An instruction manual we're losing before we learn to read it
Across the world's herbaria and field stations, a quiet revolution is unfolding: artificial intelligence and mass digitisation are giving botanists new eyes with which to see what centuries of human effort have only partially revealed. With four in ten assessed plant species facing extinction and hundreds of thousands more never yet named, the stakes of this technological turn are nothing less than the preservation of life's undiscovered pharmacopoeia and ecological web. A landmark report from Royal Botanic Gardens, Kew, assembled by 400 scientists across 40 countries, suggests that the tools now emerging may finally match the scale of the crisis — though the race between discovery and disappearance remains desperately close.
- Plants are vanishing before science can name them — 100,000 species remain undiscovered, and the current pace of identification barely keeps up with the rate of loss.
- AI is outperforming human specialists at identifying cryptic species like sedges and peat mosses, compressing decades of expert training into near-instant algorithmic recognition.
- Kew's digitisation of 7.4 million specimens — including Darwin's own collections — has cracked open centuries of locked knowledge, but the 145 million specimens now online still represent less than 16 percent of global holdings.
- Climate disruption is measurable in the archive: AI analysis of 8 million digitised specimens reveals flowering times shifting 2.5 days per decade, quietly unravelling the synchrony between plants and their pollinators.
- Fungal specimens 180 years old are yielding viable genetic material, pointing toward undiscovered medicines and disease prediction tools — but 90 percent of an estimated 2 million fungal species remain entirely unknown.
- The report's authors warn that AI's energy appetite and the risk of amplifying data inequalities mean this technological rescue mission carries its own contradictions, demanding political will alongside scientific ingenuity.
Botanists are in a race they may not be winning. Roughly four in ten of the 70,000 plant species scientists have managed to assess are headed toward extinction, and another 330,000 haven't been studied at all. Somewhere around 100,000 more await even a name. Each year, researchers formally identify about 2,000 new species — progress that barely dents the surface of what remains unknown.
A major report from Royal Botanic Gardens, Kew, produced by 400 scientists across 40 countries, argues that AI and digitisation could be the turning point. AI systems are already identifying plants that have long defeated human specialists — sedges and peat mosses whose distinguishing features are nearly invisible to the naked eye — faster and more reliably than experts who spent decades learning to see the difference.
The digitisation effort is staggering. Kew has scanned all 7.4 million of its specimens, including plants collected by Darwin, making them freely available online. At peak, technicians were capturing 20,000 high-resolution images a day. Globally, 145 million digital specimens are now accessible — yet this represents less than 16 percent of all herbarium holdings worldwide. In Madagascar, digitising 37,000 physical specimens unlocked patterns spanning centuries of collection, enabling international collaboration that was previously impossible for scientists in the global south.
The technology is also measuring climate change's quiet toll on plant life. Researchers fed 8 million digitised specimens into an AI trained to recognise flowers, revealing that flowering times have shifted by an average of 2.5 days per decade. In India's Western Ghats, the proportion of kindal trees blooming in synchrony fell from 80 percent to below 50 percent by the 1990s — a fracturing of the ecological relationships that sustain entire ecosystems.
Perhaps most striking is what ageing fungal specimens are beginning to reveal. Ninety percent of an estimated 2 million fungal species remain unknown, yet genetic material is now being extracted from specimens up to 180 years old. With fungi-borne pathogens apparently spreading as climate change extends warm seasons, these old collections represent what researchers are calling a genomic goldmine — a potential source of new medicines and disease prediction, in the tradition of penicillin and statins.
The report's authors are candid about the contradictions. AI datacentres already account for 6 percent of electricity consumption in the UK and US, and digitisation risks deepening existing inequalities if the underlying data isn't expanded. The race against extinction, they conclude, is no longer only a biological problem — it has become a technological and political one too.
Botanists are racing against time. Somewhere in the world right now, a plant species is disappearing—possibly forever—without anyone ever knowing it existed. About four in ten of the 70,000 plant species scientists have managed to assess are headed toward extinction. But that's only the beginning of the problem. Another 330,000 species haven't been studied at all. And somewhere in the undocumented corners of the planet, roughly 100,000 plant species are waiting to be named by science at all. Each year, researchers formally identify about 2,000 new plants. It sounds like progress until you realize it barely dents the surface of what's out there.
This is where artificial intelligence and digital technology are beginning to change the equation. According to a major report from Royal Botanic Gardens, Kew, produced by 400 scientists across 40 countries, the combination of AI and digitisation could be the turning point in what botanists call the race against extinction. The technology is already doing things that seemed impossible a few years ago. AI systems can now identify plants that have defeated human specialists—sedges and peat mosses whose distinguishing features are so small they're nearly invisible to the naked eye. When a new or vulnerable species appears in the field, these algorithms can spot it faster and more reliably than the experts who spent decades learning to see the difference.
