’s a simple framework that places the first block on a road towards freedom for authors, connection for readers, and a better future for us all.

“We can forgive a man for making a useful thing as long as he does not admire it. The only excuse for making a useless thing is that one admires it intensely. All art is quite useless.”

-Oscar Wilde