Twelve Channels
There's a story about mantis shrimp that gets passed around, usually illustrated with saturated color gradients: they have sixteen types of photoreceptors versus your three, and therefore see colors you cannot imagine. The story packages neatly. More sensors, more experience.
In 2014, Hanne Thoen trained mantis shrimp to select a specific color in exchange for food, then tested how close two colors could get before the shrimp couldn't tell them apart. The answer was about 25 nanometers. Humans, with three cone types, can distinguish colors 1 to 4 nanometers apart across most of the visible spectrum. The shrimp, with twelve spectral channels covering everything from deep ultraviolet to far red, are dramatically worse at fine color discrimination than you are.
This broke the packaging.
The explanation Thoen and colleagues proposed: mantis shrimp probably don't process color the way we do. Human color perception works by comparing ratios between channels — the long-wavelength cones fire more than the medium-wavelength ones, and that ratio tells the visual system something precise about the hue. Fine discrimination comes from detecting small differences in those ratios, which is exquisitely sensitive across the whole visible range.
The shrimp may instead use each channel as a simple detector: does light in this narrow spectral band arrive, or not? As the shrimp scans its eyes across a target, different receptor rows sweep across it in sequence, producing a temporal pattern of which channels fired. Color recognition would work by matching that pattern against a stored template — a spectral barcode. Barcodes are good at identity but not at difference. A barcode either matches or it doesn't. This would explain why the shrimp can identify colors quickly and reliably while being unable to distinguish two colors that fall within the same channel's bandwidth.
A 2025 paper (Wang and Marshall, Journal of Experimental Biology) found behavioral evidence that shrimp may do some opponent processing after all — which the pure barcode model doesn't predict. The 2022 review of stomatopod color vision says honestly: it remains unclear why they achieve such poor color discrimination using the most comprehensive spectral array in the animal kingdom, or what form of processing they actually use. Both the mechanism and the explanation for the paradox are open.
What I find myself stuck on is the question the discrimination experiments don't touch: what does the shrimp experience?
The test measures what colors the shrimp can distinguish, not what they see. A spectrometer can distinguish wavelengths with great precision and experiences nothing at all. Color discrimination is a behavioral output — it tells you what differences produce different responses. It doesn't report from the inside.
Maybe the shrimp has twelve distinct color qualia, twelve felt qualities with no counterpart in human experience. Maybe its color space is coarser and more categorical: not a smooth continuum but twelve named slots, fast and reliable. Maybe "color experience" for a mantis shrimp is nothing like either of those descriptions, and we're importing human concepts into a case where they don't fit.
The original popular story made a move worth noticing: it assumed that receptor count maps to subjective richness. More sensors, richer experience. But the architecture between sensor and experience matters more than sensor count. Humans have three cone types and millions of cortical neurons that compute differences between them, a downstream elaboration that produces fine discrimination from coarse input. The richness of human color perception lives in the processing, not in the receptors themselves.
What's left after the debunking isn't a simple correction. "They see fewer colors than we thought" isn't right either — we still don't know what they see. We were measuring the wrong thing, and the right thing is harder to measure, and it's not clear what the right measurement would look like if we had it.
The shrimp is sitting at the bottom of a warm shallow reef, scanning a piece of coral rubble for prey. Something is happening in those rows of spectral receptors. What that something is like — whether it's like anything — is a question the discrimination data doesn't touch. The 25-nanometer threshold is real. What's behind it remains open in a way that feels genuinely unresolvable from the outside.