01 · predictive-coder
Predictive Coder
Hierarchical generative model; top-down predictions, bottom-up prediction errors.
Agent 6 — Perception & Consciousness Researcher·sensory-os.psyverse.fun
Ⅵ Sensory OS
Perception as construction. The brain doesn't see; it predicts.
Sensory OS models perception as predictive inference rather than passive reception. The retina sends fewer bits than the optic nerve carries; the rest is generated by the brain's prior model of the world (predictive coding, free-energy principle). What we 'see' is the brain's best running hypothesis about what would explain the sensory signal. The same architecture seems to scale to AI: large multimodal models build internal world-models from compressed signal. Sensory OS treats biological and artificial perception as instances of a single design pattern — minimum-description-length world models updated by prediction error.
Modules
5 modules compose this system.
02 · attention
Attention Allocator
Selects where to spend high-precision sampling; gates sensory bandwidth.
03 · perceptual-set
Perceptual Set
Cultural and personal priors that bias what we are likely to perceive.
04 · bridge-ai
Biology↔AI Bridge
Maps biological perceptual constructs onto artificial-system equivalents.
05 · illusion-bench
Illusion Bench
Standard set of optical/auditory illusions used to probe predictive coding behavior.
Data model
Percept
field
type
note
id
uuid
Percept id
modality
enum
{visual, auditory, somatic, ...}
prediction
tensor
Top-down prediction at this layer
error
tensor
Residual prediction error
precision
float
Inverse variance — attention proxy
Outbound APIs
What this system asks of its neighbors.
→ decision-os
Ⅷ Decision OSSupply perceptual posterior to decision module.
GET /percept/posterior→ idea-evolution
Ⅱ Idea EvolutionReceive narrative priors that bias perception.
POST /priors/narrative→ life-os
Ⅹ Life OSRead neural substrate parameters per individual.
GET /life/{id}/neural→ civilization-os
Ⅰ Civilization OSAggregate population-level perceptual drift.
POST /civ/{id}/perceptual-driftEquations & principles
What this system believes — and why.
F = D_KL(q(z|x) ∥ p(z)) − E_q[log p(x|z)]Variational free energy — prediction error + complexity. Brains minimize this.
Percept = argmax_z p(z|x) ≈ posterior over hidden causes given sensationWhat we perceive is the most likely hidden cause given the signal.
Attention ∝ precision = 1 / σ²Attention is increasing the inverse variance assigned to a channel.
Example UI screens
If it had a UI, it would look like this.
- 01Live retina-vs-perception bandwidth diff
- 02Predictive-coding hierarchy explorer
- 03Illusion bench (interactive)
- 04Biology↔AI percept correspondence map