For almost a decade, iNaturalist has been helping curious nature lovers around the world get better acquainted with their local flora and fauna through its network of natural history experts. Now, the team behind the popular online community is exploring how artificial intelligence can help improve the speed and accuracy of its crowd-sourced identifications.
iNaturalist, which began in 2008 as the final project of three UC Berkeley School of Information Master's students, relies on a relatively small community of 'identifiers' to browse photos posted by users and identify the species. However, the popularity of the program — iNaturalist currently has about 500,000 registered users — means that the number of photos posted daily has long since surpassed the number of people available to identify them. It currently takes an average of 18 days for a species to receive a positive ID, but the iNaturalist team is now testing an app that uses a form of machine learning called computer vision to produce suggestions more quickly.
Using iNaturalist's existing database of around five million 'research-grade' images (images that have been vetted by the community), the computer vision model was initially trained to recognize several thousand species. When the app is fed a new photo, it instantly churns out a list of possible identifications based on what it knows. The community still has to verify the identification, but as it receives more research-grade images, the model “learns” to recognize more species and produce more accurate identifications.
“We’re essentially building a big data sensor, powered by kids exploring their backyard,” explains Scott Loarie, co-director of iNaturalist. He hopes the new, easy-to-use feature will inspire even more people to become citizen scientists and contribute to important research on species distribution and threats to biodiversity.
Andrew Simon, a naturalist from Galiano Island, B.C., has been using iNaturalist for years to engage his community in documenting the island’s species through bioblitzes. “It's proven to be the perfect platform for the Bio Galiano project, which is devoted to raising awareness about our local flora and fauna while creating an opportunity for people to contribute to science,” he says. Two species, a weevil and a moth, are thought to be newly discovered thanks to the Bio Galiano project.
Simon is skeptical of apps that claim to be able to identify species instantly and accurately, which he says does a disservice to aspiring naturalists, but is optimistic about the computer vision tool. “The great thing about iNaturalist is that it is an interactive process," he says. "It makes you think through your identifications.”
Testing the app
The question is, does the computer vision model work? I decided to test the app by uploading an image of a lime-green ladybug with a Rorschach pattern that I encountered in Ecuador. The program declared itself "pretty sure" my bug belonged to the tribe Chrysomelini — not exactly a species-level ID, but at least it wasn’t wrong. The program then offered a list of ten possible species. When I shared the same photo with the iNaturalist community at large, someone was able to identify it as Calligrapha fulvipes after a few days.
Loarie says at this stage, the app's performance is based on data input, not on how recognizable a species is. It could struggle to identify a common mammal of which only a few poor-quality photos exist in the database, but nail a rare plant of which there are many high-quality photos. For the same reason, it’s also biased to be more accurate in areas with lots of observations. And, for many taxonomic groups such as plants, fungi, or arthropods, the difference between species can lie in the minutest of details, which the program may not pick up on.
“The artificial intelligence is only as smart as the community that we train it on, and it’s sometimes wrong, but we still have this great vibrant community who will correct it,” Loarie explains.
Curious to try iNaturalist for yourself? They have a dedicated bilingual Canadian site run by the Royal Ontario Museum and the Canadian Wildlife Federation, with more than 14,000 registered users.