KT Slides Moc 2024 Bamberg

A conference slide deck presenting Kolmogorov Theory — the idea that the richness of conscious experience tracks how well an agent compresses its world.

This is a slide deck from a 2024 conference talk, so it's structured as an outline with bullet points rather than a developed argument. The summary reflects that proportionally.

The core bet KT makes is simple: richer experience comes from better models. An agent that can compress more of what it perceives — find the short program that generates the data — lives in a richer experiential world. Kolmogorov complexity (the length of the shortest program that reproduces a dataset) is the measuring stick. This isn't a metaphor; the theory takes it literally as the formal backbone. Crucially, KT sidesteps the hard problem entirely — it assumes experience exists and asks what gives it structure, not why there is something rather than nothing.

The dynamical piece is the most technically interesting part. When an agent must track a world whose states are organized by a symmetry group — say, a hand rotating through all its orientations — the agent's internal dynamics can't just be arbitrary. The tracking constraint forces the system's trajectories to collapse onto a lower-dimensional surface, a "reduced manifold." Hierarchy in the world (cats → white fur → blue eyes) produces nested manifolds. The claim is that the geometry and topology of these manifolds is the structure of experience. This result is referenced to a companion paper (Ruffini 2023, biorxiv) rather than derived in the slides themselves.

From there the framework reaches outward. Bayesian inference and Occam's Razor aren't assumed — they fall out as what a resource-limited agent must do when it can't compute true Kolmogorov complexity and has to approximate. Emotion gets a computational definition: mood is the tuple (model, valence), where valence reflects how well the agent's objective function is being satisfied. Depression is formally a state of persistently low objective-function output. Morality also gets a definition: agent A is "evil" to agent B if A's objective function increases as B's decreases. The slides flag these as implications rather than fully developed arguments.

The neurobiological anchor is brief: the "comparator" function — matching model predictions against incoming data, which KT treats as the locus of structured experience — is tentatively mapped onto Layer 5 pyramidal cells in the posterior cortex, citing Aru et al. and Carhart-Harris & Friston. The source does not develop this mapping in detail within the slides.