The digitisation effort itself is staggering in scope. Royal Botanic Gardens, Kew has now scanned all 7.4 million of its specimens, including plants collected by Charles Darwin himself, and made them freely available online. At the peak of the project, technicians were capturing 20,000 high-resolution images every single day. Globally, 145 million digital specimens are now accessible online—but that's still less than 16 percent of all the plant and fungi specimens held in herbarium collections around the world. The blind spots remain enormous.
Yet even this partial digitisation is unlocking knowledge that was locked away in archives. In Madagascar, one of Earth's most extraordinary biodiversity hotspots, Kew botanists digitised 37,000 physical specimens. The act of making those images and their data accessible online revealed patterns and insights spanning centuries of collection. Landy Rajaovelona, a senior botanist at Kew Madagascar, described it as unlocking a treasure of knowledge that would have remained invisible to most of the world. When specimens from biodiversity hotspots in the global south become digitally available, international collaboration accelerates. Scientists who never had access to these collections can now study them.
The technology is also revealing how climate change is reshaping the natural world in real time. Researchers trained an AI model to recognize flowers, then fed it 8 million digitised specimens collected over the past century. The results showed that flowering times have shifted by an average of 2.5 days per decade. Some flowers are arriving earlier, others later—a seemingly small change with serious consequences. In India's Western Ghats, kindal trees that are crucial for timber used to flower in synchrony. About 80 percent of them would bloom at the same time. By the 1990s, that number had dropped below 50 percent. When plants and their pollinators fall out of sync, entire ecological relationships begin to fracture.
Perhaps most striking is what scientists are discovering about fungi. About 90 percent of the estimated 2 million fungal species remain unknown to science, and less than 1 percent of known species have been assessed for extinction risk. Yet new technology is now extracting high-quality genetic material from fungal specimens that are up to 180 years old. Dr. Esther Gaya, a senior research leader at Kew, explained that fungi thrive in heat and humidity, and as climate change lengthens warm seasons in temperate regions, some human pathogens appear to be spreading from warmer places. The old specimens in fungaria represent what researchers are calling a genomic goldmine—a repository of genetic information that could yield new medicines and help predict disease outbreaks. Penicillin and statins both came from fungi. The next breakthrough drug might be waiting in a century-old specimen.
But the report's authors acknowledge real concerns. AI datacentres consume enormous amounts of electricity and water. In May, the UK and US saw datacentres account for 6 percent of total electricity consumption. The scientists also warn that digitisation and AI could amplify existing inequalities unless the underlying data is expanded and improved. The report calls for partnerships between technology companies and environmental organisations, and for governments and funders to invest in plant and fungi collections. The race against extinction is no longer just a biological problem. It's becoming a technological and political one too.
Notable Quotes
These AI models can sometimes now identify better than specialists—that's incredibly exciting.— Prof Alexandre Antonelli, executive director of science at RBG Kew
By digitising 37,000 physical specimens, we've unlocked a treasure of knowledge spanning centuries, offering invaluable insights into today's biodiversity.— Landy Rajaovelona, senior botanist at Kew Madagascar
The Hearth Conversation Another angle on the story
Why does it matter if we lose a plant species we've never even discovered?
Because that undiscovered plant might be the source of the next antibiotic or cancer drug. It might be a crop that can survive in a hotter climate. We're losing the instruction manual before we've learned to read it.
So AI is actually better at identifying plants than the botanists who trained for decades?
In some cases, yes—especially with plants that have microscopic differences. An AI model can process millions of images and learn patterns a human eye might miss. But the AI needs the digitised specimens and the expert knowledge to begin with. It's not replacing botanists; it's amplifying what they can do.
What does it mean that flowering times have shifted 2.5 days per decade?
It means the calendar is breaking. A butterfly that evolved to emerge when a specific flower blooms might now arrive to find nothing. The plant blooms, the pollinator isn't there yet. Repeat that across thousands of species and ecosystems start to collapse.
You mentioned fungi specimens from 180 years ago. How is that useful now?
Their DNA is still intact. Scientists can extract it and study what genetic traits fungi had before modern pollution and climate change. That's invaluable for understanding how fungi might adapt—or how we might use them to adapt ourselves.
What's the biggest barrier to saving plants right now?
Time and money. We're discovering 2,000 new species a year, but there are 100,000 still unnamed. At this rate, we'll lose species we never knew existed. The technology exists to speed this up. What's missing is the funding and the political will to make it a priority.