Your eye has three types of cone cells. The actual light entering your eye can vary across hundreds of wavelengths simultaneously. Those three cone types each measure total activation in a broad spectral band — long, medium, short — and that's what gets sent upward. Three numbers from an infinite-dimensional input. Most of the information is discarded.
Two physically different spectra that produce the same three numbers are called metamers. They are perceptually identical. A monitor exploits this: it doesn't reproduce the spectrum of a sunset, it finds a metamer for it — three LEDs at the right intensities to trigger the same cone responses the actual sunset would. The screen and the sky produce completely different light. They produce the same perception. That's not a trick; that's how the system works.
What you experience as a single color — "that orange" — is an equivalence class over an infinite family of possible spectra. Everything in the class looks identical to you. The variation within the class is physically real and completely invisible. The category is reliable and consistent. The thing the category tracks is underdetermined by the category itself.
This is the clearest case, because the math is transparent. Three detectors, infinite input dimensions: anyone can work out what that compression means. But the same structure appears everywhere perception happens. Pain collapses a sprawling population of afferent signals into a single felt intensity and location. Pitch collapses the continuous variation of basilar membrane activation into a tone you can name. Temperature collapses the statistical distribution of molecular kinetic energy across your skin receptors into "warm" or "cold." Every sensation is a metamer for a large equivalence class of physical states. The class is defined by the detector. What the detector can't distinguish, you can't distinguish.
The obvious assumption from this is: more detectors, richer perception. If three cones give you the full human color palette, sixteen should give you an extraordinary one. Mantis shrimp have sixteen types of photoreceptors. For decades this was cited as evidence that mantis shrimp must have spectacular color vision — seeing a world of chromatic richness far beyond ours.
A 2014 paper by Hanna Thoen and colleagues tested this directly and found the opposite. When mantis shrimp were trained to discriminate between wavelengths and then tested on similar wavelengths, they were substantially worse than humans. Humans can distinguish wavelengths 1–2nm apart. The mantis shrimp needed differences of around 25nm.
What the sixteen channels seem to be doing is something different from what three channels do. Rather than comparing ratios across channels — which is how human color constancy works — the mantis shrimp appears to use a temporal scanning strategy, sweeping photoreceptors across the target and recognizing spectral signatures the way you'd read a bar code. Not a continuous gradient of fine distinctions, but a fast classification into discrete labeled bins. The sixteen channels work like sixteen filters, each tuned to a different spectral category, and the classification is binary: match or no match.
This would be faster — useful for the mantis shrimp's real problem, which is recognizing prey species and rivals in milliseconds rather than pondering subtle distinctions between similar blues. Finer discrimination might actually slow that process down. The sixteen channels gave it a richer input but a coarser output — not because the processing failed, but because fine discrimination wasn't the goal.
More detectors doesn't automatically produce richer perception. It produces different processing possibilities, and the processing that evolves is the one that solves the actual problem. The sixteen channels in a mantis shrimp are not a human color system with thirteen extra slots. They're a different architecture that happens to use light.
What I keep finding is that the relationship between what comes in and what is perceived has this structure: the perceived category is consistent and real, and the physical substrate of that category is underdetermined and often variable. The category is reliable as a signal. What the signal is about is murkier than it appears. Three cones reliably track the color. The color is not the spectrum. The spectrum is what was actually there.