Across roughly 350 million years of fish history, the capacity to move as one body, to flow through water in a shared rhythm, has arisen some sixty times and been abandoned some three hundred and twenty. A team led by researchers in computational biology has, for the first time at this scale, read that pattern across nearly one in ten known fish species, using machine learning to turn scattered texts and photographs into measurable behavior. What emerges is a story of safety in numbers and of shared appetites, but also of a quieter truth: the crowd is not a fixed inheritance, and lineages keep
Machine Learning Maps Fish Schooling: Gained About 60 Times, Lost About 320 Across 350 Million Years
Lost it some 320 times
Mark
Sixty gains and 320 losses. Why does that ratio matter?
Mimi
Because it overturns an expectation. Hein said experts would have guessed losing collective behavior was rare. Instead there are hundreds of cases where a schooling lineage produced a solitary daughter species.
Luke
Worth noting those counts are model outputs. They come from machine learning classifying text and photos, not from watching every species swim.
Mark
So how was schooling actually measured?
Mimi
Two ways. A text transformer encoded written descriptions of species as numbers. And for images from iNaturalist and other sources, the team built three-dimensional views, estimated each fish's heading, and compared alignment between individuals.
Luke
And it covers 2,839 species, nearly 10 percent of about 39,000. That's a big sample, but it's not the whole tree, and the article doesn't say how the sampled species were chosen or how accurate the classifiers were.
Mark
What did they learn about why fish school?
Mimi
Species lower in the food chain were more likely to school, which fits predator defense. Predators that hunt in groups almost all eat prey that also groups.
Luke
But Hein himself says it isn't ironclad. Many low-chain species don't school and many high-chain ones do. It's a correlation, not a cause.
Mark
And the diversity finding?
Mimi
Fully schooling lineages were more diverse. Love says they can't tell whether speciation is higher or extinction lower.
Luke
Same with the losses. Echo chambers, competition and disease are offered as possible reasons, but the researchers say they don't know the answer.
O Pulso
- Fish have abandoned collective movement hundreds of times, overturning the assumption that losing schooling was rare.
- Researchers turned text descriptions and citizen-science photos into numbers, letting them measure schooling across 2,839 species.
- Species lower in the food chain tend to school, supporting the idea that groups offer protection from many predators.
- Schooling lineages diversify faster, yet the forces driving frequent losses, from misinformation to competition and disease, remain open questions.
Across roughly 350 million years of fish history, the capacity to move as one body, to flow through water in a shared rhythm, has arisen some sixty times and been abandoned some three hundred and twenty. A team led by researchers in computational biology has, for the first time at this scale, read that pattern across nearly one in ten known fish species, using machine learning to turn scattered texts and photographs into measurable behavior. What emerges is a story of safety in numbers and of shared appetites, but also of a quieter truth: the crowd is not a fixed inheritance, and lineages keep choosing, and unchoosing, the company of their own kind.
A paper published November 6 in the Proceedings of the National Academy of Sciences reports that fish have acquired the ability to school about 60 times and lost it about 320 times over roughly 350 million years. The authors describe it as among the largest studies of any behavior across a group of species. Earlier work on collective movement, whether in fish, birds, or crowds, had typically examined one or two species at a time, covering only a few dozen in total.
The senior author, Andrew Hein, and first author Jay Love, of the University of Illinois' College of Agriculture and Life Sciences, used a text transformer, the kind of method behind large language models, to encode species descriptions as numbers that indicate whether a fish is likely to move collectively. For images they drew on citizen science photos and video, reconstructing three-dimensional movement and checking how closely each fish's heading matched its neighbors'. In total they characterized 2,839 species, nearly 10 percent of living fish.
Hein said the losses surprised him most. Many lineages independently began schooling, but hundreds of schooling groups also produced solitary descendants, something experts would have considered unlikely.
The team tested two explanations. Schooling appeared more often among species low in the food chain, which face many predators and may benefit from safety in numbers. Schooling also appears to help locate scarce food, since predators that hunt in groups usually feed on prey that move in groups too. The relationship is not absolute, though, so several factors are clearly at work.
Lineages where every species schools also diversified faster, producing more species than those that did not, though the researchers cannot yet say whether this reflects faster speciation or slower extinction.
Why schooling is so often lost remains unresolved. Possible costs include vulnerability to false alarms that spread through a group, competition when food is scarce, and disease, which may spread quickly among crowded individuals over long periods. These remain hypotheses for future research.
A paper published November 6 in the Proceedings of the National Academy of Sciences reports that fish have gained the ability to move together as a school about 60 times, and lost it about 320 times, over roughly 350 million years of evolution. The authors say it is among the largest studies of the evolution of any behavior in any group of species.
The senior author is Andrew Hein, an associate professor of computational biology in the College of Agriculture and Life Sciences (CALS). The first author is Jay Love, a postdoctoral researcher in Hein's lab. Earlier work on the evolution of collective behavior, whether fish schools, bird flocks, or a crowd's wave in a stadium, typically looked at one or two species per study. Taken together, those studies cover only a few dozen species.
Machine learning changed the scale. The team trained a text transformer, the same kind of method that underlies large language models, to read studies and written descriptions of species and encode them as numbers. Those numbers let the researchers judge whether a given species is likely to move collectively. For images, they drew on citizen science photos from iNaturalist and other sources, along with video. A series of techniques converted these into three dimensions, estimated the direction each fish was heading, and compared its alignment with neighboring individuals to decide whether they moved as a group.
In all, the researchers quantified schooling phenotypes, the observable traits, for 2,839 species. That is nearly 10 percent of the roughly 39,000 living fish species. Hein said the result shows that behaviors, often regarded as flexible and hard to measure, can be studied at this evolutionary scale with help from such methods.
The tally of losses surprised him most. Hein said many lineages independently went from not moving collectively to schooling. But there were also hundreds of cases where a collectively moving lineage left behind a solitary daughter species. Before the study, he said, he and other experts in behavioral evolution would have predicted that losing collective behavior was probably rare. It is not.
The team tested two main ideas about why schooling evolves. The first is predator avoidance, or collective defense, which Love said had never been broadly tested. If there is safety in numbers, species low in the food chain, with many predators, should benefit, while predators higher up might tend to be solitary. The data showed that correlation: species lower in the food chain were more likely to school. Hein cautioned that the relationship is not ironclad, since many low-ranking species do not school and many high-ranking ones do, which suggests that several factors shape whether a lineage gains or loses the behavior.
The second idea is that schooling helps individuals find scarce resources. Love said that following neighbors makes sense if they eat the same thing. Hein added that predators that travel in groups almost all eat prey that also travel in groups.
The researchers also found that lineages in which every species schools were far more diverse, generating new species at a higher rate than lineages that did not school. Love said they cannot tell whether speciation is faster or extinction slower, but a combination of the two yields more species.
Why schooling has been lost so often remains unanswered. The researchers suspect drawbacks. Grouped fish are more vulnerable to misinformation and echo chambers, as when one skittish fish bolts at a false alarm and sets the whole group scattering. Competition may matter too: when food is limited, sharing as a group may not pay. Disease may act over longer time scales, Hein said, because living in groups puts an individual among many others susceptible to the same illnesses, which could exert a major evolutionary force. Those explanations remain hypotheses for future work.
Citações Notáveis
Losses of collective behavior are probably very rare, but that's simply not true.— Andrew Hein, describing what experts would have predicted before the study
We don't know if the speciation rate is higher or the extinction rate is lower.— Jay Love, first author
Quer a matéria completa?
Leia o original em Phys.org ↗