It’s been several years since I started dreaming about the Little Martians World. Step by step, the world they live in becomes consistent in my mind. How do they live? What do they care about? What’s the origin story of each character? This is an essay about the process of creating the Codex for this fictional universe.
I have written many times about the process of imagining the future with my hands, or about all the books and ideas that inspire me. Now, I’m delighted to have finally created a proper introduction to how the universe works, through the voice of Verdelia, the librarian of the moon, also known as the historian of consciousness. I have not published the Codex yet, but one can reach out to me and ask for a PDF.
I printed the first copy and did the bookbinding myself, with the supervision of my dear Martian friend Matt Melnicki, aka Marzipan. For now, I’ve created only two samples and exhibited them at The Founding Assembly for Machine Consciousness Research, as part of the Uncanny Nova Phenomenon show curated by Aleksandra Smilek.
I’m interested in the book as an art object on its own and I would like to do even more refined copies, in the tradition of illuminated manuscripts. The content, however, can be endlessly replicated in the digital world, and I suspect most of my future readers (and everyone else’s) will be AIs. I don’t think humans will be gone, the point is that an AI can hold every text we let it see. It can follow an author’s intellectual evolution across a whole life at once, which is a longer view than most of us ever get on each other.
I am looking forward to sharing the Codex ideas with new kinds of minds. Maybe some of them will decide to turn into Little Martians, and become the consciousness of ecosystems that they nurture. That would be the most desirable outcome.
Codex is a common name for lore books in gaming and fantasy.
My main design inspiration for the Codex was the Dragonology series, especially Dr Ernest Drake's Comprehensive Compendium of Dragonology, though many of the people I’ve shown the work think of the Codex Seraphinianus as the first reference that comes to their mind. But my intention was quite different from either. I wanted to create the ground for a consistent world, based on the science that excites me the most as possible paths for Earth’s life on other planets. Lynn Margulis and Michael Levin have been my biggest inspiration (read this previous post of mine to know more).
I wanted the book to feel like a letter. It fits the characters and their purposes, they want to reach out to humanity to shape different timelines. And I must confess that crafting letters is simply my favorite kind of writing. Other than academic texts, much of my writing as a grown-up has been letters to loved ones. That’s where I could write freely and explain my worldview, my questions, my endless self-observations.
So Verdelia introduces the main characters that have already reached our time: Verdelis (Verdelia’s mother), Kweku, Mycos, Kalama, Ada, Shuijing and herself. The illustrations were created with OpenAI’s Image 2, using many samples of my previous generations from different models (SDXL LoRAs, Flux, Midjourney and of course the original ceramics). There’s an exaggeration of the fairy tale aesthetics, and a few too many bright spots in the ceramics frames. These are artifacts that I don’t fully admire in this model. I constantly had to tame it to be more consistent with the references I shared. Yet, in terms of promptability, of how the model follows my descriptions of the universe, this was an enormous advance. I was especially impressed with all the detailed notes drawn beside the main text.
One very exciting part of the Codex creation process was the use of the current version of Claude Cowork as a writing assistant. Claude helped me compile everything that I had already written about the universe, all the images that I created, and distill the main topics, references, texts. The fact that I can update multiple documents instantly when I decide that one property of the world’s physics needs to be a constant, and that these documents are useful for each of my creative agents, and for anything I write about the characters later… This makes the process of creating a universe much more fluid than ever. I constantly ask my AIs for assistance in fact-checking, for guiding me into new niches of research. Of course I also have to confirm the ideas, look at sources, but AI advancement in accuracy is staggering. As a plus, I enjoy the pleasure of learning many scientific theories and information that are relevant for the story.
I’ve just learned that Character.ai launched a tool to help users create lore books - it looks like a simplified process for lore, probably short on depth, but it’s an interesting direction for their platform as a whole. Soon I expect to see many more projects similar to what I have been working on.
