The Question

A curator opening a wide metal drawer in a museum storeroom lined with labelled fossil bones under soft light

Behind the public halls of the world's great natural history museums are corridors most visitors never see. Miles of steel cabinets, floor to ceiling, each drawer packed with bones, shells, teeth, and rock. Some carry labels written in fountain pen a century ago. Many carry no label at all beyond a field number and a place name. The specimen has waited in the dark, in some cases, since before your grandparents were born.

Here is the fact that stops paleontologists cold. The Smithsonian's National Museum of Natural History holds more than 40 million fossil specimens, yet its online catalogue lists only around 450,000 specimen records — most of the collection has not been digitised, and much of it has never been studied in detail. Similar backlogs exist in the great collections of London, Paris, Berlin and Beijing. The obvious question follows: if the drawers are already full of things no one has properly looked at, where are the great discoveries of the next decade actually hiding? Increasingly, the answer is not a remote badlands cliff. It is a cabinet in a basement.

What the Evidence Shows

This is not a hypothetical. New species are pulled from old collections all the time. Some of the most famous dinosaurs in the world were misfiled for decades. In 2015, a giant statistical study of long-necked dinosaurs concluded that Brontosaurus — long dismissed as the same animal as Apatosaurus — was a distinct beast after all, a finding drawn largely from bones already sitting in collections. In 2025, researchers named a new sauropod, Athenar bermani, from a partial skull at the Carnegie Museum of Natural History that had been collected in 1913 and long assigned to Diplodocus. In 2007, a single vertebra in London's Natural History Museum, dug up in the 1890s, was named as a new sauropod, Xenoposeidon.

The reason the backlog exists is a money mismatch. Digging fossils out of the ground is glamorous — it wins grants, headlines, and documentary crews. The unglamorous work that follows — cleaning, cataloguing, comparing, and formally naming a specimen — is slow, painstaking, and chronically underfunded. A single fossil can take a specialist weeks to describe. With far more specimens than specialists, clearing the backlog by hand would take a very long time.

"One specimen, CM 26552, is often overlooked in favor of more complete skulls, but bears an outsized historical importance as this specimen formed the basis for the modern description of braincase anatomy for Diplodocus as a genus."

— Dr John Whitlock, Mount Aloysius College and Carnegie Museum of Natural History, on the skull that became Athenar bermani, 2025

What changes the math is machine vision — software that can recognise shapes in an image the way a trained eye does. Digitisation projects are racing to photograph and scan the world's collections: the iDigBio network in the United States has passed 100 million natural history specimen records, and London's Natural History Museum, which holds about 80 million specimens, is digitising its collections too. Early AI results are promising but limited: in a 2026 study from the Netherlands, neural networks trained on standardised museum images identified fossils correctly about 85% of the time, but only about 65% when trained on photos from citizen scientists. Once a fossil is a high-resolution image or a 3D scan, an algorithm can compare it against thousands of others in seconds — flagging the odd one out, the specimen whose proportions do not match anything named. CT scanning goes further, letting researchers see the fossil still locked inside its rock without ever cracking it open.

The rarest thing in a natural history museum is not the fossil. It is the person with time to look at it.

Why This Is Happening

Digitisation has quietly reached a tipping point. For two decades, museums photographed and scanned specimens with no single tool able to make sense of the flood. Now that the images exist in the tens of millions, they become training material — a machine can learn what a normal tooth looks like precisely because it has seen a hundred thousand of them. The backlog that made manual study impossible is exactly what makes automated study powerful.

The bottleneck was never the ground; it was the bench. Field expeditions have been finding more than museums can process for over a century. Every dig adds to a queue that already stretches beyond a human lifetime. Software that can triage that queue — sorting the routine from the remarkable — attacks the real constraint. It does not replace the paleontologist; it points them at the drawer worth opening.

Citizen science multiplies the workforce for free. Platforms that let volunteers transcribe old labels, measure bones from photos, and flag oddities have already processed millions of records. Combined with machine sorting, a global crowd plus an algorithm can pre-screen a collection that a handful of curators could never finish. The discovery is then confirmed by an expert — but the expert is arriving at the right shelf years sooner.


What Could Happen

Drawer discoveries overtake field discoveries by 2035 Less likely

AI-assisted cataloguing sweeps the major collections, and a wave of new species — dinosaurs, ancient mammals, marine reptiles — is named from specimens dug up decades ago. Museums reorganise around "collections mining," and the phrase "found in the field" starts to mean the drawer, not the desert. Field expeditions continue, but the cheapest, fastest discoveries come from indoors.

A hybrid model where software points and humans confirm Most likely

Algorithms flag anomalies but generate too many false alarms to trust alone, so the real gain is speed rather than volume. Curators still make every formal identification, but they reach the important specimens far faster. Discoveries rise sharply without ever fully shifting the balance away from traditional fieldwork.

Funding and access stall the revolution Possible

Digitisation is expensive, and cash-strapped museums cannot scan fast enough. Collections stay dark, access rules keep data siloed between institutions, and the promised wave never fully arrives. The technology works, but the drawers stay closed for want of budget — a delay, not a dead end.

Our Assessment
We assign 15% probability — unlikely that by 2035, AI-assisted cataloguing will yield more major fossil discoveries from inside museums than from new field expeditions. We have lowered this forecast from 83%: museum rediscoveries are real and frequent, but they still come mainly from experts re-examining specimens, most fossil collections are not yet digitised, and AI tools for fossils are at an early research stage. The drawers will keep yielding surprises — and software will increasingly help curators find them — but fieldwork is likely to remain the main source of major new finds through 2035.

What Can We Do

A volunteer at a laptop transcribing handwritten labels from photographs of fossil specimens on screen

You do not need a doctorate to help open the drawers. Some of this work is being done, right now, by ordinary people at home — and the discoveries belong to everyone.

Join a digitisation project as a volunteer. Platforms run by museums let anyone transcribe old specimen labels, measure bones from photographs, or classify images. It is genuine science: the data you enter feeds the same databases the algorithms learn from. An afternoon of transcribing might be the reason a new species surfaces two years earlier.

Support museum collections funding, not just exhibits. When you donate or vote on cultural budgets, remember that the flashy public gallery is a fraction of the institution. The storerooms — where the science lives — are chronically starved. Ask your local museum how its collections are funded, and make the unglamorous back rooms part of the conversation.

Push for open, shared specimen data. A discovery is only possible if the images and scans are available to compare. Encourage institutions and funders to make digitised collections freely accessible across borders. Locked data helps no one; open data lets a researcher in one country spot the oddity in another country's drawer.

Visit, and ask what is behind the wall. Curiosity funds curiosity. When you tour a natural history museum, ask staff about the collection behind the scenes and the backlog waiting to be studied. Public interest is what keeps the lights on in the corridors the public never sees — and those corridors hold the next decade of headlines.