AI is speeding up otherwise time-consuming digitization of its herbarium, bringing a collection of nearly 8 million plant specimens, dating back to 1642, online through a searchable database. They’ve also started using it in efforts to preserve tree populations in the US.
“Things that take humans a long time, you can teach an AI model to do, and it’s like having a few million people working in parallel,” said Matthew Albrecht, director of the Garden’s Center for Conservation and Sustainable Development.
Most projections surrounding AI in the future, the jobs it could do and problems it may solve, are human-health centric. MoBot’s application of this technology stands out by using it to address ecological needs — needs of the systems we inhabit, rather than those we invented.
AI has a role in conservation, said Jordan Teisher, curator and director of the MoBot herbarium. “It’s exciting, and slightly scary — but I think the sky’s the limit.”
MOVING ECOSYSTEMS
As the climate continues to change, many organisms will need to relocate to avoid the heat. Trees are no exception.
“Trees actually move,” said Adam Smith, an ecologist at the center for conservation and sustainable development at the gardens. “They don’t pick up their roots and walk around, but seeds allow them to move around.”
“One of the big differences between this century and the many hundreds of previous centuries is that climate change is happening at such a fast rate,” said Smith. “Most species won’t be able to move fast enough.”
Smith said a potential solution to this problem is assisted migration: helping animals and plants move to new habitats.
However, to keep ecosystems intact, scientists need to know how to move everything — ideally, following the ways organisms would move naturally. Humans are good at making predictions based on one variable at a time, but the way an entire ecosystem moves between one place and another depends on hundreds, if not thousands, of variables.
AI and machine learning — a kind of artificial intelligence — are powerful tools for dynamic problems like this. Using them is a little bit like playing the warmer-cooler game, but with statistics to guide you in the right direction more quickly, Smith said.
MoBot’s team has run thousands of simulations, combining information from herbarium specimens, fossilized pollen samples and DNA, to create maps of how green ash trees have moved since the last glacier retreated from the St. Louis area around 21,000 years ago.
“Once we have a handle on the past, we can have confidence about what will happen in the future,” Smith said.
PRESERVING THE PAST
Herbaria have collected plant specimens for thousands of years. All collections have a label with information like the collector’s name, the plant’s name, the location where the plant was found and a description of its environment.
Missouri Botanical Garden plans to use A.I. to help digitize herbarium archive
Daniel Bagby, a data processor with the Africa and Madagascar group in the Science and Conservation Department of the Missouri Botanical Garden, examines a label on a specimen from the herbarium on Wednesday, July 2, 2024.
The problem is, not all labels are made the same way, which makes getting all of their information into a database a mammoth task, especially at MoBot, which houses around 7.9 million specimens.
Labels can be in different places on the specimen (some in the bottom right, others on the top left). Some specimens have one label, but others have two, or more. Labels can be handwritten or typed, and some have smeared over time. This diversity would make it impossible for a scanner hooked up to a computer to do a good job — but this type of complicated task is what AI was built for.
And with every new specimen, it gets better at determining where the useful information is.
“The world’s herbaria have the only physical record of plant life on earth that we are ever going to have,” said Teisher, the MoBot herbarium director. “We are just starting to wrap our heads around how much data there is within each one.”
Digitizing specimens presents a quick and cheap way to make them accessible to the broader scientific community, something botanists have struggled with for centuries. Without an example of plants that have already been discovered (or eaten, or used as medicine), it’s hard for botanists to know if they’ve found a new species, or how that species might be helpful to humanity.
“When I started my research back when I was a student, there was no Internet and no cell phones. In Ecuador, I had to rely on whatever specimens they had there,” said Carmen Ulloa, senior curator for the Garden’s Science and Conservation Division. “Nowadays, since so many herbaria have digitized their collections, they are at everyone’s fingertips.”
Ulloa also pointed out the security afforded by digitizing specimens, citing the Berlin herbarium that was bombed in 1943 during World War II. The only specimens that survived were those on loan to other institutions or stored in the basement — or, that had been photographed prior to the bombing.
“The single greatest thing you can do on a day-by-day basis for global botany is digitize a specimen,” Teisher said. As the AI learns more about each specimen, it can transcribe information from the labels more quickly, easily outpacing a person doing the same task. However, the transcription is imperfect, so each AI transcribed label passes by the eyes of a herbarium employee before being released to the world.
“There’s a lot of fear around AI taking jobs, and in our case at least, this is not at all a substitute for our staff,” he said. “It acts as a supplement to them — a tool they can use in their work.”
This double-check still takes substantially less time than a person transcribing a label by hand, which means information once housed exclusively in shelves of the MoBot herbarium is quickly hitting the Internet — and absolutely all of it is accurate.
©2024 the St. Louis Post-Dispatch, Distributed by Tribune Content Agency, LLC.
