Fish self-control doesn’t seem to run in the family. In ManyFishes 1, a study from 15 research groups, 444 fish from 22 species took the same cylinder detour task, and closely related species didn’t perform alike. That differs from patterns reported in mammals and birds, where relatives tend to score similarly.
The Problem With How We’ve Studied Animal Intelligence
Chimpanzees plan ahead. Crows solve puzzles. Parrots wait for a better reward. For a century, these have been the animals that cognition science built its theories around – warm-blooded, familiar, easy to test in the lab. Fish barely entered the conversation, dismissed as reflexive rather than reflective.
That narrow focus created what researchers call a “taxonomic bias” – a picture of animal intelligence drawn largely from one branch of the vertebrate tree. A collaboration called ManyFishes 1, the largest coordinated study of fish cognition to date, set out to test whether that picture actually holds up once you look past mammals and birds. The results complicate that picture, at least for one particular aspect of cognition: inhibitory control.
What Is “Big-Team Science,” and Why Does It Matter Here?
Cognition studies are normally run by a single lab testing a handful of animals, using methods that shift slightly from one research group to the next. That makes cross-species comparison shaky at best. ManyFishes solved this the way large physics or genomics collaborations do: get everyone to run the exact same protocol. Fifteen research groups across 10 countries did just that, together testing 444 fish spanning 22 species and 19 genera – the largest standardized dataset ever built for fish cognition.
The Test: A Fish-Sized Puzzle Box
The core measurement tool was something called the cylinder detour task, which is a classic way to measure inhibitory control – essentially, an animal’s ability to stop itself from acting on impulse.
Here’s how it works: a fish sees a piece of food sitting inside a clear, see-through cylinder that’s open at both ends. The obvious, instinctive move is to swim straight at the food – but since the cylinder wall is in the way, that just results in bumping into glass. The “smart” move is to resist that urge, swim around to one of the open ends, and enter that way instead. Succeeding requires the fish to override its own impulse, which is exactly what “inhibitory control” means.
To get fish ready for this test without cheating the results, researchers ran them through four careful steps:
- Plate training – Fish first learn to eat from a marked plate (a green dot on a white base), so they know where food will appear. They needed to succeed in at least 5 out of 6 tries before moving on.
- Cylinder familiarization – The clear cylinder is placed in the tank for 48 hours with no barriers, so the fish gets used to it as just another object in its environment, not something scary.
- Forced trial – The cylinder is placed perpendicular to the doors, with clear panels blocking the space on either side, turning it into the only path through to the food. This confirms the fish understands the tunnel can be entered and exited safely before facing the real detour test.
- The actual test – Now the cylinder is turned sideways. The fish can see the food through the clear wall but has to detour around the side to reach it.
A “success” meant the fish got the food without ever touching the cylinder. Touching the walls counted as a failure. If the fish did nothing at all – no touching, no eating – within five minutes, that was logged separately as a “null response,” since it likely reflected low motivation rather than poor impulse control.
One detail worth noting: the cylinder wasn’t a single fixed size. Since the study spanned species ranging from small guppies to largemouth bass, each cylinder was scaled to the species being tested – built to be at least as long as, and as tall as, the largest individual of that species in the sample, with wall thickness of about half a centimeter. Placement mattered too: the cylinder sat at least one body length from the tank wall, with five to ten body lengths of swimming distance from the starting doors. The goal was to make sure a fish’s performance reflected its self-control, not its ability to physically maneuver a tube built for a different-sized animal.
One design choice sets this apart from similar mammal studies: researchers skipped the step where animals first practice with an opaque cylinder. That step normally lets an animal learn the physical detour habit before ever facing visible temptation. Cutting it out made the fish version stricter – arguably a purer test of raw self-control than the one used on primates and birds. It also came at a cost. Fish scores here landed lower overall than the mammal and bird scores from the comparable MacLean study. A separate guppy study using the easier, opaque-cylinder-first version found guppies performing on par with primates and birds. So at least part of the fish/mammal gap in this study may be an artifact of a harder test, not a real difference in self-control.
How Does a Fish “See” a Clear Wall?
But if the cylinder is transparent, how does a fish know it’s even there?
