Fish Don't Vote. They Follow One Influencer at a Time

A rummy-nose tetra swimming among four virtual companions behaved as if it paid attention to only one of them at a time, an international study found, hinting at a simpler rule for how schools hold together ^[2].

The work was published in PRX Life on October 29, 2026, under the title "Decoding Social Integration in Schooling Fish Using Closed-Loop Real-Virtual Interactions" ^[2]. Its authors are Zhen Kang, Ramón Escobedo, Maud Combe, Stéphane Sanchez, Clément Sire and Guy Theraulaz, with affiliations in China, France and Spain ^[2].

The question it tackles is an old one in the study of animal groups. A school of fish turns, speeds up and slows as one body. Does each fish weigh everyone around it, or does something plainer hold the whole thing together?

The problem with watching fish

Film a school and you get trajectories. They show that animals move together, but they do not show which neighbor made a given fish change course. Correlation is everywhere in a school, and cause is hard to pull out of it ^[2].

The team's answer was to build the neighbors. The researchers put one real tetra, Hemigrammus rhodostomus, into a closed-loop virtual-reality system with four model-driven virtual fish. "Closed-loop" means the loop runs both ways: the virtual fish reacted to the real fish in real time, so the real fish was in a live exchange rather than watching a recording ^[2].

Because the virtual fish run on rules the researchers write, the researchers could set how they interact. They varied how many neighbors were influential and combined the experiments with simulations to see how the real fish's responses related to social information and group coordination ^[2].

One neighbor, and it can change

The real fish behaved as if it responded mainly to the single most influential neighbor, according to the paper. That neighbor's identity could change over time, so this is not a fish that picks a leader and sticks with it ^[2].

The authors say this selective interaction was enough to sustain coordinated motion in the group. They also report that coordination grew as social coupling, the strength of the pull between fish, increased. The conclusions draw on analyses of movement, spatial organization, correlations and turning dynamics ^[2].

Theraulaz drew the line between seeing and following when he spoke to New Scientist. "A fish in a group can perceive, simultaneously, many, many individuals, but perception is not equivalent to influence," he said ^[1].

He also suggested a reason a fish might work this way. Focusing on one neighbor may reduce the sensory and cognitive demand on the animal, he said ^[1]. That is a proposed explanation. Nothing in the published material shows the experiment measured that saving.

From observation to test

The virtual setup grew out of earlier work. Theraulaz told New Scientist that his group had studied tetras in observational experiments and had developed a mathematical model of how the fish interact. Virtual reality let them put that model to a test it could fail ^[1].

That matters because a model that fits recorded swimming can still be wrong about why the fish swim that way. When a researcher can control what the neighbors do and watch what the real animal does back, the argument moves from fitting a curve to running an experiment. The paper points to such biohybrid experiments, in which real and artificial animals interact, as a route to the causal mechanisms behind group behavior ^[2].

The wider appeal is economy. If coordinated motion can emerge from each animal tracking one influential neighbor, group behavior may need far less from the individual animal than a rule built on averaging everyone nearby. It also gives a concrete way to think about leadership. In this picture a leader is whichever neighbor happens to carry the most influence for a given fish at a given moment, rather than a fixed individual at the head of the group ^[2].

What the study does not show

The result belongs to a particular setup, and the paper should not be read as saying that every fish in every school follows one leader. The experiment described in the abstract involves a single real fish interacting with virtual ones ^[2].

How many individual fish were tested across all the experiments is not given in the abstract or in Theraulaz's comments to New Scientist. The journal's supplementary material, which runs to a text, 11 figures, five tables and seven movies, may carry those details ^[3].

There are further open questions. Whether the one-neighbor pattern holds in freely swimming groups of real fish, or in other species, has not been tested here. Virtual fish also behave by rules the researchers chose, so they are a cleaner stand-in for a school than the real thing.

What to watch

The researchers have a next experiment in mind. According to New Scientist, they want to test whether virtual fish behaving as if stressed by a predator affect the real fish ^[1]. A threat is the situation in which schools most need to act together, and the one in which a rule of following a single neighbor would be pushed hardest. If the identity of the influential neighbor shifts under threat, that would say a good deal about how information ripples through a group when it counts.