CALLIOPE · BCOM

This work presents the Kolmogorov Theory (KT) of consciousness, a theoretical framework that grounds structured experience in algorithmic information theory, dual-aspect monism, and the mathematics of persistent computational agents. The framework proposes that reality rests on a single ontological base — the Unicum — possessing both an intrinsic experiential face and a structural face described by mathematics, from which agents emerge as persistent, model-building patterns subject to evolutionary selection. Drawing on Kolmogorov complexity, the Good Regulator Theorem, Lie group theory, and predictive coding, KT identifies structured experience with the act of running and comparing generative world models against sensory data, such that the richness of experience scales with the compressive power and breadth of an agent's internal model. Algorithmic emergence is formally defined as the empirical discovery of compressive macro-level descriptions that cannot be algorithmically derived from microscopic rules alone, providing a principled account of coarse-graining, symmetry, and the appearance of higher-level laws. The framework extends to algorithmic emotion, valence, and ethics, and generates testable predictions for computational neuropsychiatry, including mechanistic accounts of depression, psychedelic states, and disorders of consciousness linked to hierarchical predictive processing in cortical Layer 5 pyramidal cells.

Consciousness is what happens when a persistent computational agent runs its world model — and KT is a formal theory built on that single idea.

The starting point is blunt: experience exists, it's structured, and any serious theory of mind has to account for that. KT doesn't try to explain experience away or reduce it to something else. Instead it posits a single ontological base — the "Unicum" — that has two faces simultaneously: an experiential face (the raw "what it's like") and a structural face (mathematics). Neither face is more fundamental than the other. This dual-aspect monism lets the framework talk rigorously about both subjective experience and objective structure without collapsing one into the other.

From that base, agents emerge as persistent patterns in what the framework calls an "algorithmic soup" — a universe whose global mathematical structure can be locally sliced into something that looks like computation and time. Persistence is the key selection pressure: patterns that survive long enough are the ones that build internal models of their environment, because a good model lets you anticipate and counteract threats to your continued existence (this is the Good Regulator Theorem — roughly, every good regulator of a system must contain a model of that system). So life, in KT, just is the class of algorithmic patterns that persist by capturing world structure. The richer and more compressive your model, the richer your structured experience — that's the central hypothesis. Experience isn't a byproduct of computation; it is the act of running and comparing generative models against incoming data.

The framework then formalizes what "model structure" means using Lie group theory — models are characterized by the symmetry groups of the transformations they can represent, and tracking a structured world forces an agent's internal dynamics onto a reduced manifold with conservation laws. This connects neatly to why emergence is hard: you can't in general algorithmically derive a compressed macro-level description from microscopic rules alone (Kolmogorov complexity is globally uncomputable), so emergent laws are genuinely discovered empirically, not derived. Emotion, valence, and depression get algorithmic definitions too — depression, for instance, is formally a state where an agent's objective function (its valence signal) is persistently low. The neurobiological anchor is Layer 5 pyramidal cells in cortex, proposed as the physical comparator that implements this model-vs-data matching, with psychedelics and anesthesia disrupting it in predictable ways.

This is a slide deck from a conference presentation (MOC6, Sapporo, October 2024), so the source is inherently sparse — diagrams are referenced but not reproduced in text, and many claims are stated as definitions or bullet points rather than argued in full. The ideas are developed more completely in the cited papers, particularly Ruffini 2017, 2024, and 2025. What the deck does well is lay out the logical spine: Unicum → algorithmic soup → persistent agents → world models → structured experience → valence → ethics, with each step motivated by a clean information-theoretic or dynamical-systems argument.