From Caws to Code: AI Helps Decrypt Animal Communication

by Abhishant Kidangur

• Scientists are increasingly using artificial intelligence models to decode the communication of other species.
• The Earth Species Project has developed a generalizable model that can be used across species; The team works with scientists around the world to create custom models for specific species.
• In northern Spain, ESP’s AI tools are helping scientists understand how cooperatively-breeding crow populations interact with each other.
• The technology is being deployed to understand how orcas communicate with each other and how underwater noise affects their communication.

What do carrion crows say to each other?

Answering this question has been the mission of Vittorio Baglione and Daniela Canestrari for decades. carrion crow (A crow with a crown) are particularly interesting because they are involved in cooperative breeding, where not only the mother and father, but the whole family is involved in raising the chicks and defending the nest. It requires complex and nuanced communication between individuals.

“They have a very complex society and they do very complex things together,” Baglione, a professor at the University of Leon in Spain, told Mongabay in a video interview. “It’s really coordinated behavior and we’ve answered why they do it, but we want to know how they coordinate and exchange information.”

The pair deployed audio recorders and biologists in northern Spain to decode crow calls and monitor associated behavioral patterns. But as information accumulates, they cross a huge hurdle. “Each microphone lasts six to seven days,” Professor Canestrari of the same university told Mongabay in a video interview. “We realized we had a lot more data to analyze.”

Since 2024, scientists have collaborated with the Earth Species Project (ESP). A team at a US-based non-profit helped them develop artificial intelligence models to classify crow calls and create a data set of different types of calls.

How different animals interact with each other has been a subject of human fascination for generations. The advent of AI in conservation has accelerated the process of identifying patterns and structures in massive data sets, taking us one small step forward in understanding how animals interact with each other. ESP has developed a model that can be used to analyze communication across species. The group also works with scientists around the world, such as Baglione and Canestrari, to develop custom AI tools specific to the animals they are working with.

“If you look at the biodiversity crisis, there’s a huge disconnect from nature,” David Robinson, ESP’s senior AI research scientist, told Mongabay in a video interview. “Communication studies can be a window into the inner life and complexity of other species and help us gain insight into their behavior.”

For example, with carrion crows, the technology has helped scientists identify more than 127,000 vocalizations. It has also helped distinguish between the calls of adults and children as well as other species of birds. When multiple crows are calling, the model also helps synchronize data from different loggers. “In many regions, we had multiple people with dialogue at the same time,” Canestrari said. “They needed to synchronize to see exactly what one bird was doing while the other was calling.”

Although data analysis is still ongoing, scientists have found that most crows’ voices are soft, low-amplitude murmurs. This, they say, potentially indicates that a greater proportion of communication occurs at close range rather than long distance. The scientists are continuing to collaborate with the team at ESP to create a data set, a type of semantic map, that will combine audio data with video as well as data from accelerometers, devices that are used to measure speed and motion. “We can match behavior and calls to give clues about the effectiveness of those calls,” Baglione said.

The technology is not limited to terrestrial species.

At the Raincoast Conservation Foundation in Canada, scientists are collaborating with ESP to understand the communication and behavioral patterns of orcas (Orcinus orca) species live in tight groups and are known to use their calls for movement and hunting together. Decoding their calls will help scientists understand the species’ behavioral patterns. Valeria Vergara, co-director of the Raincoast Conservation Foundation’s cetacean conservation research program, told Mongabay in a video interview that it would also provide insight into “the issue of underwater sound conservation and how it affects the ability of these animals to chat with each other.”

As part of a pilot project, ESP and Raincoast Conservation are using data from drones as well as acoustic recorders to create a data set that can link orca vocalizations to their behavior and environmental factors. While this work is ongoing, the team has already used AI models to speed up the process of identifying orcas from massive amounts of ocean audio data. “Instead of listening to a recording every minute, this tool can just pick out the vocalizations so we can then look at the vocalizations and label them,” Vergara said.

The technology has also helped synchronize voice notes that scientists record in the field with corresponding whale audio or drone clips. “It would have taken us months in the past to synthesize all these data sets, but the tool can do it in an hour,” Vergara said. “What it does is it prepares the data to analyze and quantify our very qualitative voice notes.”

Vergara said the ultimate goal is to use technology to go beyond identifying dialects and call types to understand the meaning of those calls. However, he urged caution over the growing public fascination with whether AI could enable humans to interact with other animals.

“I’d say let’s be happy to understand the whales on their own terms,” ​​he said, “without inserting ourselves into the conversation.”

Abhishant Kidangur is a staff writer at Mangabe. Find him in 𝕏 @Abhishyanthpik.

–

Previously published with news.mongabay Creative Commons Attribution

***