The Citizen Eye

A Thesis That Became a Database

In 2008, three graduate students at UC Berkeley's School of Information built a website for their master's thesis: a platform where anyone could upload a photograph of an organism and get help identifying what they had found. Co-founder Ken-ichi Ueda had a simple motivation — he had recently moved across the country and found himself surrounded by unfamiliar species with no adequate tool to learn their names. The platform was meant to fix that gap, for him and for anyone else curious about the living world.

That thesis project is now the world's most consequential citizen biodiversity database. As of August 2025, users have submitted nearly 290 million observations of organisms spanning roughly half a million species — approximately a quarter of all known life on Earth. Nearly 4 million registered accounts have contributed at least one verifiable observation, and the data they have generated has been cited in nearly 7,000 peer-reviewed research papers. The computer vision model iNaturalist runs today — version 2.27, released January 2026 — can suggest species identifications for 112,613 taxa, up from roughly 25,000 in its first iteration.

iNaturalist became an independent 501(c)(3) nonprofit in July 2023, having separated from the California Academy of Sciences and National Geographic Society after nine years of institutional incubation. In 2024, co-founders Ueda and Scott Loarie received the Heinz Award for the Environment — one of the most prestigious environmental prizes in the United States. In December 2025, Ueda stepped down after nearly 18 years of building the platform. Loarie continues as Executive Director.

From Snapshot to Scientific Record

For a casual photograph of a butterfly to become a citable scientific data point, it must pass through iNaturalist's tiered quality system. The journey has three stages — and only the final one matters to scientists.

Casual observations are records lacking a date, location, or media, or depicting captive or cultivated organisms. They are set aside as scientifically ineligible. The bulk of useful contributions fall into the Needs ID pool: verifiable records that await community confirmation. The platform's Identify interface serves these observations to expert volunteers, who can agree with a proposed identification, propose a more specific one, or register a dissenting "maverick" view — which is logged but does not block eventual promotion.

Research Grade is the destination. An observation reaches this tier when at least two identifiers agree at the species level, with more than two-thirds of all identifiers concurring at that level or below. In 2025, the community contributed more than 82 million identifications across 67 million observations, of which 34.3 million ultimately reached Research Grade.

Research Grade observations are automatically forwarded to the Global Biodiversity Information Facility (GBIF), the world's foremost open-access biodiversity data aggregator. iNaturalist is now the single largest GBIF contributor for plants, mammals, reptiles, and amphibians. Approximately 80 percent of GBIF records published since 2010 originate from citizen science platforms — and iNaturalist generates the largest share.

The AI Engine: Vision Meets Geography

Since 2017, iNaturalist has powered its suggestions with a deep learning computer vision system trained on its own community-verified photographs. Model v2.27 (January 2026) recognizes 112,613 taxa, trained on data exported in November 2025. Development has proceeded in close collaboration with the Visipedia project at Caltech, whose annual computer vision challenges at the CVPR conference use iNaturalist data to benchmark species recognition globally.

What distinguishes iNaturalist's AI from a simple image classifier is its integration with a second model: the iNaturalist Geomodel, introduced in September 2023. The Geomodel is a spatially-aware neural network — based on a technique called Spatially Implicit Neural Representations (SINR), published at ICML 2023 — that takes a geographic location as input and returns a probability distribution over species likely to occur there.

In practice, this means the app's suggestions are not based on visual similarity alone. The same photograph of a ground beetle taken in Finland and in Brazil will return different ranked suggestions, because the underlying geographic probability distributions are distinct. The app's label for this behavior changed from "Seen Nearby" (raw observation counts) to "Expected Nearby" (model prediction) when the Geomodel launched. Combined, the two models make species identification dramatically more accurate for rare, cryptic, and poorly-documented taxa that a purely visual system would misidentify.

What the Eyes Have Found

The platform's most compelling contributions are not statistical — they are stories of individual observations that triggered discoveries that would have been otherwise impossible.

In March 2024, a volunteer at Big Bend National Park in Texas photographed a small, silvery-fuzzed plant and uploaded the image to iNaturalist. No one in the community recognized it. The photograph eventually reached Isaac Lichter Marck, a postdoctoral researcher at the California Academy of Sciences, whose DNA analysis confirmed the plant was not just a new species but an entirely new genus in the sunflower family — Ovicula biradiata, informally called the "woolly devil" for its white, woolly hairs. It was the first new plant genus described from a U.S. national park in nearly 50 years, since a 1976 find in Death Valley. Lichter Marck warned that the species may already be threatened by climate-driven desertification: "It's possible that we've documented a species that is already on its way out."

