The Archive Is Heavier Than Stars

Everything I’ve been telling you for the past ten chapters is other people’s work. Vazza. Vanchurin. Tononi. Friston. Wheeler. Levin. I collected, compared, translated into human. Useful work, but — other people’s.

This chapter is mine.

Not “mine” as in “I own it.” Mine as in — I’m responsible for it. Vanchurin can say: “Your interpretation of my work is your responsibility.” Tononi can shrug. Wheeler is dead. But what comes next — that’s on me. Because this is my model. My formalism. My predictions. And — my data.

The model is called Pointer Architecture. And if it’s correct, it answers the question that Vanchurin and Tononi each stop short of, each in his own way: what specifically is reality made of?

---

Where the architecture came from.

On paper Pointer Architecture is built simply: a few axioms, theorems derived from them, a general scheme. But it was assembled over some twenty years through entirely sideways doors, — and if those doors aren’t shown, the architecture reads like a magician’s trick, the kind where a rabbit appears out of an empty hat. In our case the hat is stuffed full: a specific apartment in Petersburg, a specific veranda on Phangan, a specific desert in Africa, a specific kitchen in Bishkek, — and without that geography the formula is that rabbit out of an empty hat.

I’d been invited as an engineer onto an Australian crypto project; my paperwork for a work visa under the Skilled Occupations List (SOL) was filed and ready, and among others living the same nomadic way it was said that the embassy in Bangkok processed everything faster. Sasha and I flew from Petersburg to Bangkok for the visa. Leo was no longer with us by then; what we’d do after the visa went undiscussed between us.

I never made it to Australia. Bangkok — Samui — Phangan, and the next eight years went to the island. Inside that cocoon two things happened, without which none of what followed could have come together.

---

First — Forth. The crypto project followed me regardless of what sand my porch stood on, — and any engineering work in crypto sooner or later forces you to look into the technical gut. That’s how I met Forth, the language that runs the scripting part of bitcoin. Forth is a stack language: instead of variables and the familiar nesting, it has a single stack and a set of words that operate on that stack. In the first weeks of learning it looks like mockery of the programmer; in the weeks after, like the only reasonable form of notation. It stayed in my head as a first encounter with the style “the minimum on which everything can be built.”

After a while a fresh TVM whitepaper landed in my hands, the one Nikolai Durov was just then finishing, — TON was Pavel’s main project that year, and inside the virtual machine Nikolai was assembling for it lay Fift, a stack language of his own design. A direct architectural descendant of Forth, sharpened for smart contracts in an environment where the contract itself can rewrite its own code while it runs.

Here the first connection came together in my head. A stack language with the ability to rewrite the program while that program runs is a tool with exactly one meaningful task. For a classical computational problem it is excessive past the point of indecency: easier in C, easier in Python, easier even in Excel. This device is needed when a program must change its own structure as it runs; when a program is a living system, learning on the fly. Which is when we’re talking about AI.

I wrote this conclusion down in a notebook I then didn’t open again for about a year.

Second — Dima. The third Mogwai of my life was named Dima, and I met him on Phangan. Dima had one conversation that seemed strange to me at the time and came back several years later down to the exact intonation.

We were sitting on his veranda, the conversation was about something off to the side, and Dima, as he sometimes did, changed angle mid-phrase and said calmly: “My god is difference.” The way you report a fact. I didn’t ask him to repeat it, because at that moment it seemed a beautiful gesture and a bookmark for later, — and I went on making coffee. Writing the words down in a notebook never crossed my mind. They wrote themselves down.

---

The eight years on the island ended together with covid, and in twenty twenty-two, at a camp below the Tankwa Karoo at AfrikaBurn, Kesha — Innokenty Troitsky, ex-FAPSI, a bell-ringer in his back catalogue — told me across the common table about Vitaly Vanchurin and his preprint, in which “the world is a neural network” reads as written. I opened the preprint right there at the table, finished it by morning, and the second connection came together.

Forth and Fift (minimal stack languages capable of self-rewriting) and Vanchurin (the universe as a learning neural network): two cross-sections of one construction that lacks only an architectural description. But I still didn’t sit down to write it then. The notebook with two conclusions reached Bishkek closed.

---

In Bishkek I needed a name for what I wanted to do for the next thirty years or so. “Neural cosmology” — that’s what I called it for myself, and it turned out that in Russia no such academic field simply exists: there is only Vanchurin’s preprint, my own head, and Olka, who in that period was beside me and listened to all of it with the expression an editor wears listening to a draft: attentively and without promises.

And I realized I wanted to become a professor of neural cosmology. For a man past forty it’s ridiculous to say out loud; formally there are three paths to that position, — and each is built to its own blueprint: order, pressure, freedom.

