Scientists scanned 15,000 bird skeletons and found evolution doesn’t work the way you’d think

A computer can measure a dead songbird’s skeleton in about 45 seconds. That number is doing more work in this study than almost anything else, because it’s what turned a question researchers have been arguing over for a century into something you can actually run the numbers on.

University of Michigan scientists used artificial intelligence to pull a pattern out of the evolution of Passeriformes, the enormous bird group that covers most songbirds. The birds didn’t evolve at a steady clip. Big changes showed up in fast bursts, and several of those bursts line up with shifts in Earth’s climate.

That idea isn’t new. What’s new is where the evidence came from.

The fossils hinted at it. Modern skeletons proved it.

Evolutionary theory has proposed for decades that life diversifies in stretches of rapid change followed by slower phases. Fossils gave clues. This study found the same signal by measuring bones from modern bird specimens sitting in museum drawers.

“This is really important for evolutionary theory because there’s a long history, going back 100 years, that predicts the emergence of new groups, called evolutionary radiations, is often associated with an explosive burst of diversification. Evolutionary theory predicts that adaptive radiations may account for a large portion of the diversity of life on Earth,” said Jake Berv, lead author of the study and postdoctoral fellow in the U-M School for Environment and Sustainability.

“This could be because of a new ecological opportunity, or it could be because a group dispersed to a new continent, resulting in dramatic accelerations in their rate of evolution. The idea is that, over time, there’s less opportunity as evolution proceeds, and so it slows down, and that this occurs in pulses across time. That’s what theory predicts, and that seems to be what we see in the data as well,” Berv said.

The results came out of AI plus a large statistical model. The findings are published in Nature Ecology & Evolution, with primary support from Schmidt Sciences and the David and Lucile Packard Foundation.

Seven years to teach a model to measure 12 bones

The Michigan team, with senior author Brian Weeks, worked through more than 2,000 species and put together over 170,000 individual skeletal measurements. Nobody does that by hand.

They did it with Skelevision, an AI tool built in Weeks’ lab alongside David Fouhey’s lab at New York University. It photographs specimens, in this case bird skeletons, against a grid that keeps the measurement scale consistent. Over a seven-year collaboration, Weeks and Fouhey trained a model that can accurately measure 12 bones across a single bird’s skeleton.

More than 15,000 museum specimens went through the system. Most came from the U-M Museum of Zoology’s collections. At roughly 45 seconds a specimen, digitizing an entire museum collection stops being a career-length project.

The cooling event that shows up in the bones

Berv also built a new statistical method called bifrost, which let the team analyze each species’ complete skeleton instead of picking apart individual bones one at a time. With it, they estimated how passerine body shapes shifted across roughly 45 million years.

“The whole organism is an integrated, complex morphology, and each of the individual pieces is interrelated to every other part in the body,” Berv said. “The question from the model’s perspective is, ‘What is the sequence of evolutionary changes that needs to happen to explain the variation we can see today?'”

Around 35 million years ago, body-shape evolution went unusually fast. That burst sits on the Eocene-Oligocene transition, a stretch of intense global cooling. The statistical results also flagged a cluster of slowdowns around 15 million years ago, matching up with another major geological event.

“Our findings have definitely shifted my thinking about how the world works,” said Weeks, associate professor of ecosystem science and management at U-M’s School for Environment and Sustainability. “This pattern we found with rare, big increases in the rates of evolution and lots of small decreases in the rate of evolution is really consistent with a pattern where lineages are exploring new ecological space and changing rapidly to take advantage of that opportunity.”

Birds farther from the equator evolve faster

Then the team ran a second check, looking at where the birds in the dataset live now. Geography turned out to predict the average rate of morphological evolution too.

Communities at more extreme latitudes, where seasonal temperatures swing harder, include species that evolve faster than the ones closer to the equator. The same pattern showing up both across millions of years and across present-day geography is what makes it interesting. It points toward environmental variability driving changes in body shape.

“It looks like there’s a connection between latitudinal gradients and rates of morphological evolution that has been underappreciated,” Weeks said. “I hope our findings will inspire a new integration of rates of morphological change into other big areas of interest, things like the very well-known latitudinal gradients in biodiversity.”

A pitch for museum funding, hidden in a bird study

There’s a second argument buried in this paper, and Weeks makes it directly. AI is pulling information out of preserved specimens at a scale that wasn’t realistic before.

“It’s especially clear how important it is to invest in museums when you think about the scale of an analysis like this; it’s so far beyond the scope of what can be done using specimens contributed by an individual collector,” he said. “It’s also fun to imagine what early collectors would make of how we’re using the specimens they collected — I imagine it would blow their minds to learn that a computer has analyzed a photograph of these specimens. It’s just another example of how impossible it is to foresee the full future value of a specimen.”

The uncomfortable part

The researchers say the work may help scientists think about how species respond to the climate change happening now. Berv is careful about the scale mismatch, and he should be.

“Right now, we’re in this moment in human history where there’s dramatic global climate change, and we don’t know what’s going to happen over even a 10-year period, let alone over a 10-million-year period,” Berv said. “To have a chance of understanding the long-term impact of human activity on Earth, we have to study the relationship between events in Earth’s history and evolutionary transitions.”

Additional support came from the Michigan Institute for Data & AI in Society, the Natural Sciences and Engineering Research Council of Canada, and the National Science Foundation.

If you want the one detail to hold onto, it’s this: the birds that changed fastest were the ones living where the weather refused to stay the same. That’s a 45-million-year record saying instability is the engine. We’re currently running the fastest version of that experiment anyone has ever run, and nobody has the results yet.