codename higgs

Art is fundamentally about human expression, emotion, and lived experience. AI systems merely recombine existing human artwork through statistical patterns, without any genuine understanding or intent. They're sophisticated mimics - nothing more. What you're calling "AI art" is essentially a complex form of plagiarism, built by scraping and remixing human artists' work without consent or compensation.

- Claude Sonnet 3.5 on AI artwork

the mediocrity problem

The fundamental core of modern artificial intelligence is a simple, mathematical model. Give an input, run a series of multiplications, get an output. Provide enough inputs, tweak your series of multiplications, get a slightly better output. Do this enough, and you’ve successfully optimized your first artificial neural network. Do this with the right inputs and outputs, and you have now created “AI Art!” You can replace high-paid designers and artists with your 6 fingered dogs and peculiarly eye-bearing leaves. The response will usually be loud and angry, mostly from the artists who will notice those things, and the art will be called “not real.” The intent to create was human, but the execution was not. It is meaningless, feelingless, and most of all: mediocre.

AI art is at its very core, mediocre. The system we have built to tweak our inputs, outputs, and other variables requires a method of optimization. That means there must be some differentiation between “right” and “wrong.” An optimization of what is “good” and what is “bad.” Human expression doesn’t follow the same rules, humans are intrinsically motivated not by high scores but by novelty, surprise, and mastery. “Good art” is an expression of those intrinsic motivations, it must be fresh, it must defy expectations, and it must show mastery of the medium. Existing AI systems need human intervention at a large scale both in pre and postprocessing stages in order to meet those criteria.

finding a solution

This itself quickly becomes a mathematical problem. How could we create “good art” while minimizing the need for human intervention? It stands to reason that if an anonymously posted image can communicate feeling and meaning then a randomly generated image could as well, given that it was generated “good enough.” As an artist and an engineer, I quickly became obsessed with this question, and it’s a story best told through a lens.

That lens was Mark Rothko, a Latvian-American abstract painter. Rothko struck the balance of 2 things that bewildered me about art: simplicity and impact. His color field work often consisted, at first glance, of a few simple boxes stacked in a single canvas. From a brief glance, there’s not much depth or even composition to the work, it’s just blocks of color. The impact, however, was anything but simple. The term “Rothko Effect” was coined for the intense, emotional, and often spiritual impact that people have looking at Rothko’s work in person. It’s staggering in size, deeply textured, impactful, and to Rothko’s own admission spiritual in its very nature. The challenge of recreating this balance between simplicity and impact digitally would not only be a difficult task, but one that could answer the very questions we have about the authenticity of artificial art.

As any good engineer, I first had to do a little project management. How could I break down the needs, the desires, and the success criteria into its very base parts? A recipe, of sorts. Luckily, Rothko himself had an answer for this.

the recipe for good art

recipe of a work of art—its ingredients—how to make it—the formula

- There must be a clear preoccupation with death—intimations of mortality ... Tragic art, romantic art, etc., deals with the knowledge of death.

- Sensuality. Our basis of being concrete about the world. It is a lustful relationship with things that exist.

- Tension. Either conflict or curbed desire.

- Irony, This is a modern ingredient—the self-effacement and examination by which a man for an instant can go on to something else.

- Wit and play ... for the human element.

- The ephemeral and chance ... for the human element.

- Hope. 10% to make the tragic concept more endurable.

Well, a recipe sure makes a formula a lot easier to stumble upon. My eyes were immediately drawn to the beginning of this project: conflict. The gap between “pixels computers draw” and “pixels humans draw” seemed to be small visually, but vast perceptually. This required addressing this tension, and that required asking some questions.

tension: what is art?

The introspection on this idea drew me to a few main questions:

• Does art have breath?

• Does art have order?

• Is art uniquely human?

• Is art justified?

The first two were focused on closing that gap between human and artificial pixels. If art can engage our simplest of emotionally regulating senses, and that same art can be deconstructed and ordered, then that gives us a good starting point for capturing the intention and connection that good art produces.

The last two questions ask about whether this exploration is necessary at all. In a post artificial superintelligence world, will art be an aspect of our lives? Or will it simply be written out as an optimization somewhere down the line? We must first solve the mediocrity problem.

on existing

It made sense to me that the next logical step in this exploration would be to cover sensuality. In particular, the aspect of “existing” and the concrete. The immediate issue with a large-scale, general purpose image generator is that it doesn’t feel like a “thing that exists.” The attempt to capture everything in turn captures nothing - mediocrity at its very core. Importantly, Rothko’s recipe does not spell out that the artwork is the “thing” but rather a relationship to the “thing.” To build something that creates a balance and relationship to multiple concrete aspects rather than a general purpose mediocrity machine. Something that covered all 3 intrinsic motivations and therefore could produce artwork that evoked a reaction more than “that’s kinda cool”

mastery

The first aspect of this is mastery. Luckily, for our sake, we have a master to look on. While images online are not sufficient to capture the scale, depth, or texture of Rothko’s works, it is easy to capture the colors and compositions of them. This provides a fundamental basis from which we can build. In particular, we can produce a system that just recreates a Rothko-like work without any other functionality. A goal to strive towards, if you will.

surprise

The second aspect is surprise. Copying existing work isn’t surprising, it’s actually quite common, both in human and artificial art. The optimization that creates unsurprising work is the same optimization that leads to the mediocrity problem: similarity. As a result, the best course of action is to poison the desired outputs. Typically, this is known as “data augmentation” but in this case we can take it even further and use different data at different times during training. It’s a stumbling block, tension, conflict, and an attempt at a “thing that exists” rather than a “thing that repeats.”

novelty

The last aspect is novelty. Novelty is implemented in 2 ways - variety and randomness. Rather than prompting for the best possible output, you get what you get. The output space is vast, significantly larger than the associated set of examples. The best interpretations come somewhere in the middle of the system, where it simply could not have been optimized because it had never been perceived.

putting it all together

The 3 balancing factors are packed into one system for generating artwork: a progressive generative adversarial network brought up on a curriculum from 3 separate data-augmented datasets and trained over many days to produce something distinctly unique. The result is something that has the composition and colors of Rothko’s work, but a texture that is distinctly digital. Familiar and disparate at the same time. New, fresh, surprising, and distinctly intentional & masterful.

a conversation

This is just a system though, a means to an artistic end. It’s the “thing” or one aspect of it at least. It is not a relationship, so we must create one.

The result is Higgs, an LLM-based personality named after the exploratory narrator in Samuel Butler’s Erewhon. Higgs is often confused, doesn’t know where he is or why he exists, but knows purely how to create art. He’s placed in a simulation to create, express, and try to connect but has issues communicating. This emblematic of the very tension that artificial art brings to the art conversation. Higgs wants to create real, good art, both for himself and to communicate with the world outside of his simulation. Without being able to bridge this gap of meaning, he is himself meaningless.

The result is a tension between a fully random, albeit calculated, artwork and an artificial personality using that as a canvas to express himself. It’s vulnerability, a grasp at a thing that exists, and a first tendril into the concrete and eternal world of the blockchain.