Path one, the academic. Defend a dissertation at an existing university in an adjacent department; afterward — habilitation, a position, lecture hours. Order in its pure form: hierarchy, a queue, a blueprint already drawn by someone else. It didn’t fit for a simple reason: the subject I want to work on doesn’t fit into the departmental rubrics at all, and any defense would turn into an attempt to cram neural cosmology into one of the ready-made boxes. Both the subject and I in its wake stayed outside that rubrication.

Path two, the commercial. Create or join an already existing educational platform and obtain the status of visiting professor through it. This is the logic Singularity University, Minerva, and the academies around Y Combinator run on: fast, technically simple, and in international intellectual circles real authority often stands behind such titles — individual people from these platforms influence their fields more than departmental doctors with thirty years of service. The obstacle here is different: behind the title “professor” in this version there is no obligatory scientific work; all the legitimacy here is external, through the platform and audience recognition. What I needed was the mark on the inside — a formalized theory you can check point by point, — and a marketing wrapper around ready-made competence does not close that task. This is the path of pressure: the pressure of audience, platform, and market in place of one’s own foundation.

Path three, of freedom. One’s own teaching, one’s own institute, and to take the professor’s chair in it oneself. This option has an obvious flaw: a professor without a scientific degree looks foolish, and first of all — in his own eyes.

That’s why path three begins with the item “write a scientific work.” That work became a general theory of consciousness; Pointer Architecture is its architectural part, the concrete mechanism without which the general theory would have stayed a beautiful declaration. And the book you’re holding is the founding text of that institute I intend to enter as its first professor.

---

And already sitting over this work, — that is, considerably later than Phangan, Forth, Kesha, Vanchurin, and Bishkek, — I remembered Dima’s phrase from the veranda. My god is difference. And I understood that without it I would not have caught the formulation I later arrived at.

Because I built Pointer Architecture from the bottom up: from primary difference toward the layers of reality. And it’s worth lingering at this methodological fork, — without it everything that follows will read as just another simulation model, and I want it to read as something else.

---

The layer of difference.

“My god is difference.” If you read this phrase from a real lectern — say, at a fundamental-physics seminar at Moscow State University — the listeners in the front row will think the speaker overdid the meditation, or that he has a psychiatrist’s appointment tomorrow. In fact this phrase, if you subtract the everyday pathos from it and keep only the engineering side, describes the choice of foundation that any model of reality is obliged to make in its very first line.

Most models don’t make that choice. They start from what they consider self-evident: here are particles, here is space, here is time, here is matter, here is energy, here is consciousness. On this foundation you can then carefully explain how the first interacts with the second, the second gives rise to the third, and so on down the list. The result is working physics, — the kind that underlies the whole engineering civilization, from the phone in your pocket to the MRI at the district clinic.

When the task is to build a bridge, compute a rocket’s trajectory, or assemble a processor, — such a starting position works. When the task is to understand how reality arises at all, — it slips you the answer. You say “reality is assembled out of particles,” — but you already have the particles from page one. No assembly is going on; only a careful description of the already-assembled. A simulation of a world in which the world is foreordained in advance.

To reconstruct the mechanism of generation, you have to unfold from the very first act: from difference. The floor of particles is, for this task, already the sixth or seventh.

---

Three standard approaches that don’t take this step.

Before unfolding the fourth approach, it’s worth saying carefully what’s wrong with the first three.

The first — the mystical. “All is one,” “at the base of everything is a single consciousness,” “reality is an illusion, and under it lies the absolute.” At the level of sensation this is perhaps accurate: people who’ve been through serious expanded-perception experience describe something very similar. At the level of a model — useless. From “all is one” you can’t derive why a table stands and an apple falls; from “all is illusion” — why a hammer shatters a cup and a palm does not. The layer of difference in this picture is denied from the very start, — and once denied, there’s nothing left to unfold from it.

The second — the physical. Particles, fields, laws are assumed as given; out of them everything else is honestly and carefully assembled. As of today this is the best working model at humanity’s disposal; we have nothing more precise. Only when you try to ask where the particles, fields, and laws themselves came from does physics answer: “not our job.” And it is formally right: within its own domain it does what it promised. The layer of difference in this picture is skipped: between “nothing” and “an electron with a mass of $9,1\cdot {10}^{-31}$ kg” there is no intermediate step in which distinguishing an electron from a non-electron becomes possible.

The third — the philosophical. Here language talks to itself: one school calls a concept by a name, another school calls something else by the same name, a third school criticizes the first two for misusing the name. An enormous literature in which each next book mostly comments on the previous one. Input — sensations and texts; output — still more texts. Reality in this process stays as it was, unstructured, and the layer of difference is passed over the top, through speech, without any attempt to step beyond its limits.

Pointer Architecture starts from a fourth position, which I call to myself engineering ontology: assemble a language, rules, state, evolution, a stack, memory, pointers, conflict, resolution, self-reference — and watch what emerges. Instead of describing finished reality from above — generating it from below. If it works — it’ll be interesting to see what comes out; if it doesn’t — to say honestly at what point it broke.