A few things help. Underwater, clear acrylic or glass isn’t perfectly invisible – light refracts and reflects off its edges, creating faint outlines and glare that give away its shape. The cylinder also sits on a solid white base plate anchored to the tank floor, which acts as a visible marker of where the object is, even if the cylinder walls themselves are hard to see.
Just as important is the run-up to the actual test. By the time a fish faces the real detour trial, it’s already spent 48 hours freely exploring the cylinder and has physically swum in and out of it during the forced-tunnel step. In other words, it has already built a working sense of the cylinder’s shape and boundaries before ever being tested.
This matters for interpreting the results. The test isn’t measuring whether a fish can perceive a physical barrier – it can. What’s being measured is whether the fish can override its instinct to swim straight at visible food, and instead act on what it already knows: that reaching the food requires going around the side.
Family Tree Doesn’t Predict Fish Performance
Here’s the finding that got researchers’ attention. In mammals and birds, closely related species tend to score similarly on cognitive tasks – a pattern called a phylogenetic signal, measured with a statistic called Pagel’s lambda. A value near 1 means ancestry strongly predicts performance; near 0 means it predicts almost nothing.
For the fish in this study, lambda came out statistically indistinguishable from zero. Knowing how closely two fish species were related told you nothing about whether they’d perform alike.
That’s a real departure from patterns in warm-blooded animals – but it comes with an asterisk. The species tested here were an opportunistic sample, not a systematic one, and coverage across the fish family tree was uneven. A weak signal in data like that can mean there’s no real pattern, or it can mean the pattern was too hard to detect with the species on hand. The finding is genuine and striking; it’s also not the final word.
This is a genuinely unusual finding. It suggests that, unlike in mammals and birds, inhibitory control in fish isn’t something locked in by deep evolutionary history. Instead, it looks like a trait that can shift rapidly depending on what a particular species needs to survive in its particular environment.
Performance Was All Over the Map
Beyond the phylogenetic result, the study found huge variation at every level:
- Between species: Some species did dramatically better than others. Zebrafish and the bluestreak cleaner wrasse (a fish known for its specialized cleaning behavior on reef fish) were among the higher performers, with roughly half their trials ending in success. At the other extreme, species like the largemouth bass, gilthead seabream, and white-banded triggerfish performed close to zero – essentially failing almost every trial.
- Between individuals of the same species: Even within a single species, some individuals were far more consistent than others, with certain species showing much wider gaps between their best and worst performers.
- In how fish learned over time: Researchers also tracked whether fish improved with practice across repeated trials. Some species got noticeably better as trials went on, suggesting they were actively learning the “trick” of the task. Others showed no improvement at all, hovering at the same performance level (sometimes at zero) from the first trial to the last. This suggests different species may be relying on entirely different mental strategies – or none at all – to solve the same problem.
Same Species, Same Protocol, Different Results
The clearest illustration of how much environment matters came from guppies. Tested at three universities – Macquarie in Australia, Stockholm in Sweden, Ferrara in Italy – using an identical protocol, they didn’t perform identically at all: Macquarie’s guppies outperformed both other sites, and Stockholm’s outperformed Ferrara’s. Since the procedure was controlled for, the gap likely comes down to something more mundane – housing, handling routines, water conditions. Standardizing the protocol, it turns out, doesn’t standardize the fish.
Why Might Fish Cognition Work Differently?
One explanation researchers point to: fish brains don’t stop developing. Mammal and bird brains largely plateau in adulthood; many fish keep generating new neurons throughout life, a process called continuous neurogenesis, with brain size continuing to scale up alongside body size well past the point where a mammal’s would level off. That ongoing plasticity could let fish continually retune their neural wiring – and their behavior – to match their circumstances, rather than locking into a fixed cognitive style early on.
It’s a compelling story, but the researchers don’t treat it as the only one. An equally plausible read is that inhibitory control is simply a highly plastic trait in fish generally – one that would produce this exact pattern, high variation paired with weak ancestry effects, regardless of what’s happening at the level of individual neurons.
Where This Leaves Us
ManyFishes 1 proves two things. The practical one: large-scale, standardized cognitive testing works even in animals that are notoriously hard to compare across labs. The more interesting one: a rule that looked universal – that intelligence runs in the family – turns out not to hold for roughly half of all living vertebrate species. The researchers’ own next step says as much about their confidence as any of their statistics: test more species, and test the same species in more than one lab, before deciding how far this rule really breaks down.
Leave a Reply