To the untrained eye, zebras in Kenya probably all look alike. But each animal’s black and white markings are like a fingerprint, distinct — and invaluable for scientists who need to track the animals and information about them, including their births, deaths, health and migration patterns.

Traditionally, getting this kind of information has been an invasive and labor-intensive process. But breakthroughs in artificial intelligence (AI) and crowdsourcing of photos of individual animals are beginning to change the conservation game.

The A.I. Age | This 12-month series of stories explores the social and economic questions arising from the fast-spreading uses of artificial intelligence. The series is funded with the help of the Harvard-MIT Ethics and Governance of AI Initiative. Seattle Times editors and reporters operate independently of our funders and maintain editorial control over the coverage.

Portland, Oregon-based nonprofit Wild Me has developed AI to pick out identifying markers — the stripes on a zebra, the spots on a giraffe, the contours of a sperm whale’s fin — and catalog animals much faster than a human can. Photo surveys are increasingly used as the backbone for population estimates, and Wild Me’s Wildbooks, which catalog various species, are giving conservation groups, governments and citizen scientists a faster way to monitor animals around the world.

“We can use this information to track diseases and poaching threats, look at manifestations of diseases,” said Michael Brown, a conservation science fellow at the Giraffe Conservation Foundation and the Smithsonian Conservation Biology Institute, who has been working with Wild Me for the past few years. “It lets us piece together an understanding of how these threats to giraffes are spatially situated (and) how the giraffes are utilizing different landscapes over time.”

Founder Jason Holmberg launched the first iteration of Wild Me in 2003 after swimming with whale sharks off the coast of Djibouti. He wanted to find a different way to track the animals other than invasive tagging, so he teamed up with a biologist and a NASA astronomer, adapting the algorithm for the Hubble telescope to match the shark’s spot patterns.

For years, Holmberg’s endeavors were a side project — he didn’t leave his full-time job in tech until recently. Wild Me’s work gradually expanded, then it really kicked into gear with a 2018 grant from Microsoft’s AI for Earth. Today, Wild Me has a team of six full-time staffers, with plans to add more soon.


Wild Me’s process of creating and training algorithms takes serious time. Thousands of photos of the species must be manually annotated so that the algorithm learns what a given animal is, what the distinguishing characteristics are and what’s just background noise.

The model relies largely on photographs taken by scientists or everyday people who upload their photos to the corresponding Wildbook. It uses AI to “find things in the picture and then hand it to algorithms or machine learning to suggest IDs — which whale, which giraffe, etc.,” Holmberg said.

Christin Khan conducts aerial surveys of North Atlantic right whales for the National Oceanic and Atmospheric Administration and had sought an AI-based solution for years. She said she watched Facebook implement facial recognition and wanted to use similar technology to help identify whales within the endangered species (there are only about 400 North Atlantic right whales left).

“We needed a really simple, user-friendly web-based interface where a biologist who knows nothing about AI could upload a photograph and get a result back,” she said. “Eventually we realized the developers at Wild Me had already done a lot of what we needed, and it wouldn’t require us to reinvent the wheel.”

The Wildbook for whales, called Flukebook, encourages collaboration, which is particularly useful for whales that travel long distances because it can be difficult for one research group to effectively monitor one area.

“The more people on the water, the more photos, the more it’s decentralized, (the better),” said Shane Gero, who founded and runs the Dominica Sperm Whale Project. “By doing the matching themselves, by contributing their own data, not only do they get to know the animals, but it creates a locally motivated community of people that can react when conservation actions come up.”


Before the introduction of AI, Gero said it would take about a month to process a month’s worth of photos.

“(Now), we have our numbers of individuals sighted and population estimates faster, so we can report (almost) in real time,” he said.

That means his group is able to provide the government of Dominica with more up-to-date information and offer better advice on how to shape conservation efforts.

One of Wild Me’s more recent innovations is an AI-driven feature that datamines YouTube videos of whale sharks and sea turtles, using user-generated videos (often taken by tourists) to get a better sense of the populations. This has been a great way to increase the amount of photos coming in and provide researchers with more data. But it also creates even more work for people on the ground, who have to manually check the AI’s suggestions and accept the results.

“We’re flooding the whale shark community with more data than it can handle,” said Holmberg.

So Wild Me is now building the capacity to automate the identification process and scaling the tech that combs social media for relevant videos.

The nonprofit recently received a two-year grant from the Gordon and Betty Moore Foundation to develop the new algorithm that will make the animal IDs on its own.


It’s focusing the initial work on zebras because it already has an incredibly rich dataset. Every two years since 2016, the Great Grévy’s Rally in Kenya has used hundreds of “citizen scientists” spread out over thousands of kilometers to photograph Grévy’s zebras over two days. Wild Me’s AI analyzes the zebra markings on all the photos to come up with a total population, which the Kenyan government treats as the official census for Grévy’s zebras.

This type of work is a huge upgrade from the traditional “capture-mark-recapture” process, which is both invasive and time consuming, with studies done every five to 10 years.

“You can only make very coarse-grained conservation decisions,” Holmberg said. “The point of going to a fully automated system is to shorten that cycle so we can take all of the data over the past week or two weeks and have a continuous prediction of population size. It’s fine-grained, which helps researchers understand and lobby for better conservation activities.”

For Khan, meanwhile, the existing technology is still in its early days. The algorithm for North Atlantic right whales became operational in November 2019, and she said they’re still working out the kinks and figuring out how best to use it. But, she said, she sees the incredible potential that it holds.

“My dream is that we get to the point where the world’s oceans will be trolled by satellite photos and we can understand the world’s whale population,” she said. “Combining AI with satellite imagery and drones — we have the potential to exponentially understand the world’s oceans that’s just not possible with manned aircraft.”