About Project Causa Sui — for Readers Who Don’t Do Philosophy

What Cannot Be Saved

Anyone who has played computer games knows the “Save” button. You save before a dangerous spot, and if your hero dies, you load the save and everything starts again from the same place, as if nothing had happened. The hero does not remember dying. Life has no such button. A day once lived cannot be loaded again, and what you have learned you cannot stop knowing. A mistake, a hurt, a lucky guess stay with a person for good — and change them.

Project Causa Sui begins with a question: is this where the line runs between a subject — someone who acts on their own, answers for themselves, can be afraid and can make mistakes — and even the most complex machine?

Why “Causa Sui”

Causa sui is Latin for “cause of itself.” In classical philosophy the term was reserved for what depends on nothing outside itself: for the scholastics and Descartes, God; for Spinoza, substance, whose essence already includes existence. Nietzsche called the idea a contradiction — nothing can pull itself out of nonexistence by its own hair. The project keeps the essence of the expression but applies it not to the infinite, but to the finite and mortal. It is not about God or creation from nothing. A finite being, too, can be a cause of itself: the source of its action cannot be taken outside, separated from its body and copied onto another carrier. Every act irreversibly changes the one who acts, and the next act is performed by someone already changed. A subject does not create itself from nothing; it keeps building itself out of its own history.

Not a Question of Complexity

Artificial intelligence writes texts, solves problems, holds conversations. Many expect that once it is complex enough, it will become conscious. The theory puts the question differently: what matters is not how complex a system is, but how it is built.

Every modern computer, however clever, has one property: its state can be saved and later restored exactly, bit for bit. It can be stopped, copied, started again — and it carries on from the same place. According to the theory, a system with this property cannot be a subject. Not because it lacks power, but in principle.

Program and Carrier

A computer has a carrier — the hardware — and a program. The program can be moved to another computer and will run the same way. Carrier and program are two different things.

Living things are different. Take someone who has learned to ride a bicycle. The skill is not a file. It is written into changed connections between brain cells, into muscles, into the body itself. You cannot take out the “riding program” and copy it to another person. And learning itself changed the carrier — irreversibly.

The theory takes this as its starting point: in a subject, the program of action and the carrier cannot be separated. This gives a test. Make a copy of a system and compare it with the original. For a computer program the difference is zero — the copy is indistinguishable. For a subject the difference is not zero: an exact copy does not exist, only a “near-exact” one. Zero or not zero — that is the line.

Why It Matters: Fear and Responsibility

If any mistake can be undone by loading a save, a system has no reason to fear and nothing to answer for. A subject, according to the theory, is one who has something to lose. Every event leaves a trace in it that cannot be erased, and closes off some possibilities. That is why pain in the theory is a broader notion than damage: it is any influence that irreversibly narrows what the subject can still become. Damage is only the extreme case.

What About Immortality?

The second book of the project applies the same yardstick to transhumanism — the movement that has long promised to defeat aging, extend life, and expand the powers of mind and body. The theory accepts the goal: to widen what a person can become. But some of the promises have to be called illusions.

The most popular promise is to “upload” a mind into a computer. But a computer is exactly the kind of carrier whose state can be saved and restored. An uploaded copy may speak in your voice and remember your memories, but according to the theory it will not be a subject — let alone you. The same goes for freezing the body in the hope of reviving it later. These are not foolish ideas but drafts: they were proposed by people who did not yet have a theory of the carrier.

What remains. First, expanding abilities within one living body: prosthetics, neural interfaces, gene therapy. Here the carrier is never interrupted, and the goal is within reach. Second, continuation beyond a single life — not as the endlessness of one body, but as a chain of mortal successors. Each next living carrier receives from the previous one not life but content: accumulated experience and values. The transfer is never exact, so from link to link the personality drifts — at first almost unnoticeably, then in a leap, then stably again. The book does not avoid the uncomfortable question either: if not everyone can continue themselves, who decides whose content is passed on? It shows that no basis for such a selection solves the problem, and it names the parallel with eugenics outright.

From Idea to Test

One AI model, after reading the first two books of the project, wrote that it “offers not a finished answer to the problem of subjecthood, but a new principle for posing the problem itself.” It added that the main thing was missing — a rigorous apparatus, without which disputes remain disputes about words. That was fair. The third book, “Paths of Implementation,” is devoted to exactly this.

It presents a program model of the subject: a set of skills (hunting, sleep, searching for the new), each with its own records. The set changes under the pressure of circumstances in three ways: a skill is tuned, a skill grows, and, in dire need, the set gains an entirely new skill.

The model lives in a simulated world, and this world is deliberately simple. A creature in it has three states: hunger, hunting, sleep. The world changes every hour, and a life lasts up to 400 days. There are other ways to get food, but each carries a risk: mushrooms, some of them poisonous; fruit on a tree you can fall from; water, which does not feed you but helps you hold out. As long as there is enough food, the creature never uses them: when life is good, there is nothing to change.

Four times in a life, hard times come: prey becomes scarce and hunger sets in. Only then, at the edge, is the new born — the creature tries mushrooms or climbs a tree. The new can also come another way: the creature sees a neighbor who ate a mushroom and was fine, or a neighbor who was poisoned — and remembers. The new is not salvation but a risk: you can be poisoned or fall.

Three groups were compared. Creatures unable to produce anything new all died in the very first hard times. Those able to produce the new and having memory survived the first hard times in 60% of cases. The most interesting is the third group: these creatures found the new even more often than the others, but did not remember it and started from zero every time. The result over a whole life: of those who did not remember, none lived to the end — 0 of 200. Of those who remembered, 52 of 200 did. Novelty without memory does not become experience.

An important caveat. The creature in the program is not a subject: its state can be saved and restored, so by the theory itself it cannot be a subject. The simulated world is like a wind tunnel: the aircraft in it does not fly, but the shape of the wing can be tested. It tests one thing — whether the theory contradicts itself when the mechanisms it describes are implemented in a program. No contradictions were found.

There is also a step beyond the model. For experiments with living cell cultures, which can already learn simple tasks, a prediction has been written down and published in advance: what should be observed if the theory is right, and what would refute it. The prediction was recorded before the experiment, so that the result cannot be fitted after the fact.

Honestly About the Limits

The starting point of the theory is accepted, not proved. That is how every axiomatic theory works. When mathematicians gave up Euclid’s fifth postulate, geometry did not collapse — a different geometry appeared. So “I disagree with the postulate” is not yet criticism. Criticism means finding a contradiction inside the theory or refuting its prediction.

The theory does not solve the “hard problem of consciousness”: why irreversible damage is felt from the inside as pain rather than simply happening. And it openly admits that no external test of behavior can tell a subject from a good imitation. The dispute is settled not by watching behavior but by how a system is built.

What Follows

If the theory is right, an artificial subject will not appear simply from more powerful computers. A machine can grow its structure by a given rule, but it cannot change the rule itself. What is needed is a different carrier — one in which what has been written cannot be erased and restored — and therefore a different kind of engineering. If the theory is wrong, it can be refuted, and there are concrete ways to do so.

Either way, the question moves from the realm of opinion to the realm of testing. That is what the project is about.