Artificial intelligence is being used to ID goose faces : NPR

A few years ago, Sonia Kleindorfer was interviewing to become the director of the Konrad Lorenz Research Center for Behavior and Cognition in Vienna, Austria.

“My predecessor was telling me this story,” she says. Konrad Lorenz was a famous Austrian biologist who spent much of his career studying the behavior of local Greylag Geese. Famously, his students presented him with a plaque that had 30 goose faces on it.

“He went on to correctly name each goose,” she said. “He made one error, and that was two sisters.”

Another longtime researcher at the institute could do it too, and Kleindorfer felt a certain amount of pressure, as the new director, to learn how to tell the geese apart. But she just couldn’t get it.

“I can do five, but when the next five come, I start to have a mental meltdown,” she says. “So I’m actually not as good as I would like to be.”

It was embarrassing, frankly, so she contacted a more technically-minded colleague and asked him: Could he write a program to distinguish these faces?

He said, yes, but he’d need a database of geese photos to work with. Kleindorfer got her team out there, snapping pics of the geese from every angle. After building the database, they wrote a piece of facial recognition AI that could ID a goose, by looking at specific features of its beak.

It took a couple of years, but, writing in the Journal of Ornithology, the team reports that their goose recognition software is now about 97% accurate.

“So we have nailed the AI, but then you have to ask yourself … does it matter in the life of a goose?” she asks.

Goose drama

And here let’s just take a moment to talk about the lives of geese, because as even a casual observer can attest, they are not the most pleasant animals. Kleindorfer says that’s in part because they have a lot going on:

“Geese have such drama — there’s arch rivals, and jealousy and retribution,” she says.

To find out how faces figured into this drama, Kleindorfer presented the geese with full-sized cutouts of themselves, their partner, or another member of the flock. She showed evidence that geese seemed to recognize photos of their partners and friends, but not themselves (since geese cannot see their own reflection easily, they presumably treated their own image like a stranger).

She thinks that makes sense, because all that goose drama can only happen if they can tell each other apart.

“This facial recognition, we think, might be a key component in higher-level social organization among unrelated individuals,” she says.

That’s probably why humans are good at recognizing each other too.

This is not the first facial recognition program for animals. In recent years, researchers have used it on everything from lemurs to bears.

AI for conservation

“One thing that’s really exciting about this is that it’s AI being used for good,” says Krista Ingram, a biologist at Colgate University in New York.

Ingram has developed SealNet, an AI tool that can tell harbor seals apart. Before SealNet she says, the only way to identify individual seals was by tagging them, but that was difficult. The best way to do it was to try and shoot them with tracking darts.

“It’s very time consuming, costly and to be honest it stresses them out,” she says.

SealNet can ID seals with high accuracy using just a photo. It’s easier, faster, and better for the seals and the scientists.

Both Ingram and Kleindorfer think that facial recognition is going to play a really important role in conservation and ecology. Researchers will be able to tell how many individuals are in a population, they’ll see who’s hanging with whom.

“I do think it’s the wave of the future,” Ingram says, though she notes that field biologists struggle to compete financially with Silicon Valley for the researchers who truly understand how to build AI systems.

“We need more computer scientists trained in behavioral ecology and we need more conservation scientists trained in computer science,” she says. “But working together, I think we can do this.”

And citizen scientists could be part of it too. Sonia Kleindorfer hopes birdwatchers will someday be able to snap a picture of a goose, ID it, and share its location with scientists.

But she adds, just remember, her new research suggests that birdwatching goes both ways: Geese can remember faces too.

“If you are ever not kind to a goose,” she warns, “woe to you the rest of your life, it shall not be a happy one, if that goose finds you again.”