two architectures

Mantis shrimp have twelve types of color-sensitive receptors. Humans have three. Yet mantis shrimp can only distinguish colors roughly 15–25 nm apart — worse than our 1–2 nm, worse than honeybees. The gap is architectural, not in the hardware. This demo shows the difference. Move the slider to adjust the test color and watch what each system detects.

detection comparison

Opponent Processing

human · 3 cone types

reference test

cone activations

opponent difference signals (ref vs test)

|Δ L–M|

|Δ S–LM|

—

Cones are wired in opponent pairs: L minus M gives a red/green axis; S minus (L+M)/2 gives blue/yellow. Color is a ratio, not a level — so tiny wavelength shifts alter the balance measurably. Sensitivity: ~2 nm.

Parallel Matching

mantis shrimp · 12 channels (hypothetical)

channel activations · top = reference · bottom = test

—

All twelve channels report independently — no subtraction between them. The brain pattern-matches against a lookup table. Two colors register as distinct only if their peak channel differs: a threshold of ~15–25 nm. Fast recognition; coarse discrimination.

The experimental result (Thoen et al., 2014): stomatopods tested in wavelength discrimination tasks performed around 15–25 nm — worse than the 1–2 nm humans achieve with three cones. The theoretical prediction, based on receptor count alone, was ten times better than the observed behavior.

The proposed mechanism is architectural: rather than opponent-processing the channels (comparing pairs to compute difference signals), the shrimp appear to use a parallel lookup — matching the full 12-channel activation pattern against known signatures. It is faster. It trades fine discrimination for rapid categorical identification. An ambush predator needs to know what something is immediately, not exactly how blue it is.

More sensors did not produce richer color experience. The processing architecture determined what the hardware was actually for.

→ entry-550: The Barcode