The Algorithmic Weltanschauung: Agency and Emergence in an Algorithmic Soup

Reality is just persistent compression: a unified theory of life, mind, and physics built from the single idea that agents are patterns that survive by modeling their world.

Kolmogorov Theory (KT) starts from an unusual place. Instead of assuming an objective universe and then trying to explain why there are observers in it, it starts with the undeniable fact that something is being experienced and asks what kind of computational structure could produce that. The answer it gives: an agent is any persistent pattern that compresses its sensory history into a model, uses that model to predict the future, and takes actions to keep itself from dissolving. The authors call this drive telehomeostasis — not homeostasis in the sense of maintaining body temperature, but maintaining the pattern itself, the algorithm that defines what the agent is.

The "algorithmic soup" metaphor is central. Imagine a primordial sea of interacting programs — mutating, replicating, competing. Natural selection filters for programs that persist. The ones that survive are the ones that build good internal models of their environment, because good models yield better predictions, better actions, and therefore longer survival. This is not a metaphor for biology; in KT it is biology, and also cognition, and also physics. Every agent — bacterium, brain, corporation — is described by the same three abstract modules: a modeling engine that compresses history into a finite state, an objective function that scores possible futures, and a planning engine that picks actions to maximize that score.

What makes this more than philosophy is the connection to Kolmogorov complexity (a formal measure of how much information is in a string, defined as the length of the shortest program that produces it). Physics, in this view, is just the set of maximally compressive regularities that resource-bounded agents can discover. Space, time, and causality are not fundamental — they are the compression artifacts that happen to shrink description length the most. Emergence, similarly, is redefined: something "emerges" when an agent empirically finds a concise macro-level model from coarse-grained data that it could not have derived from microscopic laws alone. Randomness and probability are not properties of the world; they are what you get when your compression budget runs out.

The framework also scales upward naturally. Tightly coupled agents can collectively satisfy the same four criteria, forming meta-agents — ant colonies, immune systems, firms — without any additional machinery. Markov blankets (the statistical boundary separating an agent's internal states from the outside world) simply stack. The paper is a late-breaking abstract, so many claims are stated rather than derived in full, and the source does not make all mathematical details explicit. But the conceptual architecture is clear: life, agency, emergence, and subjective experience are four faces of the same thing — the survival of compressive patterns in an algorithmic soup.