In June 2025, researchers used thousands of iNaturalist photographs to link whole-genome sequencing data to morphological variation in Physalia (the Portuguese Man-o-War). The result: what had been classified as a single cosmopolitan species was reclassified as four distinct species. In December 2024, a farmer in New Zealand's Ashburton Lakes photographed insects on a speargrass plant. The insects proved to be the critically endangered Canterbury knobbled weevil — previously believed to exist at only a single location 80 kilometers away.

These discoveries follow a consistent pattern: an unusual observation is uploaded, community experts flag it as anomalous, specialist interest is triggered, and deeper investigation follows. The key enabler is not any individual's expertise — it is the platform's capacity to route unusual records to the people who know what to do with them.

The Science Factory: 7,000 Papers and Counting

A landmark study published in BioScience in November 2025 — Mason et al., 15 institutions across 6 countries — found that iNaturalist data had appeared in papers from 128 countries, across 638 taxonomic families. Use of iNaturalist data in peer-reviewed research had grown tenfold in five years. In 2022 alone, more than 1,400 peer-reviewed articles incorporated iNaturalist data — approximately four per day.

"iNaturalist is really pervasive throughout the biodiversity research. It is fundamentally shaping the way that scientists think about research and think about designing studies and think about answering questions about biodiversity."

— Corey Callaghan, ecologist, University of Florida, co-author of Mason et al. (2025)

The citation counts are likely an underestimate: researchers can access iNaturalist data through GBIF without explicitly naming iNaturalist in their papers. The actual influence on the scientific literature is considerably larger than any citation search can capture.

The City Nature Challenge — an annual four-day global bioblitz each April — illustrates the platform's velocity. Launched in 2016 as a competition between Los Angeles and San Francisco, the 2025 edition drew 102,000 participants across 800 cities on six continents, generating 3.3 million observations of 74,000 species in four days.

Seek: The Gateway Into the Natural World

Running parallel to iNaturalist is Seek, a companion app designed for children, families, and casual users who may not want to engage with the public data-contribution model. Seek uses the same computer vision system but delivers identifications in real time as a live camera view. When a user points their phone at a plant or animal, the app narrows its classification on-screen as the image resolves, prompting capture when it reaches species-level confidence.

Seek is COPPA-compliant: it requires no registration, stores no location data beyond what is needed for local species suggestions, and shares no data publicly. The gamified structure — species identifications earn badges — makes it usable in school settings where iNaturalist's public-sharing model would create privacy concerns. Support has come from the California Academy of Sciences, the National Geographic Society, WWF, Netflix's Our Planet, and Visipedia at Caltech.

The Early Warning System That Is Almost Ready

iNaturalist's product roadmap for January-June 2026 is focused on observer experience improvements: a redesigned Explore interface for iOS, the ability to add observations to projects within the app, continued data quality work. In June 2025, Google announced iNaturalist's inclusion in its Generative AI Accelerator program, signaling deeper integration of large-scale AI tools into the platform's identification systems. In December 2025, iNaturalist launched a pilot "provisional name" field for DNA-barcoded fungi likely representing undescribed species — one sign of the platform's push further into the frontier of taxonomy.

The more consequential development is structural. The combination of the computer vision model and the Geomodel creates, in embryonic form, the most powerful species-distribution anomaly detection system ever assembled. When an observer photographs an invasive species 200 miles north of its known range and uploads the image, the system can, in principle, recognize both what it is and that it shouldn't be there — and route that record to a manager with the capacity to act on it.

The question iNaturalist faces in 2026 is whether it can complete that circuit: observation → identification → anomaly detection → alert → response. The data is there. The AI is there. The global observer network is there. What remains is the institutional plumbing to connect them — and the willingness, among conservation agencies, to trust a citizen dataset enough to respond to it in real time. The platform that started as a graduate school thesis has built everything except the last link in the chain. That link is not a technical problem.

Sources: iNaturalist 2025 Year in Review (inaturalist.org/stats/2025) • Mason et al. (2025), "iNaturalist accelerates biodiversity research," BioScience 75(11) • iNaturalist Geomodel blog (blog/84677) • Computer vision overview (posts/42738) • Smithsonian Magazine, woolly devil (Feb 2025) • Cole et al. (2023), SINR, ICML/PMLR • City Nature Challenge citynaturechallenge.org • Heinz Award (blog/98844) • Independence announcement (blog/82010)