iNaturalist's AI Has Identified 90,000 Species From a Single Photo

Scientists estimate that humans have formally described only a fraction of the species that share the planet, and even among known species, distribution data is patchy and expensive to collect. A single field survey requires trained biologists, permits, and weeks of work. iNaturalist, a nonprofit social network for nature, replaces some of that effort with something almost everyone already carries: a smartphone camera and, behind it, an AI that can name what the camera sees.

The results are staggering in scale. As of August 2025, iNaturalist users had contributed nearly 300 million observations of plants, animals, fungi, and other organisms worldwide, with 250 million of them verified as evidence of a specific organism. More than 5 million people have registered accounts. The computer vision system at the heart of the app now recognizes more than 90,000 species, and its suggestions are accurate enough that they have become the first step in a pipeline that turns casual photos into data serious enough for peer reviewed science.

A Field Guide That Learns

iNaturalist began in 2008 as a master's project at the UC Berkeley School of Information. For its first years, species identification depended entirely on the community: a user posted a photo, and other members, including professional biologists, proposed or confirmed identifications. That worked, but it was slow, and it depended on an expert happening to see the photo.

Around 2017 the platform teamed up with computer vision researchers, including Grant Van Horn and the Visipedia team, who had recently built the bird identification app Merlin with the Cornell Lab of Ornithology. The collaboration produced a deep learning model trained on iNaturalist's own research grade photos. The team also published the iNaturalist species classification dataset at the CVPR computer vision conference in 2018, which became a widely used academic benchmark for a hard problem: distinguishing thousands of visually similar categories where a few species have millions of photos and most have only a handful.

Today the model is retrained regularly on fresh data exports, and each run adds thousands of new taxa as the pool of verified observations grows. When a user photographs a beetle or an orchid, the AI returns a ranked list of candidate species within seconds, each with a confidence score and example photos, so the observer can compare and confirm.

Location Makes the AI Smarter

A photo alone can be ambiguous. Two butterflies can look nearly identical while living on different continents. In 2023 the iNaturalist team addressed this with a geographic model that weighs the computer vision scores against which species are known to occur near where the photo was taken. According to the team's published evaluation, adding this spatial prior improved the accuracy of the first species suggestion from 75 percent to 87 percent on average, a large jump for an identification system that already worked well.

The combination matters for the community verification loop. Accurate first suggestions mean the AI proposes the right species more often, which speeds up confirmation by human identifiers, who review and agree or correct the suggestion. Once two thirds of identifiers agree on a species and the photo meets quality standards, the observation is labeled research grade.

From Backyard Photos to Published Science

Research grade observations flow automatically into the Global Biodiversity Information Facility, the international repository that aggregates biodiversity data from museums, governments, and citizen science platforms. As of early 2024, more than 4,000 research papers had cited iNaturalist data, many in ecology, conservation biology, and climate change research, including studies of climate driven shifts in species ranges and timing.

The platform has also contributed directly to species discovery and rediscovery. Observations posted by amateurs have brought attention to organisms that turned out to be new to science, and have documented species in places where they had not been seen in decades. A single hiker with a phone can now generate a verifiable biodiversity record that would once have required an expedition.

Where the Data Matters Most

iNaturalist has become a primary source of crowdsourced biodiversity data in regions where formal survey capacity is limited, including Mexico, southern Africa, and Australia. The annual City Nature Challenge, coordinated through the platform, mobilizes hundreds of cities to document urban wildlife in a single weekend. A companion app called Seek, aimed at families and classrooms, uses the same identification technology without requiring an account, lowering the barrier for young naturalists.

The platform is not without caveats. Identification confidence varies across taxonomic groups, and the data skews toward charismatic species and places where people carry phones. The AI can be confidently wrong, which is why human verification remains part of the process rather than an afterthought. But as a system that converts curiosity into scientific evidence at planetary scale, iNaturalist has no real precedent.

The Flywheel Effect

The deeper story is how the AI and the crowd feed each other. Every confirmed identification becomes training data for the next version of the model. Every improvement in the model makes identification faster and more fun, which attracts more observers, which generates more data. Since 2012, participation has roughly doubled year over year in sustained periods, and in 2023 the project became an independent nonprofit after years of support from the California Academy of Sciences and the National Geographic Society.

Three hundred million observations later, the lesson is simple. Biodiversity monitoring used to be limited by the number of trained biologists on Earth. Now it is limited by how many people decide to point their cameras at a bug on a leaf. The AI handles the naming, the crowd handles the checking, and science gets a living map of life on Earth that updates every single day.