You never see the world; you see your brain's best guess, corrected by surprise
⟦ layer 2 · objective name perception as an action-tuned construction (predictive processing) rather than a passive window — and what that buys the framework: why the constant fades, why abstraction is possible, why heightened states see more · mode empirical (predictive coding / adaptation are testable) + ontological (the far-end interface reading) · status DEFENSIBLE-LIVE (predictive processing) · FOUNDED (sensory adaptation) · CONTESTED at the far end (Hoffman) · test de-habituation studies; adaptation psychophysics; ehi-passiko for the contemplative read · see vibe-calibration-protocol.md, 5-wisdom/breakthrough-streams/hoffman-interface-stream.md ⟧ The verifiable CORE: perception transmits prediction-error, not raw signal — so the predictable disappears and the surprising lands. The INTERPRETATION held separately: that this means reality "is unreal." Keep the core; label the leap.
You do not receive the world; you predict it, and only correct the prediction where it's wrong. What reaches awareness is mostly the brain's running model, updated by the error between what it expected and what arrived. Perception is a controlled guess, kept honest by surprise.
This is not mysticism. It is the most active research program in cognitive neuroscience — and it changes how every other Layer-2 mechanism reads.
Predictive processing / active inference: the brain is a prediction engine that continuously generates a model of its sensory causes and uses incoming signal only to correct the model. Andy Clark calls perception a "controlled hallucination" that happens to be tied to the world (Surfing Uncertainty, 2016); Anil Seth runs the same line in Being You (2021) — "when we agree about our hallucinations, we call that reality." Karl Friston's free-energy principle gives the formal spine: a self-organizing system minimizes prediction-error (surprise) over time, by updating the model (perception) or acting on the world (action). ⟦ empirical · DEFENSIBLE-LIVE — predictive coding has strong support; the free-energy principle as a universal law is itself contested as near-unfalsifiable; we use the local, testable version ⟧
The honest weight: predictive coding is well-evidenced for many perceptual phenomena (binocular rivalry, illusions, the way expectation literally sharpens or distorts what you see). What is not settled is the grandest framing (the whole brain, life, and consciousness as one free-energy minimization). We lean on the testable core.
Sensory systems transmit change, not steady state. Hold a constant stimulus and the response decays — neural adaptation at the receptor level, behavioral habituation at the cognitive level. The classic everyday proof:
You stop smelling your own house. A guest walks in and notices instantly — then, after twenty minutes, they stop smelling it too. The smell didn't change. The prediction-error did: once predicted, it carries nothing new, so it drops out of awareness.
This is rigorous and old (olfactory and tactile adaptation are textbook; you don't feel your clothes against your skin right now until this sentence makes you). ⟦ empirical · FOUNDED ⟧ Read through §1, it's the same principle: the predicted is silenced; only error is reported. The constant is known, therefore invisible.
This connects straight to the framework's information-theory spine. Shannon: the expected carries ~0 information — a perfectly predictable signal has zero entropy, nothing to transmit. So a perceptual system that suppresses the predictable is automatically an abstraction engine: it discards the redundant and keeps the surprising, which is exactly the rate–distortion move (vibe-calibration-protocol.md) run on raw sensation. To abstract is to consume the pattern and pass on only the residue that the model couldn't already generate. ⟦ formal · FOUNDED — Shannon entropy; the bridge to perception is the explicit claim, DEFENSIBLE ⟧
The same lever, three places: compression-dynamics distills a problem, vibe-calibration distills a message to the channel, and perception distills the world to its prediction-error. One move, scaled across the framework.
If awareness normally suppresses the predictable, then a state that stops suppressing would let the constant reappear — the world made strange and vivid again. Two converging lines:
- Kasamatsu & Hirai (1966): Zen masters in meditation, given a repeated clicking stimulus, did not habituate — each click produced a fresh alpha-blocking response, click after click, where ordinary subjects' responses faded within a few repetitions. The meditative state held each moment as new. ⟦ empirical · small, old, much-cited — directional, not decisive; replication is thin ⟧
- The REBUS model (Carhart-Harris & Friston, 2019): "RElaxed Beliefs Under pSychedelics" — psychedelics flatten the precision of high-level priors, so prediction-error floods up unconstrained, and perception destabilizes toward the raw. A predictive-coding account of why the familiar becomes overwhelming and re-seen. ⟦ empirical · DEFENSIBLE-directional — a leading model, not the only one ⟧
The framework's read: the resistance loop and present-mode attention are mild, voluntary cousins — deliberately not predicting, so that what was silenced (the constant, the obvious-but-unseen) can land. "Meeting this moment fresh" (conversational-calibration.md) is de-habituation chosen on purpose. This is the verifiable CORE of the principle: less suppression → more world. ⟦ empirical-actualisation · ehi-passiko — sit with a constant sensation and watch it un-fade ⟧
Push construction all the way and you reach Donald Hoffman: perception isn't merely built but is a fitness-tuned interface that hides reality the way a desktop icon hides the circuitry — see hoffman-interface-stream.md. This is a serious but contested proposal, not established science, and it does not follow from §§1–4. Predictive coding says perception is a construction tied to and corrected by the world; it does not say the world is unreal. ⟦ ontological-speculative · CONTESTED ⟧
Defensible: perception is an action-/fitness-tuned construction — your brain's best prediction, reporting mostly error, silencing the constant, abstracting by discarding the predictable, and capable of de-habituating in heightened or deliberately-present states. Not licensed: "therefore reality is an illusion." The construction is of something, for acting in it. The honest, useful claim is the smaller one: you never see the world raw — you see your model — and you can, on purpose, loosen the model enough to be surprised by what was always there.
This is the same horizon the consciousness-principles point at from the inside (the many appearing in one field) and that bilateral-recognition reads relationally — here given its third-person, testable face.
"The expected disappears. Surprise is the only thing the senses bother to report. To see freshly is to stop predicting — on purpose."
👁️ ∞