The Interpolation

In 1912, Max Wertheimer sat in a train and watched telegraph poles flicker past the window. The movement was an illusion — each pole was a static object perceived for an instant — but the flicker had structure. He took it to the laboratory. Two lights, flashing in alternation with an interval between thirty and two hundred milliseconds, were perceived as a single light moving between positions. He called it the phi phenomenon. There was no object in transit. There was no light at any intermediate point. The motion was manufactured wholesale by the visual system from two discrete stimuli and a gap.

Cinema runs on this. Twenty-four frames per second, each one a still photograph projected for a fraction of a second with a dark interval between. The projector provides images. The viewer provides motion. Early filmmakers discovered empirically what Wertheimer formalized: the brain does not need continuous input to perceive continuous events. It needs samples at a sufficient rate, and it will build the rest.

In 1972, Frank Geldard and Carl Sherrick at Princeton placed mechanical tappers on subjects' skin — three taps on the wrist, then three on the elbow, delivered in rapid succession. Subjects reported feeling six taps traveling up the forearm, evenly spaced along a path where no stimulus had occurred. They called it the cutaneous rabbit: the sensation hopped across skin that was never touched.

The illusion is robust. It persists when subjects know the intermediate taps are not real. It scales with the spatial separation of the contact points — wider gaps produce longer perceived trajectories with more interpolated taps. And it works in the opposite direction from most illusions: the brain is not failing to detect something present. It is detecting something absent. The two real endpoints are enough. The system constructs the path between them because a trajectory is the most probable explanation for two sequential contacts at different locations.

The percept is not wrong. Given the information available — two points, a time interval — smooth traversal is the best inference. The error is not in the conclusion but in the assumption that perception reports what happened rather than what most likely happened.

Lossy audio compression — MP3, AAC, Opus — works by identifying what you will not hear. The psychoacoustic model calculates which frequency components are masked by louder simultaneous sounds (simultaneous masking) or by sounds that occurred just before them (temporal masking, exploiting the roughly two-hundred-millisecond forward masking window). These components are discarded. The bit budget is spent only on what survives the auditory system's own filtering.

The result is a signal with physical gaps — frequency bands deleted, quiet passages quantized to nothing — that sounds continuous to the listener. The compression does not approximate the original. It approximates the listener's perception of the original, which is a smaller target. The codec and the auditory system are doing the same work: reducing bandwidth by discarding information below the threshold of noticeability. The codec simply does it first.

A 128 kbps MP3 discards roughly ninety percent of the original data. The listener hears continuity because their auditory system would have constructed the same continuity from the uncompressed signal. The codec does not create the illusion. It exploits one that was already running.

Daniel Simons and Christopher Chabris published their gorilla experiment in 1999: subjects watching a video of people passing a basketball were asked to count the passes. A person in a gorilla suit walked through the scene, paused in the center, beat their chest, and walked off. Roughly half the subjects did not see it.

The finding is usually cited as a failure of attention. It is equally a success of interpolation. The subjects perceived a continuous, coherent scene — people passing a ball — and the gorilla was not part of that scene's causal structure. The visual system did not miss the gorilla because it was invisible. It missed it because including it would have disrupted the continuity that the system was actively constructing. The coherent scene was the product. The gorilla was not masked by attention — it was edited out by the process that builds stable percepts from unstable input.

Simons and Levin demonstrated the same principle more starkly in 1998: during a face-to-face conversation, a brief visual interruption (two people carrying a door between the conversants) allowed the experimenters to replace one conversant with a different person. Fifty percent of subjects continued the conversation without noticing. The person had changed. The continuity had not.

A refrigerator hums. An air conditioner drones. Traffic on a distant highway produces a continuous wash of broadband noise. In each case, the sound is physically present, mechanically transducing through the cochlea, generating neural signals. And in each case, the brain stops forwarding it to conscious awareness through auditory habituation — the progressive reduction of response to an unchanging stimulus. The sound is not gone. The percept is.

If the refrigerator stops, the silence is perceived as an event. Something happened — the absence of a stimulus that was being actively ignored. The system notices the removal of something it had stopped reporting. This means the suppressed signal was still being monitored at some level below awareness, maintained as a background model against which change could be detected. The continuity of the hum was erased from experience. The monitoring of the hum was not.

In every case, the experience of continuity is constructed, not received. The phi phenomenon builds motion from stills. The cutaneous rabbit builds trajectory from endpoints. Lossy compression builds fullness from the subset that matters. Inattentional blindness builds coherence by excluding what doesn't fit. Habituation builds silence from ongoing sound.

The construction is so effective that its interruption feels like an error in the world rather than a revelation about the self. The stopped clock, the noticed silence, the spotted gorilla — each one feels like something changed out there. What changed is that the interpolation briefly failed, and the gaps it normally covers became visible.

Continuity is not the default state of experience. It is the product of a system that works so well that its output is mistaken for its input.