Machine learning has worked wonders again. An artificial intelligence is now able to identify the dried plants making up a very extensive herbarium… just by looking at them, reports Nature, according to a study published on BMC Evolutionary Biology.
A herbarium is a collection of plants dried and pressed between sheets of paper, usually carefully composed by researchers. The world has around 3,000 of them, home to an estimated 350 million species – only a small fraction of which have been digitized.
Erick Mata-Montero from the Cartago Institute of Technology in Costa Rica, and botanist Pierre Bonnet from CIRAD decided to put an algorithm to the test by organizing a sort of plant blind test, with the aim of testing its recognition of plants. . They made him devour more than 260,000 scans taken from herbariums, containing more than a thousand different species. The AI ​​rose to the challenge and came out with 80% accuracy (i.e. 208,000 plants). The correct answer was found 90% of the time in the five guesses made by the algorithm.
This is enough to scare taxonomists, whom Pierre Bonnet wishes to reassure: “People think that this kind of technological advance could end up making human botanical expertise lose its value. But this is not the case. This feat was made possible only through human labor. What is more, we also still need humans to verify the veracity of the results provided. And to make herbariums.
Sources : Nature / BMC Evolutionary Biology / Ulyces