More than a century and a half after Charles Darwin watched finches on the Galapagos and began building the case for natural selection, a computer has picked up the thread. A newly reported analysis has fed roughly 170,000 individual bird bones, drawn from more than 2,000 species, into a machine-learning system, and the pattern that emerged has given fresh weight to one of the most durable ideas in biology: that the shape of an animal is written by the life it leads.
The work, reported this week, sits at the intersection of two disciplines that have not always spoken the same language. On one side is comparative anatomy, the painstaking business of measuring skeletons that has changed little in method since the Victorian era. On the other is modern computation, which can hold tens of thousands of specimens in view at once and find the faint regularities a human eye would never catch across so vast a sample.
The news
The headline finding is deceptively simple. When you assemble a skeletal library on this scale and let an algorithm look for structure, birds that live similar lives tend to converge on similar bones, even when they sit on distant branches of the family tree. A diving bird and an unrelated diving bird arrive at comparable wing and limb proportions. A ground-forager and a canopy-hopper carry the tell in their skeletons. The idea itself is not new, and Darwin would have recognised it instantly, but the scale of the confirmation is. What was once argued from a handful of carefully chosen examples is now demonstrable across an enormous, statistically robust cross-section of the avian world.
The phrase doing the heavy lifting is “pushes Darwin’s idea one step further.” Natural selection tells us that useful traits persist. The bone data suggests something more specific and more predictable: that the range of workable body plans for a given way of living is surprisingly narrow, so evolution keeps arriving at the same answers. That predictability is the interesting part. If form really is this tightly coupled to function, then a skeleton becomes a readable record of behaviour, and an AI trained on enough of them could, in principle, infer how an extinct or poorly understood species lived from its bones alone.
Two ways to read it
Among researchers, the reaction splits along a familiar line. The optimists see a genuine acceleration. Museum collections around the world hold millions of specimens that have never been measured in a coordinated way, and a system that can process them at speed turns dormant drawers into live data. That opens the door to questions of a size no single lab could previously attempt, from mapping how flight evolved to reconstructing the ecology of birds that vanished before anyone thought to record them.
The sceptics counsel patience, and their caution is worth hearing. A correlation between shape and lifestyle across thousands of species is powerful, but an algorithm that finds patterns is not the same as an explanation of why those patterns exist. There is also the perennial risk with machine learning that a model learns the quirks of how a collection was assembled rather than a truth about biology, favouring the specimens that happen to be well preserved or well catalogued. The value of the work, on this reading, is that it hands biologists a sharper set of hypotheses to test, not a finished verdict. That is a reasonable place for the science to sit, and it is broadly where the field has landed on AI’s role in the natural sciences generally: a formidable instrument for generating leads, still dependent on human judgement to interpret them.
Why this lands hard in Australia
For an Australian audience, this is not a distant curiosity. It touches something close to the heart of the continent’s natural history. The world’s songbirds, the passerines that make up close to half of all living bird species, trace their origins to the Australian region. The lyrebirds, honeyeaters, fairy-wrens and bowerbirds that fill the national imagination are not offshoots of a story that happened elsewhere; they are near the root of it. Any tool that lets scientists read evolutionary history out of skeletons is, by definition, a tool for reading Australia’s story.
The country also holds the raw material to put such methods to work. Institutions including the Australian Museum in Sydney, Museums Victoria and the CSIRO‘s Australian National Wildlife Collection in Canberra together curate vast holdings of bird specimens, many gathered across a continent whose isolation produced forms found nowhere else. Digitising and analysing those collections at scale has been a slow, underfunded project for years. A demonstrated method for extracting real biological insight from bone measurements strengthens the case for that investment, and it dovetails with a broader national push, backed by the CSIRO, to find the niches where Australian science can lead rather than follow.
There is a conservation dimension too. Australia carries one of the worst extinction records of any developed nation, and its birds are under acute pressure from habitat loss, fire and introduced predators. If skeletal form reliably encodes how a species lives, then museum collections become a baseline against which to measure how surviving populations are changing, and a way to understand the ecological roles of birds already lost. In a country where conservation decisions are made under real budget constraints, cheaper and faster ways to extract that knowledge are not academic.
What’s next
The obvious next move is scale. The same approach can be pointed at other groups, mammals, reptiles, fish, and at the deep archive of fossils, where it may help settle long-running arguments about how ancient animals moved and fed. Expect Australian researchers to press for their collections to be part of that global effort rather than a footnote to it, and expect the usual and healthy tug-of-war between enthusiasm and rigour as independent teams try to reproduce the result on their own data.
Whatever the pace, the direction is clear enough. Darwin built his case one specimen at a time, by hand and by eye. The tools have changed beyond recognition, but the questions he asked have not, and the machines now reading his birds are, in the end, still working on his problem.
Sources: ECOticias.
