A group of brown-headed cowbirds has had their every move and song recorded in an advanced aviary in South Philadelphia over the past several months.
University of Pennsylvania scientists hope to use the unpreceded trove of data to unlock some secrets of animal behavior.
“We’re interested in the social group and how they kind of align themselves, and how eventually that leads to reproductive success,” said biologist Marc Schmidt, speaking near the enclosure. It’s maybe 80 square feet, with faux nests and eggs meant to more closely mimic nature than the lab.
“So we can ask questions, you know, given these two pairing up, and their entire prior history — how does that translate into them laying more eggs,” he said.
Traditionally, Schmidt said, a scientist studying bird behavior would have to simply sit in an aviary and take notes.
“But if you’re just observing, all you can do is observe one bird at a time. You can’t observe the entire network of birds,” he said. “So the approach here is a very agnostic approach, where we just observe everybody and try to figure out what are the rules of the game.”
He said agnostic, but the view his team gets is closer to omnipotent. The wire aviary, located at the Pennovation campus in the Forgotten Bottom section of Grays Ferry, is equipped with two dozen advanced microphones and eight high-definition cameras. The system can record half a terabyte of compressed data a day.
“This season was the first season where we could record continuously, and we recorded for a hundred days nonstop,” said Schmidt. “So that is half a terabyte times 60, over 120 days — which is a lot of information.”
The system is especially useful when comes to studying very rare behaviors, he said.
“These very rare behaviors that occur maybe 2% of the time under very specific social contexts,” Schmidt explained.
One such event is called a “wing stroke,” he said. Female cowbirds assume a mating position in response to a strong song from a male. But sometimes, a juvenile male tries out a tune, and a female very subtly fluffs up her feathers, as if to say, “Almost there.”
“Then the next 10 songs that he sings are going to be exactly the same. So basically, she reinforced and said, ‘I liked that song.’” said Schmidt. “And so he then he keeps practicing — he’s kind of exploring song variants, and she’s telling him what song to produce.”
There are also second- and third-order interactions. Schmidt said it’s simple enough for researchers to study two birds interacting with each other: Bird A sings to Bird B, and she responds one way or another. But how does a watching Bird C feel about all that?
The project brought together a team of biologists, engineers, physicists, and of course, with all that data, computer scientists. Postdoctoral researcher Marc Badger said the challenge now is to use machine learning to train computers to sift through all the information they gathered.
“The data are there, like there in the videos … we could know where every bird was over the entire day for 90 days and where, you know, where they’re looking, when they call, what type of call it was,” he said. “That data is there, and we just need to develop algorithms to extract it.”
Step One is to teach a computer to track individual birds, then to recognize movements and postures among individual birds.
The data collected from just one season, Badger said, will be studied for years to come.