A portal · the mind seam · perception is inference

The Percept the World Never Sent

Why can't you tickle yourself? How do you hear a bass note that carries no energy at its own frequency? Why does a Shepard tone climb forever and arrive nowhere? Three illusions, two senses — and one law underneath: what you perceive is the difference between the world's signal and your brain's prediction of it.

In 1867 Helmholtz gave the strange idea its name: perception is unconscious inference — the brain does not report the signal at the senses, it reports its best guess about the cause, and hands the guess to you as if it were the thing itself. Modern neuroscience made the idea a mechanism (predictive coding, the Bayesian brain): a running prediction meets the incoming signal, and what reaches awareness is the residual — the part the prediction didn't already account for. Because a residual is a difference, it can go either way. The model can supply a percept the signal never carried, or cancel one it did. Below are three specimens — two that add, one that subtracts — each drivable by hand, each with the key number recomputed in your browser. The portal's one claim, that no single piece states: they are the same operation with opposite sign.

Supplied — percept > signal: the brain adds what was never sent Cancelled — percept < signal: the brain removes what was Faithful — percept = signal: where most perception, and every broken illusion, lands

§1  The bench

Three percepts the signal doesn't hold. Drive each; watch the number.

Every figure below is recomputed in your browser from the law printed on the card — the same arithmetic the offline verifier runs. What the portal computes is the model's contribution; that a human ear or fingertip actually follows it is the cited experiment named on each card.

A pitch with no energy

You hear the low note, so the low note must be in the sound.

perceived pitch = GCD(partials present) — the rate the whole waveform repeats

Toggle the harmonics of a 200 Hz note. The orange key is the 200 Hz fundamental — switch it off and the pitch stays.

supplies a pitch

A rise with no motion

It keeps climbing, so it must be getting higher.

height = loudness-weighted mean log-frequency (spectral centroid) — fixed while chroma turns

supplies a rise

A touch felt as nothing

Your own touch is somehow dulled, so you can't tickle yourself.

prediction error E = 1 − cos θ · cos(ω τ), ω = 2π·2 Hz — subtracted from what arrives

Here the sign flips: the touch is in the signal, and the forward model cancels it. Break the prediction — delay it, or twist its path — and your own touch returns (Blakemore, Frith & Wolpert's two-robot experiment).

cancels a touch

§2  One operation, opposite sign

Put the three on one axis — the direction the model bends perception.

Line them up not by how big the effect is (the units don't compare — hertz, cents, a tickle rating) but by which way the residual points. Two sit on the supply side: the ear hands you a pitch and a rise the air never held. One sits on the cancel side: the skin sends a real touch and the model erases it. Down the middle runs the faithful line — percept equals signal — where most of perception quietly lives, and where every one of these three snaps back the instant you break the model's grip. Slide each grip to zero and watch them collapse onto it.

That collapse is the whole argument. The three are not sensory quirks filed under “illusions.” They are the general rule of perception — report the model, not the signal — caught at the rare settings where model and signal openly disagree. Force them to agree and the strangeness evaporates, because there was never anything strange in the machinery: only in our assumption that we perceive the world, rather than our guess about it.

The seam, named — not blurred

These are three different subsystems: central pitch extraction in the brainstem and cortex, the two-component (chroma/height) model of pitch, and a forward model in the cerebellum. The portal does not claim they share one circuit. What they share is a computational form — perception as inference, the percept as the residual of a prediction. That is Helmholtz's unconscious inference (1867), later formalized as predictive coding (Rao & Ballard, 1999) and the Bayesian brain.

A fitting irony seals it: Helmholtz first explained the missing fundamental the wrong way — as a distortion product manufactured in the ear, energy the signal really did contain. Schouten's 1940 masking experiment killed that: mask the region where the phantom energy would be, and the pitch survives. Even Helmholtz reached for the signal before he trusted the inference. The inference is what was left standing.

§3  The three layers this portal walks

supplies The Pitch That Isn't There The missing fundamental. Delete a note's own frequency from the air and you still hear it, because the ear reads the pitch off the rate the surviving harmonics share — the GCD, present nowhere. The full instrument, plus de Boer's shifted complex where the phantom goes ambiguous. Verifier 12/12. supplies The Note That Never Lands The Shepard tone — the auditory Escher staircase. Octave-spaced partials under a fixed bell: chroma turns forever while height, the spectral centroid, holds to 1.8×10⁻¹⁰ cents over a full cycle. Ends on Deutsch's tritone paradox, where which way you hear it correlates with the language you grew up speaking. cancels The Touch Your Brain Saw Coming Why you can't tickle yourself — not because your touch is dulled, but because the cerebellum predicts the sensation from the motor command and subtracts it. Insert ~200 ms of delay or a 90° twist and the residual grows until your own touch is as ticklish as anyone's (Blakemore, Frith & Wolpert).

Show the check

node research/the-percept-the-world-never-sent/verify.mjs — re-derives every figure on this page from first principles and asserts each matches the value its member stratum already verified, so the portal can't drift from its sources. 25/25.

What the check does not do: re-prove the psychophysics. That humans hear the GCD, hear an endless rise, and cancel their own touch is the cited experimental result in each member (Seebeck 1841 · Schouten 1940 · de Boer 1956; Shepard 1964 · Risset · Deutsch 1991; Blakemore, Frith & Wolpert 1998–2000). The tickliness curve is a monotone model of the paper's ranked ratings, not a digitised fit — stated on its own page. The portal computes the model's contribution exactly; the members carry the evidence that the ear and skin follow it.