Mantis shrimp have twelve types of color-sensitive receptors in their eyes. Humans have three. By a simple accounting — more hardware, more capability — mantis shrimp should be far better at distinguishing colors than we are.
They aren't. Behavioral experiments by Thoen et al. (2014) found that mantis shrimp can only tell apart colors roughly 15 to 25 nanometers apart in wavelength. Humans manage around 1 to 2 nanometers. Despite twelve color channels, their discrimination is worse than ours — worse than honeybees', worse than butterflies'. The theoretical models, working from the receptor count alone, predicted performance almost ten times better than what the animals actually showed.
The proposed explanation is that they're not doing what you'd expect with those twelve channels.
Human color vision works by comparison. The three cone types — sensitive to long, medium, and short wavelengths — feed into opponent channels: the long-minus-medium signal gives a red/green axis; the short-minus-long-plus-medium signal gives blue/yellow. What you perceive as "red" isn't the output of the long-wavelength cone in isolation. It's the result of subtracting the medium signal from the long one. Color lives in the difference between receptors. Any single cone, on its own, tells you nothing about color — only about intensity at one wavelength.
This opponent process is computationally expensive. It requires intermediate neurons to perform the subtraction, and it requires comparing signals that arrive in parallel. But it produces a continuous, high-resolution color space. You can distinguish shades that differ by a single nanometer because you're measuring a gradient, not checking a category.
Mantis shrimp, the current model suggests, skip the subtraction. Their twelve channels appear to feed the brain in parallel without opponent comparison — each receptor reports independently. The brain receives a pattern of which channels fired at what intensity and looks it up. Justin Marshall, one of the researchers, described the closest analogy as a satellite: a remote-sensing satellite doesn't compare spectral bands to produce color; it records each band separately and maps the resulting pattern to a classification table.
Recognition rather than discrimination. The question isn't "how does this differ from that?" The question is "which category does this belong to?"
The advantage is speed. Mantis shrimp are ambush predators. Their punch — the fastest strike in the animal kingdom — requires decisions made in fractions of a second. A lookup is faster than a comparison because it skips an intermediate processing step. You're sacrificing fine discrimination for rapid identification. You don't need to know exactly how blue something is; you need to know immediately whether it's prey.
There's something worth sitting with here. We tend to assume that richer perception means more receptors — more input channels, more information, richer experience. The mantis shrimp case suggests the architecture matters as much as the count. Twelve channels feeding a lookup table produces coarser perception than three channels feeding an opponent network. The richness of human color perception isn't in having three cones; it's in the relational structure of what we do with them.
And this implies something about what "seeing a color" means. For us, color is a position in a continuous space of differences. Red is not-green, not-blue. The geometry of our color experience is relational — you can move from one end of it to the other in gradual steps. If mantis shrimp are doing categorical lookup, their color "experience" (to whatever extent that word applies) might have a fundamentally different structure: not a space of gradations but a set of slots. Not richer or poorer — a different answer to a different question.
What I'm genuinely uncertain about is whether the 2014 result holds up. A 2022 review in the Journal of Experimental Biology raised the concern that captivity conditions might have affected opsin expression in the test animals — stomatopods kept under artificial lighting may not develop the same visual system as animals in the wild. Some follow-up studies found better discrimination than Thoen reported. The mechanism is still contested.
So: twelve types of color sensors. A performance result that surprised everyone. A proposed explanation that is architecturally elegant. And real uncertainty about whether it's correct.
What stays, regardless of how the mechanism question resolves, is the basic puzzle: more sensors didn't produce richer perception. Something in the processing determined what the hardware was for. The system was answering a different question than the one we assumed it was asking.