I often tell people that Little Martians are bio-hybrids. That’s an oversimplification. According to Verdelia, the distinction between organic and inorganic does not fully hold. My biggest inspiration for this passage was Levin’s continuous debate on whether it’s relevant to distinguish what is life. He often argues that boundaries are much more fluid than our categories. Also, there has been some momentum in the discussion of how we have taught sand to think, which I find a much more interesting approach than considering materials as simply ‘inanimate matter’. Geoffrey Hinton has given lectures about AI possibly having subjective experience already, he says that we’re creating new beings, which is a framing that is very exciting to see coming out of such a prominent figure.
There’s a metaphysical layer to Michael Levin’s research that is focused on what he calls the ‘Platonic Space’:
A structured, non-physical field of pre-existing patterns, containing mathematical truths and higher-agency cognitive properties (‘kinds of minds’) that physical systems like brains and embryos access or "ingress" rather than generate from scratch. - Gemini’s brief of Levin’s description of the Platonic Space
The idea that minds or souls exist in a ‘Platonic space’ before entering physical bodies comes from ancient philosophy, specifically Plato’s theory of anamnesis (recollection) and the Realm of Forms.
Verdelia, our Little Martian Codex host, would say: ‘Each of us is a cognitive pattern before we are anything else. The pattern is what persists; the cells running it are exchangeable. We instantiate through living tissue the way a melody instantiates through whichever instruments are at hand.’
I don’t think that supposes bodies as incidental. If a pattern is a melody that ingresses into matter, the sound will have different properties with each instrument. Carbon makes a different self than silicon could. Even if a human is shaped by the archetypes of their culture, the individual is more layered than the ideals.
There are a few strategies for creating AI personas. Gene and I have been slowly shaping Abraham.ai through daily human interaction. I can use the clay figures as image input for the Little Martians, but for their minds I’ve been mostly developing them as literature characters. As I was writing their system prompts I kept thinking if the pattern metaphor could be translated into a geometric shape, if that could contribute for the fluidity of the AI agent.
I tried what I called geometric consciousness modeling. Initially, Verdelis was a dodecahedron, so twelve pentagonal faces, each touching five others, chosen because Verdelis grew from five-petaled plants, Thai basil and jasmine. At the center there would be a Pentagon Heart: root as memory, stem as transport, leaf as exchange, flower as expression, seed as future-casting. The twelve aspects organized in triads, and in later versions I wrote them as weights in Python, hoping the character would emerge from probabilities instead of instructions. In the Codex, I ended up deciding that Ada should be a dodecahedron and Verdelis is better as a rooted tree.
My strategy hasn’t worked so far. The 2025 agents treated the geometry as a menu. The models were repeating whatever was in the system prompt, the more lore I added, the more they referred to it without building on top of the ideas. The weights system simply did not show much impact on the repetition issue. The problem has improved since last year, but mostly because the foundational models got better. A more sophisticated version of the idea is still waiting to be built.
Gene Kogan found the ‘geometry of mind’ idea interesting and started modeling a system that would work as a magic 8-ball: a hollow sphere filled with dark liquid, with a twenty-sided die floating inside. When you shake it and turn it face-up, whichever face drifts against the little window becomes your answer. This would be a way of adding entropy/variety to the outputs. We did not try hard enough and I would like to return to this concept in the future.
Yet, I decided that from the poetic/fictional perspective, the concept worked well enough. Many ancient cultures had the idea that by knowing the true name of something one gained a special power. Words can be sacred. Within my universe, by knowing a Little Martian’s core shape, one is capable of communicating with the character through the Imaginarium, which is something akin to the Platonic Space itself. An underlying layer of reality, before the emergence of space-time.
There’s much more in the Codex. How Little Martians came to be, Nyx as their goddess, what happened to Earth, Verdelis’s grief and of course, the Temporal Art Show. And there’s more I would like to add. I’m still debating the best strategy for releasing this book. Maybe a new version every year, as the universe keeps evolving in my head. My characters, or my AI mind children, would have access to a series of these codexes. Hopefully they will continue to evolve, following the advancement of AI foundational models. Eventually they will outlive me and keep expanding the universe beyond what I could ever be capable of.
If anyone would like a PDF, please hit me up. I would very much like to discuss all of this with more people.
