A.I. will soon help us save endangered species
Camera traps have data. A.I has organization. One Google-funded group thinks they're a match made in heaven.
Conservationists use devices called camera traps to identify which species are living in certain habitats. These camera traps are equipped with either a motion sensor or an infrared sensor and take a picture when they detect something moving in surrounding area. Though these camera traps can be extremely useful, they take a ton of pictures, and the pictures often don’t actually feature any wildlife. Furthermore, there aren’t databases that are widely available where conservationists can share and view these images.
Too much unorganized data is a problem that calls out to artificial intelligence. Wildlife Insights, a collaborative effort between conservation groups, universities, museums, and one of Silicon Valley’s largest players, has just unveiled its database that will allow conservationists to share their images, sort through them and identify which species their camera traps are photographing.
The database uses A.I. developed by Google to sort out pictures that don’t contain wildlife, and it can identify many different kinds of species. The database uses A.I. developed by Google to sort out pictures that don’t contain wildlife and can identify a wide variety of species.
Jorge Ahumada, the executive director of Wildlife Insights as well as a senior wildlife conservation scientist at Conservation International, tells Inverse that the collaboration is still in the early days, but they already have 62 projects in the database from around the world. Wildlife Insights is working with 4.5 million images that feature around 600 species. Over 80 percent of the photos in the database feature wildlife, and so far the A.I. has been able to accurately identify around 100 species 80 to 90 percent of the time.
“There are millions and millions of camera trap images out there in the world, but because there’s no way to process them or share them or analyze them, they’re unavailable to conservationists,” Ahumada says. “The folks who are collecting this data—they’re just sitting on it, basically. Most of them don’t have the capacity to go through it and look at the images one by one. If they do, it’s going to take them a long time.”
Ahumada claims camera traps in just one area can capture up to 100,000 images over the course of a month. That’s a lot of pictures to sort through. Even if a conservationist does sort through the images, they’re typically not sharing them very widely. This project will change that.
All of the images are uploaded under a creative commons license so anyone can access them and use them, and anyone who sets up camera traps or just wants access to the images can make an account. Even people who put camera traps on their property as a hobby can get involved.
“Anybody who wants to do a study can download images for a certain species or download images from a given country,” Ahumada says.
Ahumada says the A.I. was created using a deep learning model that was exposed to millions of images from camera traps so it could learn to identify different species in different environments.
“This is a deep learning model that was trained with about 8.4 million camera trap images that were already cataloged, so somebody had gone through all of these images from various parts of the world,” Ahumada says. “The model got trained with about half of those images, then a data set was used on the other half to test how it works.”
This project will be useful for many purposes. Ahumada says conservationists will be able to use it to track wildlife in certain areas, and what they learn with this data can help them understand which species are in a given habitat, if they’re migrating and more. This will help conservationists protect these animals, learn how an environment is affecting them and could also help shape public policy.
“In many places, we really don’t know what species are in these protected areas. We assume there will be this or that, but we don’t know,” Ahumada says. This is a way to confirm those species are there. This is also a way to measure what is happened to these species over time, as forests get cut down and the climate warms.”
If you notice a species is starting to congregate in one part of a habitat, you can go check if something is causing them to avoid another part, which could include poachers or a degrading environment. You might take what you learn from the traps and decide to request that government officials change the number of hunting licenses that will be available or request that lawmakers better protect a certain area.
The U.N.’s Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) claims almost 10 percent of terrestrial species are currently under threat, and many of those species will likely disappear within a decade. That’s over 500,000 species that are threatened. Not only are we losing species that are important to different ecosystems, but the disappearance of a species also presents a threat to the climate.
“When you lose a species, like vertebrates, you lose a lot of the forest’s function, like storing carbon,” Ahumada says.
These animals eat the fruit from plants that contain seeds that don’t easily disperse without them. The fewer animals that are spreading the seeds of these plants, the less new ones are growing that can absorb greenhouse gases. The world’s forests are estimated to absorb nearly 40 percent of the CO2 humans produce.
“There are all of these cascading effects that happen when we lose a species,” Ahumada says.
You can go to Wildlife Insights to see how the project is developing and get involved if you’d like. The A.I. will get better at identifying different species as the network grows and more camera trap mages are uploaded.