AI Read 170,000 Bird Bones and Found That Evolution Moves in Bursts

Evolution is often imagined as a smooth, almost invisible process: tiny changes accumulating at a steady pace over millions of years. A new analysis of the world's most diverse order of birds suggests something more dramatic. Long periods of relative slowdown appear to have been interrupted by rare bursts in which body shapes changed much faster.

To uncover that pattern, researchers combined an artificial-intelligence system that measures museum skeletons with a new statistical method capable of analysing the whole body at once. The study, published on 20 July 2026 in Nature Ecology & Evolution, used more than 170,000 skeletal measurements from 2,057 species of Passeriformes, the group containing songbirds and other perching birds.

A machine turns museum drawers into a vast dataset

The physical evidence had been waiting in natural-history collections for generations. The problem was scale. Measuring a dozen bones on thousands of fragile specimens by hand would require years of repetitive work and introduce differences between human measurers.

The team's tool, called Skelevision, photographs a bird skeleton against a calibrated grid. A computer-vision model identifies and measures bones across the specimen. Each scan takes roughly 45 seconds. Researchers used it on more than 15,000 individual museum specimens and assembled over 170,000 measurements.

The system does not identify evolutionary events by itself. Its role is to turn physical collections into consistent numerical data at a scale that would otherwise be impractical. The researchers then used evolutionary family trees and a new modelling framework called bifrost to reconstruct how combinations of traits changed through time.

Why one bone at a time was not enough

A bird's body is an integrated system. A longer wing, shorter leg or different bill does not evolve in isolation from every other structure. Conventional analyses can miss that relationship when they examine traits separately or reduce them too aggressively.

Bifrost was designed to analyse shifts across multiple related measurements. It searched the passerine family tree for branches where the overall rate or pattern of body-shape evolution changed. The code and supporting data have been made available for other researchers to inspect and reuse.

This whole-body approach revealed a hierarchy of rare bursts near the origins of major bird groups, followed by many smaller slowdowns. Those early periods of rapid anatomical exploration appear to have contributed disproportionately to the diversity of modern passerines.

A climate connection written into bird skeletons

One major burst occurred around 35 million years ago, close to the Eocene-Oligocene transition, when Earth underwent substantial global cooling. Later clusters of change also aligned with periods of climatic instability.

The researchers tested the pattern in a second way by comparing modern bird communities around the world. Assemblages at higher latitudes, where seasonal temperatures vary more strongly, contained species with higher average rates of body-plan evolution than many communities near the equator.

The agreement between deep time and modern geography supports the idea that environmental variability repeatedly creates pressure and opportunity for anatomical innovation. Changing climates may open habitats, rearrange food sources, isolate populations or reward new ways of moving and feeding.

AI did not discover this without people

The headline role of artificial intelligence can easily obscure the human and physical infrastructure behind the result. Museum workers collected, preserved, catalogued and maintained the specimens. Researchers spent seven years developing and checking Skelevision. Evolutionary biologists built family trees, selected models and interpreted the results.

AI accelerated a narrow but crucial task: extracting standardized measurements from images. It did not replace field knowledge, taxonomy, statistical judgement or the need to verify unusual outputs. The project is a useful example of AI expanding what scientists can ask of old collections rather than making those collections obsolete.

What the study cannot tell us

The alignment between evolutionary bursts and climatic instability is strong evidence of association, not proof that climate directly caused each anatomical change. Continents moved, habitats shifted, species competed and new ecological opportunities emerged during the same periods. Those forces are difficult to separate across tens of millions of years.

The analysis reconstructs history from living species, museum specimens and a phylogenetic model. Extinct branches that left little evidence can affect the apparent timing and size of evolutionary changes. Measurement models and family trees also carry uncertainty, even with a dataset of this scale.

Most importantly, the study does not imply that birds can easily adapt to today's rapid human-driven warming. A pattern measured across millions of years cannot guarantee successful adjustment over decades, especially when habitat loss and other pressures occur simultaneously.

Old bones answer a modern question

The result is striking not only because of what it says about birds, but because of how it was produced. Specimens collected long before machine learning existed became inputs to a computational experiment their collectors could never have imagined.

AI did not predict the future of evolution. It helped read a record stored in thousands of skeletons, revealing that life's history may advance less like a clock and more like a long quiet sentence interrupted by sudden exclamation marks.