Artificial intelligence: Tool or distraction for bee taxonomy?
Taxonomy is a popular and important issue for us here at the Bee Report. We’ve previously talked about it on the podcast with both Zack Portman and Hollis Woodard. We've talked about the taxonomic bottleneck which, if not overcome, could hamper effective conservation plans and future science. We’ve talked about the potential solutions as well: sufficiently support and fund the taxonomists currently at work, get more qualified people engaged with identification, and develop new tools to facilitate the process of identification. And there are plenty of opinions about where we should be focused in solving the problem.
In last week’s newsletter, I included a tweet from the British Entomological Society about “a robot‐enabled image‐based identification machine, which can automatically identify, sort and estimate the biomass of invertebrates”. That’s right: Artificial intelligence to identify insects.
The system is called BIODISCOVER (an acronym for BIOlogical specimens Described, Identified, Sorted, Counted and Observed using Vision-Enabled Robotics) and is described in a recently-published paper from a group of Finnish and Danish researchers. To give you the highlights of how the system works and performs: A specimen is placed into a cuvette filled with ethanol at the top of the device. As the specimen slowly floats and spins to the bottom, two cameras capture images from two angles. The images are then fed to a neural network (a form of artificial intelligence) that has been trained on images of different invertebrate species. It then kicks out a list of species and abundances from the total collection sample you’ve run through it. The researchers tested the system with 12 insect and spider species (all of them easily identifiable from each other, and none of them bees), and reported a 98% accuracy rate – with lower rates of accuracy for species with fewer examples in the system, and difficulty identifying rare and unknown species.
“The machine will be useful for bulk identification while taxonomic experts can concentrate on more difficult specimens and rare species,” said Johanna Ärje, the corresponding author on the paper, via email. Her recent post doc work was on automatic insect identification at Tampere University in Finland.
BIODISCOVER seems like an interesting and potential taxonomic tool. The question here is: How well might a system like this work with bees?
“The specimen must have room to spin around in the liquid while it’s being imaged,” Ärje said. “If that is something that can be done with bee samples, then I don't see a reason this system shouldn't work for bees as well.”
Bee taxonomists, however, have a few additional thoughts.
“First off, in order for machine learning to be successful, you require a solid taxonomic foundation. We don’t have that in bees,” said Zach Portman, bee taxonomist at the University of Minnesota in the Cariveau Native Bee Lab, via email. “Second, I don’t think many people appreciate the sheer level of variation within species, and especially how many species vary so much in characters (such as coloration) that are meaningless in terms of species identity.”
Dialictus bees, for example, essentially all look the same. What variation does exist takes the form of microscopic characters such as the size and spacing of integumental punctures.
“Plus the specimens themselves are often in poor condition, with their bodies twisted around, gunked up, hairs matted, or integument worn down,” Portman said.
“I would think a student could be trained to do the same work in an hour and wouldn’t require storing or processing large numbers of images,” said Jason Gibbs, assistant professor and curator for the museum of entomology at the University of Manitoba. “These tools may eventually play a role in easing some of the routine identification burdens that taxonomists generously provide, but these types of tools can often distract from the larger problem of being able to delimit and describe biodiversity before its too late.”
Artificial intelligence is being proposed or implemented as a solution in countless situations these days – and taxonomy is no exception. We need to have these sorts of discussions to ensure that enthusiasm is balanced with critical consideration. When it comes to tackling the issues underlying the taxonomic bottleneck, could AI be a viable tool or is it distracting us from a more effective focus?
Many thanks to Johanna Ärje, Zach Portman and Jason Gibbs for taking the time to correspond with me on this topic.
The podcast will be back soon with a brand new episode! In the meantime, why not catch up on some previous episodes? Available on Apple, Stitcher, other major platforms and at thebeereport.buzzsprout.com.
Do you have tips, comments, questions or ideas for collaboration? Please send them to tbr@bymattkelly.com.
Conservation
Rusty patched bumble bee found at Iowa nature center
(Quad-City Times) Amy Loving was working at Davenport's Nahant Marsh Education Center, taking photos of bees and other insects buzzing around the building in preparation for an upcoming program. Spotting a bee in some white clover, she crouched down to get a shot. Back inside the center some time later, she was scrolling through her images when suddenly she felt her heart begin to race. One of her pictures showed a bee with a rusty patch in its yellow backside.
Science
Strange, spiral bee combs look like fantastical crystal palaces. Now we know why.
(Live Science) How did these Tetragonula bees become the Frank Lloyd Wrights of the insect world? Does each colony employ its own master architect, charged with guiding his comb's construction — or does each worker bee merely follow an unconscious set of individually-encoded building rules? According to a new study, “These combs follow the same basic rules that cause crystals to grow up in a spiral pattern.”
Osmia collect pollen from both coevolved and novel plant species
(Twitter, Anthony Vaudo @ThatBeeAnthony) "With metabarcoding of Osmia cornifrons larval provisions we found they have distinct family level preferences, but can switch host species native to many regions, a mechanism for how introduced bees can establish in new habitats?"
Natural history of an Australian bee community, studied for ten years
(Austral Entomology) “We surveyed bees using a blue vane trap during spring, summer and autumn from 2008 to 2017 at one location in Canberra, Australia. To the best of our knowledge, this is the longest near‐continuous record of bee activity in the southern hemisphere... Our findings relate only to our study site but are similar to findings from other long‐term studies conducted in the northern hemisphere, which collectively present a picture of high natural variability in bee communities that must be considered when interpreting findings of bee responses to anthropogenic disturbances.”
With the right mix of plants, urban green spaces can support diverse pollinators
(ScienceDaily/PLOS) Researchers from RWTH Aachen University in Germany surveyed diverse green spaces across the city of Aachen to assess plant-pollinator interactions within an urban environment. Pollinator visits varied significantly between different types of green space.
Society/Culture
College biology textbooks still portray a world of white scientists
(ScienceNews) According to a new analysis, mentions of white men still dominate biology textbooks despite growing recognition in other media of the scientific contributions of women and people of color. Some good news: Scientists in textbooks are getting more diverse. The bad news: If diversification continues at its current pace, it will take another 500 years for mentions of Black/African American scientists to accurately reflect the number of Black college biology students.
Technology
WiBee data dashboard is up and running
(Twitter, Gratton Lab @GrattonLab) “Since April 1, our WiBee citizen scientists have already submitted over 500 bee surveys! We’ve just launched a public Data Dashboard for anyone interested in wild bees to explore the data. It’s a work in progress, but we couldn't wait to share it: data-viz.it.wisc.edu/wibee/”
One More Thing…
“Clearing opium fields hurts honey bees”
From the archives of Scientific American, 1911: “Some unexpected results are found from the movement against the production of opium in China. In the Yunnan, one of the provinces where opium was produced in large quantities, it appears that the poppy is no longer cultivated, owing to the recent measures. However, this has had a disastrous effect on the honey culture of the region. As the bees find no more flowers, the production of honey is stopped. The new crops which replace the poppy, such as wheat or peas, are not such as will give as good a honey yield. On another side of the question, it appears that the opium habits of the population are not suppressed by the present legislation, as some supposed would be the case, but are again on the rise.”